The question I hear most often from teachers right now is some version of: What should I actually do about AI in my classroom?

It is a fair question. It is also, I think, the wrong one, or at least an incomplete one. Because what a teacher should do about AI depends entirely on what her discipline requires. And that is not a generic answer. What counts as genuine work in English is not what counts as authentic study in history, which is not what counts in science or mathematics. AI disrupts each of those disciplines differently, which means the response has to be discipline-specific too.

Over the past year, I have been developing a framework called DSAIL, Discipline-Specific AI Literacy, that takes this idea seriously. The larger project is developing slowly into a manuscript-length exploration of disciplinary literacy, the Five Roles, and the conditions that activate learning inside each discipline, accounting for AI’s ubiquity. The argument at its center is that expert teachers already carry tacit knowledge about what genuine disciplinary thinking requires, and that AI has made naming that knowledge newly urgent. AI can now produce the forms of disciplinary work, the essay, the lab report, the historical argument, the proof, without performing the epistemic acts that give those forms meaning. The response is not detection or prohibition. It is the deliberate design of conditions where genuine disciplinary thinking is required, visible, and assessable.

At the center of that design sits a framework I call the Five Roles.

What the Five Roles Are

The Five Roles are a discrete but important part of discipline-specific AI literacy. Assuming that teachers have established the conditions for rich disciplinary engagement, these roles describe five orientations a student can take toward AI output. They are not steps. They are not a checklist. They are positions on a spectrum that runs from critical distance to collaborative integration, and the student’s movement across them is shaped by context, by the demands of the task, and by the developing expertise the student brings to it.

The Critic maintains critical distance. The student has done her own work first, read the source, attempted the problem, sat with the poem, and then brings AI’s output into contact with that work. The purpose is not to catch AI in a mistake, though that happens. The purpose is to see what the contact between her thinking and AI’s output reveals. What did she notice that AI did not? What did AI produce that looks right but feels wrong? The Critic role builds the capacity to evaluate, and it depends on the student having something of her own to evaluate against.

The Verifier tests and confirms. The student checks AI’s claims against sources, verifies its reasoning for soundness, determines whether what AI asserts can be trusted. This is slower and more deliberate than the Critic role. It builds systematic habits of validation and strengthens the student’s knowledge of what counts as reliable evidence in her field.

The Interlocutor engages in dialogue. The student uses AI to test ideas, explore alternatives, and develop her thinking through exchange. She is not asking AI for an answer. She is using the conversation to sharpen something she already owns. The Interlocutor role requires that the student have enough of her own thinking developed to maintain intellectual ownership of the exchange. Without that, dialogue becomes dictation.

The Editor refines and revises. The student works with AI feedback to improve work she has already produced, deciding which suggestions serve her meaning and which distort it. In some disciplines, the Editor role also functions as an access role, using AI to clear a path into a difficult text or an unfamiliar register, so that the interpretive or analytical work can happen.

The Architect orchestrates production. The student designs specifications that direct AI toward a particular output, encoding her disciplinary knowledge into prompts and constraints. This role requires the deepest understanding of all five, because a student who cannot specify what she wants cannot evaluate what she gets. Most students cannot do it well, and discovering that is part of the point.

The Developmental Logic

The roles are sequenced for a reason. The sequence is not prescriptive. A student does not move through the roles one at a time like levels in a game. On any given day she might cycle from Critic to Interlocutor and back to Critic three times. The sequence describes the order in which capacities must be built. A student cannot orchestrate AI effectively until she can evaluate its outputs critically. She cannot use dialogue productively until she has enough knowledge to maintain ownership of ideas. Critic and Verifier build the foundation that makes Interlocutor, Editor, and Architect possible.

This means the most common way students use AI right now, jumping straight to what amounts to the Architect role, asking AI to produce something they then submit, is not just problematic because it bypasses learning. It is problematic because it skips the very capacities that would make AI use productive. A student who delegates before she can evaluate has not made a strategic choice. She has made the only choice available to someone who has not yet developed the earlier capacities.

The developmental logic also means that experts continuously return to the Critic role when they encounter unfamiliar territory. The goal is not to graduate from early roles to advanced ones. The goal is to move intentionally across the spectrum based on what the situation requires and what one’s developing expertise allows.

Why the Discipline Decides

Here is where the framework becomes something more than a general account of AI use. The Five Roles are the same across disciplines. What each discipline requires of each role is not.

Consider the Critic.

In a history classroom, the Critic is the student who has read the primary sources before consulting AI, who has practiced sourcing, asking who wrote this, when, why, and from what position, and who can now identify that AI’s historical account treats all information as equally positioned, all claims as equally grounded. AI does not source. It presents the past with a confidence that no historian would recognize as legitimate. The history student occupying the Critic role can name that absence because she has practiced the act AI cannot perform.

In a science classroom, the Critic is the student who has collected her own data before asking AI to explain the phenomenon. She compares AI’s explanation to what her data actually show and discovers that AI reasons from expected conclusions, from what experiments typically find, rather than from the specific data in front of her. AI’s explanation may be textbook-accurate and still wrong about what happened in her lab. The science student occupying the Critic role can name that gap because she has the empirical record AI does not.

In an English classroom, the Critic is the student who has sat with the poem before asking AI what it means. She has had her own encounter, marked lines, been confused, returned to phrases that caught her attention before she could say why, and she brings that encounter to AI’s interpretation. AI’s reading will be competent, sometimes even impressive, and entirely devoid of a reader. There is no felt experience behind it, no moment of confusion that led to insight, no personal stake in the interpretation. The ELA student occupying the Critic role can name that absence because she has been the reader AI cannot be.

In a mathematics classroom, the Critic is the student who has attempted the proof before asking AI to produce one. She evaluates AI’s proof not for whether the answer is correct but for whether each step follows from the one before it by logical necessity. AI produces proof-shaped text, steps that follow each other in sequence, without the logical entailment that makes a proof a proof. The mathematics student occupying the Critic role can name the difference between form and substance because she has struggled with the substance herself.

The role is the same in every case: the student who has done her own work first and who brings AI’s output into contact with that work to see what the contact reveals. The disciplinary act that fills the role is different in every case: sourcing, empirical reasoning, interpretive encounter, logical entailment. The framework is transdisciplinary not because it ignores disciplinary difference but because it operates at a level above disciplinary content. What each discipline requires of the Critic is different. What the Critic requires of the student, prior engagement, independent judgment, evaluation against an internalized standard, is the same.

Different Disciplines, Different Arrangements

The Five Roles do not carry equal weight across all fields. Each discipline’s epistemology shapes which roles matter most and how they are used.

History emphasizes Critic and Verifier. Historical thinking centers on evaluating sources, checking evidence, and assessing competing accounts. The Verifier role is particularly accessible in history because verification means checking whether a source says what a summary claims it says, a relatively straightforward act that students can begin practicing early. The Interlocutor role is useful for exploring competing interpretations, but the discipline’s authority runs through evidence, not through dialogue. The Architect role is available but constrained: a student can specify a structure for her argument, but she must write the interpretation herself.

Science emphasizes Verifier and Architect, and has a distinctive relationship to legitimate AI use. Scientific work requires checking empirical claims and designing systematic procedures. The Verifier role carries particular weight because the student has her own data as a check, a concrete, specific standard that AI’s general knowledge cannot replicate. What makes science distinctive among the four disciplines is the Architect role’s legitimate scope. There are genuine computational and analytical tasks, running a statistical analysis on a data set, generating a visualization, conducting a literature search, that AI can perform in service of the student’s investigation without compromising her empirical reasoning. The teacher’s design challenge is ensuring that delegated tasks remain embedded within a sequence that preserves the student’s own reasoning at every stage.

English Language Arts emphasizes Interlocutor and Editor, and resists the Architect. Literary analysis develops through articulating interpretations and refining them through revision. The Interlocutor role is particularly productive in ELA because literary meaning is constituted through exchange. A student who has noticed something in a text but cannot yet name it can use dialogue to find language for the noticing. The Editor role functions distinctively as an access role: AI can help a student get inside a difficult text, clarifying an archaic word, paraphrasing a syntactically dense passage, without doing the interpretive work for her. The Architect role is the most limited in ELA of any discipline. Meaningful literary interpretation emerges through the encounter with the text, and it resists specification. A student who can write a full specification for a literary argument has already done the intellectual work the essay was meant to produce.

Mathematics emphasizes Verifier above all, and the Verifier role carries a weight it does not carry elsewhere. In mathematics, verification means checking each step of a logical chain for entailment. This is not a supplementary check. It is the discipline’s fundamental act. A student who can read a proof for logical necessity rather than for sequential plausibility is doing the thing that mathematics is. The Architect role also fits well: specifying a proof strategy or a solution method makes mathematical thinking explicit in ways that are both diagnostic and productive. The Interlocutor role is useful for exploring problem approaches, but solutions must ultimately be verified rigorously. Dialogue does not substitute for proof.

What This Means for Teachers

The framework asks of every lesson one question before any other: What disciplinary act is this lesson designed to require? Not what AI tool students will use, not what policy will govern their use, not what generic competency will be addressed. What act. Sourcing. Interpretive encounter. Empirical reasoning. Logical entailment. Once the act is named, the role follows. And once the role is clear, the design of the assignment, what must be collected, when AI enters the sequence, what evidence of genuine disciplinary thinking the teacher needs to see, becomes specific rather than generic.

A student who has learned what it means to occupy the Critic role in history class has learned something she can carry to science, to ELA, to mathematics. She will need to learn what sourcing means specifically, what comparing AI’s explanation to her own data means specifically. But the underlying orientation, bringing AI’s output into contact with her own prior work and asking what the contact reveals, is transferable. A teacher who names that transfer explicitly is doing one of the most generative things she can do.

The Five Roles are not a policy. They are not a restriction on AI use. They are a framework for making AI use deliberate, for keeping the epistemic work of the discipline in the student’s hands, and for ensuring that what students build when they work with AI is genuine capacity rather than efficient dependence.

The DSAIL project and the manuscript it is becoming will develop all of this in far greater detail: the disciplinary arguments, the unit sequences, the assessment logic, the developmental arc. But the Five Roles are where the design begins. They are the point where a teacher’s disciplinary knowledge meets the question of what to do about AI, and where the answer turns out to be something she already knew how to do, described in language she can now use.

Nick Potkalitsky, Ph.D.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy

Educating AI publishes twice weekly. If this piece was useful, share it with a colleague who’s navigating these questions. And if you want to support the work, consider becoming a paid subscriber — it’s what keeps this going.

Check out my serial releases from my latest book, Thinking with AI: A Student’s Guide to Literacy in an AI-Rich World.

Intro and Ch. 1 Ch. 2-3 Ch. 4-5 Ch. 6-7

One district near me has been holding AI literacy nights for parents for the past two years. They started before most schools were even having the conversation. I think that kind of foresight deserves recognition, and it is part of what inspired this resource.

What those evenings accomplish goes well beyond sharing information. They are an opportunity to meet your community where it actually is: to understand their real concerns about AI, to get a read on their familiarity and comfort level, to demonstrate which tools your school has vetted and why, to be transparent about safety and data privacy, and to show families that specific decisions are being made about purpose and scope, not just reacting to whatever arrives next.

Just as importantly, community engagement around AI should not be a one-time event. Periodic surveying gives schools a way to track how parent understanding and concerns are shifting over time. That data, when taken seriously, can directly shape what future workshops address. The event design below is one suggested pathway, built to be adapted based on what your community actually needs.

The guide below is a fully designed event: four rotating 15-minute stations, facilitation scripts, sample materials, a consolidated parent takeaway sheet, and facilitator preparation notes. Each station stands on its own. Schools can run the full event in about 75 minutes or pull individual stations for other contexts, like a back-to-school night table or a department meeting.

The guide include slide decks and graphics for each station.

This post also brings the Thinking With AI series to a close. Over the course of this series, we have built out resources for students, educators, and now families. What comes next is an e-version of the full text, designed to serve as a grounding document for professional development, school-wide training, and community events like the one described above. More details soon.

Nick Potkalitsky, Ph.D.

Read more

Last week’s release of the student workshop series drew the kind of response that confirms this work is landing. Thank you for engaging with those materials so seriously, and for sharing them so widely. The suite of workshops I promised is taking shape, and this week continues that build.

This edition is for educators.

What follows is the teacher-facing complement to last week’s release: two complete workshop designs and six cross-disciplinary sample lessons built for grades 6 through 12. Where last week’s materials put students in the role of learner, this week’s put teachers in the role of designer. The question at the center of both workshops is the same one that drives the whole project: how do you structure learning so that the thinking lands with students, not with AI?

Subscribe now

The workshop materials come in two formats. The 1.5-hour introduction, “Same Technology. Different Trajectories,” is built for mixed-subject faculty audiences who are new to disciplinary AI literacy. It establishes a shared foundation, works through the five principles, and sends educators away with one concrete move they can make in their next unit. The 3-hour deep dive, “Designing for Student Thinking in an AI-Rich World,” goes further: educators analyze sample lessons, design their own, and identify the coordination gaps that keep individual classroom efforts from adding up to something systemic. Both workshops are built for cross-disciplinary audiences, because the conversation across subject areas is not a compromise. It is the point.

The real substance of this release is the six sample lessons. These are the fullest account I have offered to date of what disciplinary AI literacy looks like in practice, lesson by lesson, step by step. They span ELA, social studies, science, and math. They range from grade 6 through grade 12. Each one is built around the five DSAIL principles and designed so that student thinking is visible, documentable, and assessable at every stage. The lessons are included in the workshop as objects of analysis, but they are also ready to adapt and use.

A few things are worth naming about how these lessons are designed. Every one of them begins before AI enters the room. Students write first, read first, frame first, argue first. AI comes in after students have something at stake in the question, something they have already committed to thinking about. That sequence is not incidental. It is the whole architecture. And every lesson ends with students accounting for their own moves: what they accepted from AI, what they rejected, and why. That reflection is where the learning consolidates.

These materials are designed for educators who are not waiting for someone else to solve this. They are for the teachers and instructional leaders who understand that AI is already in their classrooms and have decided to meet it with intention rather than anxiety. The framework does not ask teachers to become AI experts. It asks them to do what they have always done: design for student thinking, assess student moves, and protect the cognitive work that makes learning real.

Next week, the series closes with the parent and guardian workshop. That release will complete the full community suite I outlined at the start of this series.

Nick Potkalitsky, Ph.D.

Read more

The response to Thinking with AI has been genuinely energizing. Your engagement with this material, your questions, your pushback, and your enthusiasm for bringing it into real classrooms, is exactly what keeps this work moving forward. Know that I am currently working on an e-book version of Thinking with AI. Be on the look-out for updates over the next couple weeks.

Subscribe now

Over the next three weeks I am developing a full suite of workshop materials designed to help bridge the AI literacy gap in your community: student workshops, teacher workshops, and parent/guardian workshops. This edition launches that series with the complete student cycle.

This release includes the full lesson plans for all five sessions of the Thinking With AI student workshop series, along with five interactive student-facing artifacts, one per session, built for direct use in the classroom. The lesson plans are the architecture. The artifacts are the working space where students do the actual building.

The five sessions move in a deliberate sequence.

Session 1 establishes what AI actually does, pattern prediction rather than understanding, and introduces the five principles that anchor the whole series.

Session 2 puts two foundational roles in students’ hands: the Critic, who evaluates AI outputs for accuracy, completeness, and depth, and the Verifier, who checks claims against real sources.

Session 3 introduces the Interlocutor, teaching students to use AI dialogue to develop thinking they already own rather than outsource thinking they have not yet done.

Session 4 adds the Editor, who triages AI feedback on their own drafts, and the Architect, who directs AI by encoding genuine expertise into clear specifications.

Session 5 brings all five roles together on a single task, closes with a structured self-assessment, and asks students to make a concrete commitment about how they will engage with AI going forward.

This process is grounded in Thinking with AI’s 5 Principles for AI Engagement and Instruction:

I want to acknowledge Elizabeth Helfant, Assistant Head and Academic Coordinator at Ravenscroft School in Raleigh, NC, whose work helped bring this material to life in the classroom. Drawing on the framework from Thinking with AI, Elizabeth developed her own original artifacts and is currently leading a 7th grade half-semester course that integrates these materials with NeuroTeach. Her implementation is a serious one, built on careful reading of the underlying ideas and a genuine commitment to doing this work well. It is exactly the kind of educator-led effort this framework was designed to support.

Both the lesson plans and the artifacts rest on the same conviction: AI does not reduce the thinking required for intellectual work. It relocates it. These materials are designed to make sure that thinking lands where it belongs, with students.

Read more

Educating AI publishes twice weekly. If this piece was useful, share it with a colleague who’s navigating these questions. And if you want to support the work, consider becoming a paid subscriber. It’s what keeps this going.

Subscribe now

Check out my serial releases from my latest book, Thinking with AI: A Student’s Guide to Literacy in an AI-Rich World.

Intro and Ch. 1 Ch. 2-3 Ch. 4-5 Ch. 6-7

Next week, I hope to release some training roadmaps for educators and administrators interested in piloting discipline-specific AI Implementation in their contexts!

Nick

One of the most dangerous sentences an AI system can produce is also one of the most reassuring: “Yes, this quote matches the original document.”

It feels definitive. Settling the nerves. The kind of authoritative confirmation that lets you move on to the next task. In my experience, it is often wrong. Not dramatically wrong. Not in ways that announce themselves. But wrong in the subtle, structurally consequential ways that undermine the integrity of research, instruction, and professional judgment. This is not a fringe failure mode. It is a direct consequence of how generative AI systems work.

The Misconception

When you upload a document into a chat and ask the model to quote from it, there is a natural assumption that the system can simply “look it up.” That somewhere inside the machine, a search function locates exact text and compares it with mechanical precision.

This is not how large language models operate.

They do not retrieve text the way databases do. They reconstruct language probabilistically, token by token, based on patterns, likelihoods, and semantic approximation. Even when the original document is fully present in the prompt, the model does not perform exact character-by-character comparison. It generates what seems right. And what seems right is often close enough to feel authoritative without being literally accurate.

Reconstruction, Verification, and Why Both Fail

Consider a scenario many of us have already encountered. A teacher uploads a policy document and asks the model to quote the section addressing student data privacy. The model produces: “Schools must ensure student data privacy when implementing AI systems.” It sounds correct. It captures the meaning. But the actual document reads: “Schools are responsible for ensuring the privacy and security of student data when implementing AI systems.” The model did not retrieve that sentence. It reconstructed it. One is a verbatim quotation. The other is a paraphrase disguised as one. In education, in policy, in research, the difference between what a source actually says and what it approximately means is the difference between evidence and interpretation.

If the teacher then asks, “Does this quote match the original?” the model will frequently respond with confidence: yes, it does. That confident verification is produced by the same probabilistic process that generated the inaccurate quote in the first place. The system is not stepping outside itself to scan the document like software running a string match. It is generating a judgment about plausibility, evaluating whether the quote feels consistent with the source, not whether it is identical to it.

Researchers call this self-verification failure. Once the model generates a quote, it treats that output as a strong prior and tends to defend it when asked to verify. Because “careful” and “thorough” are semantically similar, the model will judge a substitution of one for the other as correct. Because a sentence with a dropped clause still carries the same general meaning, the model will affirm that nothing is missing. The verification step feels like a second opinion. It is not. It is the same opinion, restated with confidence.

What About RAG?

Some readers familiar with AI architecture may be wondering about Retrieval-Augmented Generation, or RAG, which works differently from standard chat. Instead of processing an entire document at once, a RAG system breaks it into smaller chunks, indexes them, and retrieves only the most relevant passages in response to a query. The model generates its answer based on those retrieved passages rather than the full document.

This offers real advantages for quotation reliability. The model works with a smaller, targeted passage, reducing the positional bias that plagues long-context processing. The retrieved chunk can be presented alongside the response, giving users something concrete to verify against. Well-designed RAG systems can even enforce extractive behavior, requiring the model to pull directly from retrieved text rather than generating freely. For many professional and educational use cases, a well-configured RAG pipeline will produce more grounded, more traceable outputs than a standard chat upload.

But RAG shifts the failure modes rather than eliminating them. If the chunking strategy splits a key sentence across two passages, or the embedding model fails to rank the right passage highly enough, the correct text never reaches the generator. Even with the correct passage retrieved, the generator still operates probabilistically and can still paraphrase or compress unless explicitly constrained. And most critically, recent research by Wallat et al. (2025) found that up to 57% of citations in RAG systems lack genuine faithfulness: the system attaches a citation that looks correct without having actually derived the claim from that passage. The grounding is decorative. RAG moves the needle. It does not solve the problem.

What This Means for Education

Teachers are increasingly turning to AI to support grading and content workflows. A teacher might upload a student essay alongside source materials and ask the AI to check whether the student accurately quoted a primary source. The AI may confidently affirm accuracy even when the student has subtly altered the original text. The error is not caught. It is certified. Students face a parallel risk. A student writing a research paper may rely on an AI system to locate and quote supporting evidence. The system may produce language that appears authoritative and convincingly specific, but that does not actually exist in the cited source. If the student trusts the verification step, the error becomes embedded in the academic record.

We need to start distinguishing more carefully between tasks that require semantic approximation and tasks that require literal fidelity. Generative AI is extraordinarily effective at semantic tasks: summarization, explanation, synthesis, brainstorming, feedback. These benefit from the model’s ability to reconstruct meaning flexibly. Quotation, citation, and verification are fundamentally different. They require extractive precision, the ability to identify and reproduce exact spans of text without alteration. By default, language models are not optimized for this. They are optimized to produce the most plausible continuation of a sequence, which is a different objective entirely. A teacher asking AI to generate feedback on student writing is operating within the model’s strengths. A teacher asking AI to verify whether a quote appears exactly in a source document is operating outside them.

This is why AI literacy cannot be reduced to prompt engineering or tool familiarity. It must include what I would call epistemic literacy: an understanding of what kinds of questions AI can answer reliably and what kinds require independent verification. Students must learn that AI systems are not authoritative sources. They are generative collaborators whose outputs must be treated as provisional until confirmed through external mechanisms. Teachers must similarly recalibrate their trust. The presence of a document within an AI chat does not guarantee faithful reproduction.

None of this diminishes the transformative potential of AI in education. I have written candidly in this newsletter about how much I benefit from working with these tools, and I am not about to pretend otherwise. But the time saved on generation must be partially reinvested in verification, and I mean genuine verification, not the model checking its own work.

Reliability Is a System Property

We are moving from a world where verification was primarily procedural to one where verification is interpretive. The tools we use to generate knowledge can also generate plausible errors. This places new responsibilities on all of us: to design workflows that preserve human verification at critical points, to teach students not just how to use AI but how to question it, and to develop institutional norms that distinguish between generative assistance and authoritative sourcing.

AI systems are extraordinarily fluent. They produce language that feels precise, confident, and complete. That fluency is their greatest strength and their greatest epistemic risk. When an AI system tells you that a quote is accurate, it is expressing a probabilistic judgment, not a deterministic fact. Understanding this distinction is foundational to every serious conversation we are going to have about AI in education going forward.

Reliability is not a property of the tool alone. It is a property of the human-tool system. And in that system, human judgment remains indispensable.

Nick Potkalitsky, Ph.D.

Further Reading

The research conversation behind this article is growing and worth following. Here are five entry points for readers who want to go deeper:

Lost in the Middle: How Language Models Use Long Contexts (Liu et al., 2024, Transactions of the Association for Computational Linguistics).

Correctness Is Not Faithfulness in RAG Attributions (Wallat et al., 2024/2025, ACM SIGIR ICTIR 2025).

When Can LLMs Actually Correct Their Own Mistakes? A Critical Survey of Self-Correction (Kamoi et al., 2024, Transactions of the Association for Computational Linguistics).

Not Wrong, But Untrue: LLM Overconfidence in Document-Based Queries (Hagar, 2025, Computation + Journalism Symposium).

LaRA: Benchmarking Retrieval-Augmented Generation and Long-Context LLMs (Li et al., 2025, ICML 2025).

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy

I am excited to release the Introduction and Chapter 1 (What AI Is and Isn't) from a resource I have spent the past several months developing, Thinking With AI: A Student's Guide to Literacy in an AI-Rich World*.

I will release the full guide in 2 chapter installments over the next couple weeks. The full guide applies a five-role framework across critical thinking, problem-solving, reading, and writing to show how AI engagement can accelerate disciplinary learning when done critically.

Paid subscribers get access to the complete manuscript as I develop it. If this work matters to you, consider supporting it with a paid subscription. If you want to onboard your team, consider investing in a group subscription (+5 members get 40% discount).

Subscribe now

Thinking With AI: A Student’s Guide to Literacy in an AI-Rich World

Introduction

A Personal Note

I came to AI literacy through the classroom, not the computer lab.

For fifteen years, I taught English, Latin, and humanities across secondary and university settings. I spent my doctoral work at The Ohio State University studying how narratives shape meaning, how rhetoric builds arguments, how media transforms communication. My academic training was deeply humanistic: close reading of texts, careful analysis of language, attention to how ideas develop and persuade.

When AI language models emerged as accessible tools, my first instinct as an educator was skepticism. Here was technology that could generate fluent text without understanding, produce confident-sounding explanations without knowledge, organize information without judgment. Everything I valued about education (struggle leading to understanding, revision clarifying thinking, difficulty building capacity) seemed threatened by tools that made producing text effortless.

But I kept watching my students. And what I noticed surprised me.

Some students used AI exactly as I feared: delegating thinking, accepting outputs without evaluation, producing work that represented someone else’s (or something else’s) intellectual labor. Their writing became more polished but less theirs. Their thinking became less visible. Their capacity stopped developing.

Other students used AI differently. They analyzed AI outputs critically, noticing what was wrong or oversimplified. They verified claims systematically, building knowledge through checking rather than accepting. They used dialogue to develop ideas they were actively working on, not to generate ideas for them. They got feedback on their drafts, evaluated it thoughtfully, and revised toward stronger meaning. These students’ capacity accelerated noticeably beyond peers without AI access.

Same technology. Opposite trajectories.

This pattern held across practices. In critical thinking, some students let AI do their reasoning while others used AI to test and strengthen their own thinking. In problem-solving, some outsourced formulation while others used AI to explore approaches they would evaluate. In reading, some substituted summaries for engagement while others used AI to access difficult texts they would interpret themselves. In writing, some imported AI language while others maintained voice while getting strategic support.

The difference wasn’t the tool. It was how students engaged with it.

This realization changed my work. As AI Literacy Specialist at the Educational Service Center of Central Ohio, I now support 32 school districts navigating AI integration. I work with teachers across disciplines who see both the promise and peril of these tools. I talk with administrators developing policies that must balance innovation with integrity. I hear from parents trying to understand what their children’s education looks like in an AI-rich world.

The challenge they all face is the same: How do we use AI in ways that build student capacity rather than bypass it?

This book offers an answer grounded in both research and practice. It develops a framework for engaging with AI across five roles (from critical distance to collaborative integration) and four core academic practices (critical thinking, problem-solving, reading, writing). The framework helps students, teachers, and parents make intentional choices about when and how to use AI while maintaining the intellectual work that creates learning.

But this book is for everyone navigating the same tension: AI can accelerate learning when engagement is intentional, and it can undermine learning when engagement is passive. Understanding the difference matters enormously.

Guide Overview

Part I: Foundations

• Chapter 1: What AI Is (And Isn’t)

• Chapter 2: Choosing Tools and Shaping Interactions

Part II: Core Practices Each chapter examines one practice across five roles: Critic, Verifier, Interlocutor, Editor, Architect

• Chapter 3: Critical Thinking

• Chapter 4: Problem Solving

• Chapter 5: Reading

• Chapter 6: Writing

Part III: Integration

• Chapter 7: Building Disciplinary Expertise in an AI-Saturated World

Read more

The response to my post about the Claude Soul Document was revealing. Some readers saw it as an endorsement of Claude. Others recoiled at Anthropic and their machinations. Such moments remind me why I do this work on Substack. Between these diverging views sits an abiding tension worth exploring. Mike Kentz’s recent post captures more of this same complexity.

I am offering a 20% Forever Discount to readers who sign up for a yearly paid subscription before the end of the year. Thanks for all the support over the years. Together we are creating a very special collaborative community.

Subscribe now

This past Friday, I presented to the Ohio 8 Coalition—superintendents, tech directors, and union presidents from Ohio’s nine largest urban districts, including Cleveland, Columbus, Akron, Cincinnati, Dayton, Canton, Lorain, and Youngstown. This wasn’t another AI hype session. This was a room full of leaders responsible for hundreds of thousands of students, grappling with a disruption they didn’t ask for and a legislative deadline they can’t ignore.

My main message: We are in the middle of a 3-5 year rebuild. Anyone offering easy solutions hasn’t come to grips with the full complexity of what we’re facing.

The Legislative Clock Is Ticking

Ohio’s House Bill 96 requires every district to adopt an AI policy by July 1, 2026. The Ohio Department of Education and Workforce must publish a model policy by the end of this month. Most districts are somewhere between reactive and compliance-focused—policy frameworks exist, committees have formed, but clear instructional guidance remains absent.

Here’s what I told them: This deadline is not just an obligation. It’s an opportunity.

Not an opportunity to check a box. An opportunity to build a comprehensive guidance document that winnows down the flood of AI tools into a robust, actionable system. If we’re not getting student data across chats—not just inside single chats—we’re greatly underutilizing these tools. We’re experimenting without purpose, without the ability to gauge success.

Four Foundational Shifts We Can’t Avoid

1. Instructional Redesign: The New Starting Point

Here’s the foundational premise: Any work sent outside of class will be completed with the assistance of AI.

Any other starting point is a fool’s errand at this point.

This isn’t pessimism. It’s pragmatism. We need to go beyond AI assessment scales and rubrics that try to police student use. AI is a highly interactive medium of learning—it’s genuinely hard for students to stay inside a “permissible editing modality” when AI continually and unpredictably prompts them toward other kinds of engagement.

The solution isn’t more restrictions. It’s redesign. Chat rooms need to be set up that focus only on assigned purposes. Those engagements need to happen in the classroom, then be closed down when students leave. We need to accept that homework as we’ve known it has fundamentally changed.

2. Assessment Redesign: Naming the Crisis

We are in a crisis moment. Many teachers simply do not know what students know anymore.

We have two possible pathways:

• Go all in with AI: Adopt robust systems that serve as open engagement platforms where learning is proven through punctual, demonstrable moments of competency.

• Lean seriously into process pedagogy: Make the thinking visible, scaffold the journey, assess the learning process itself.

But more importantly, we need to globally decide what it is important for students to know.

I’m impressed by a small school in Connecticut that started assessment redesign with the premise that authority and research are essential parts of student wellbeing. They’re dynamically redesigning an entire assessment system centered on that value.

In public districts, we don’t have that luxury of starting from scratch. State standards and state testing serve as punctual moments where students must demonstrate different kinds of knowledge in circumscribed, non-AI-assisted formats. Until these structures give way—and they won’t, not for 10-20 years—we will continue to have to design and redesign assessment to guide students toward success in these formats.

We can dream about ideal schools where work flows from genuine engagement, where learning happens entirely through authentic real-world situations or design cycles or PBL. I love those dreams. I was one of those dreamers. But when working with the Ohio 8 Coalition, my pragmatism reignited. Our AI response plan needs to build out of the crucible of what is actually happening in schools today.

3. Purposive Technological Ecosystems: Discernment Over Accumulation

Districts are drowning in AI tools. Every edtech company is slapping “AI-powered” on their pitch deck. The result is chaos—teachers don’t know what’s approved, students toggle between platforms, data sits in silos, and no one can assess what’s actually working.

We need discernment of purpose. What are we trying to accomplish? What student data do we need to see that progress? What tools genuinely serve those purposes, and what tools are just adding noise?

This is where the policy deadline becomes an opportunity. Build a guidance document that doesn’t just say “be ethical with AI.” Build one that says: “Here are the three AI systems we’re using district-wide. Here’s what each one does. Here’s when students should use them. Here’s what teachers can see. Here’s how we’re protecting data.”

4. Teacher-Led Curriculum Rebuilding: Top-Down Meets Bottom-Up

I presented initial findings from my disciplinary AI cohort—work happening right now in central Ohio where teachers are rebuilding curriculum together across content areas. The idea is simple: complement top-down trainings (PD days focused mostly on tools) with emergent teacher cohorts rebuilding curriculum from the ground up.

This sparked a fire in several districts.

Why? Because teachers continue to be the superpower during this AI disruption.

Too many caricatures are circulating right now about teacher responses to AI—that they’re either luddites resisting change or reckless early adopters abandoning rigor. The truth is far more complex. Each teacher is grappling with a complexity they never asked for, making difficult decisions with no clear assurance of success, all for the best interests of their students and communities.

As my colleague and I have said: teachers will continue to be the drivers of change over the next 5-10 years. Any AI implementation plan that doesn’t account for their wisdom will not likely have much success.

What I Learned in That Room

The Ohio 8 Coalition represents some of the most complex, under-resourced, high-need districts in the state. They don’t have the luxury of small class sizes or experimental pilot programs. They have real constraints: state testing, tight budgets, teacher shortages, union contracts, board politics, community skepticism.

And yet, the conversation in that room was one of the most hopeful I’ve experienced in months.

Not because anyone pretended this would be easy. But because the leaders present understood that this is a rebuild, not a quick fix. They understood that rushing to compliance without instructional clarity would waste this moment. They understood that the students most vulnerable to AI’s risks—students with IEPs, students in high-poverty schools, students already disconnected from adults—deserve more than policy theater.

We talked about the data: 71% of teachers worry about student emotional dependence on AI. 52% of students in high-AI schools feel less connected to their teachers. 42% of students are using AI for mental health support, often with no adult awareness. 36% of students are aware of AI-generated deepfakes at their schools.

One finding from the Center for Democracy and Technology’s recent report stopped the room: in high-AI districts, students use AI as a companion at double the rate of low-AI districts (63% versus 25%). The same pattern holds for mental health support.

I’ll return to this report in an upcoming article for a deeper dive, but what struck most folks in the audience wasn’t a call to retreat from AI in curriculum. It was a recognition that our current approach (everyone does their own thing, often without adequate training or framing in terms of AI literacy, in the absence of a scope and sequence that defines what kinds of access are developmentally appropriate) is contributing in some way to these outcomes. The solution isn’t less AI. It’s more intentional implementation.

These aren’t abstract futures. These are present realities happening in districts across Ohio right now.

The Path Forward

Ohio districts have until July 2026 to adopt a policy. That’s seven months.

But the real work—the instructional redesign, the assessment reimagining, the purposive technology ecosystems, the teacher-led curriculum rebuilding—that’s a 3-5 year project. Minimum.

Here’s what I’m asking districts to consider:

1. Use the policy deadline to build something comprehensive. Not just a compliance document. A living guidance framework that addresses instruction, assessment, wellbeing, equity, and data governance.

2. Start with instructional redesign. Accept that homework has changed. Design for that reality, don’t fight it.

3. Get serious about assessment. What do students actually need to know? What can we let go? What must we redesign?

4. Build teacher cohorts. Top-down training alone won’t work. Teachers need time, space, and support to rebuild curriculum together.

5. Center the most vulnerable. If your AI plan doesn’t explicitly address students with IEPs, students in poverty, and students already disconnected, it’s not a plan—it’s a hope.

The shift is happening whether we’re ready or not. How we navigate it is up to us.

I left that room grateful—grateful to work with districts doing the hard, unglamorous work of real transformation. Not the Silicon Valley version. The version that happens in classrooms with 30 students, limited budgets, and state tests that still matter.

That’s the work that will determine whether AI becomes a tool for equity or another mechanism of disparity. And that work doesn’t happen in 90 days. It happens over years, with teachers at the center, guided by leaders willing to name the complexity and commit to the rebuild.

Nick Potkalitsky, Ph.D.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy

An Introduction for Parents and Educators

We’re living through a profound shift in how text gets created. For the first time in history, our children will grow up in a world where much of what they read (homework help, explanations, stories, even “personal” messages) might be generated by AI rather than written by humans.

This isn’t a distant future concern. It’s happening now, in their classrooms and on their screens.

The question isn’t whether to expose children to this reality. They’re already immersed in it. The question is: Are we teaching them to navigate it?

Most AI literacy curricula focus on older students: high schoolers learning about algorithmic bias, college students studying machine learning ethics. These are important, but they miss a critical window. By the time students reach high school, they’ve already spent years passively consuming AI-generated content without tools to evaluate it. We’re teaching them to think critically about AI after they’ve already formed habits of blind trust.

We need to start younger. Much younger.

The curriculum that follows takes a different approach. It doesn’t try to explain how AI works under the hood (that can come later). Instead, it teaches elementary students (kindergarten through fifth grade) practical detective skills for the world they’re actually living in:

• How to verify facts instead of trusting confident-sounding assertions

• How to recognize the difference between human writing and AI-generated text

• How to value specificity, detail, and authentic voice over generic correctness

• How to build habits of skepticism and verification that will serve them for life

The approach is deliberately playful. Students become “Text Detectives” solving cases, checking sources, and learning to spot what one young student brilliantly called the “flavor” of real human writing versus the textbook correctness of AI output.

This isn’t about demonizing AI. It’s about building discernment. Children who complete this curriculum won’t fear AI-generated text, but they won’t blindly trust it either. They’ll develop what we might call “textual situational awareness”: the ability to recognize what they’re reading, where it comes from, and how much weight to give it.

The stakes are high. Students who can’t distinguish between human and AI writing will struggle to develop their own authentic voices. Students who can’t verify facts will be vulnerable to plausible-sounding misinformation. Students who never learn to demand sources will accept authority without question.

But students who master these detective skills? They’ll be prepared for a world where the origin of text matters, where verification is essential, and where authentic human expression is more valuable than ever precisely because it’s becoming rarer.

The curriculum below is designed to be adapted to your classroom or homeschool setting. It’s provisional, a starting point for a conversation we urgently need to have about what literacy means in an age of abundant AI-generated text.

Let’s begin.

Nick Potkalitsky, Ph.D.

THE TEXT DETECTIVE TRAINING MANUAL 🔍

For Young Investigators (Grades K through 5)

WELCOME, DETECTIVE!

Congratulations! You’ve been selected to join the Text Detective Academy.

Your mission? Learn to tell the difference between writing that comes from real people and writing that comes from computers (we call them AI bots or robots).

Why does this matter? Because nowadays, LOTS of the words you read might come from robots, and you need to know the difference!

In this training manual, you’ll learn:

👃 How to sniff out robot writing (it has a special smell!)

🔬 How to test if facts are really true

🕵️ How to spot real human stories vs. made-up robot stories

⚖️ When to trust what you read and when to be suspicious

Ready to become a Text Detective? Let’s go!

📋 YOUR DETECTIVE BADGE

🏅 OFFICIAL TEXT DETECTIVE BADGE 🏅

Detective Name: _____________________

Badge Number: _____________________

Training Started: _____________________

My Detective Motto: _____________________

[Space for student self-portrait]

PART 1: MEET THE SUSPECTS 👥🤖

Suspect #1: The Human Writer 👤

What they’re like:

• Has real memories and experiences

• Makes mistakes sometimes

• Writes in their own special way

• Cares about what they’re writing

• Can tell you stories that REALLY happened to them

Example of Human Writing:

“My dog Biscuit ate my homework last Tuesday. I mean, he REALLY ate it. There were teeth marks and drool everywhere. My teacher didn’t believe me until my mom sent a photo. Biscuit looked so proud of himself!”

Detective Notes: See those specific details? “Last Tuesday,” “teeth marks and drool,” “looked so proud.” Those come from a REAL memory!

Suspect #2: The Robot Writer (AI) 🤖

What they’re like:

• Never actually experienced anything

• Follows patterns it learned from reading LOTS of writing

• Sounds smart but doesn’t have real memories

• Writes things that sound correct but feel... empty

• Uses fancy words but the stories feel fake

Example of Robot Writing:

“My dog is a wonderful pet. Dogs are loyal companions who bring joy to families. Having a dog teaches responsibility and provides opportunities for outdoor exercise. Pets are important members of the household.”

Detective Notes: This sounds nice, but it could be ANYONE’s dog! No real details. Just general facts about dogs. This is robot writing!

THE DETECTIVE’S FIRST BIG QUESTION:

When you read something, always ask:

“Is this from a PERSON or a ROBOT?”

🎯 INVESTIGATION #1: THE FACT CHECKER CASES

Detective Skill: Can I Check This?

Sometimes robots tell us facts. But here’s the thing: Just because a robot sounds confident doesn’t mean it’s right!

Your Job: Learn to check if facts are TRUE.

CASE #1: The Confident Robot 🦜

The Story:

A robot told a student: “Parrots can live to be 200 years old and can learn to speak 500 different words.”

Wow! That sounds amazing! But is it TRUE?

Your Detective Work:

1. Check your science book: What does it say about parrots?

2. Ask your teacher: Do they know about parrots?

3. Look in the encyclopedia: What do the experts say?

What You’ll Find:

• Parrots can live 50 to 80 years (not 200!)

• Most parrots learn 20 to 100 words (not 500!)

🚨 THE ROBOT WAS WRONG! 🚨

Even though it sounded really confident!

⭐ DETECTIVE’S RULE #1: IF I CAN’T CHECK IT, I CAN’T TRUST IT!

YOUR TURN: Fact Checker Training

The robot told you these “facts.” Can you check them?

Put a ✅ if you can verify it, or a ❌ if you can’t find proof!

1. ⬜ “The sun is a star.”

   • Where I checked: ________________

2. ⬜ “There are purple tigers in Australia.”

   • Where I checked: ________________

3. ⬜ “George Washington was the first U.S. President.”

   • Where I checked: ________________

4. ⬜ “Chocolate comes from cacao beans.”

   • Where I checked: ________________

5. ⬜ “Your teacher’s favorite color is blue.”

   • Where I checked: ________________

Detective Tip: Notice how some facts you can check in books, but some you need to ask a real person? That’s being a smart detective!

CASE #2: The Missing Source Mystery 📚

The Mystery:

Two students wrote reports about dolphins. Let’s investigate!

Report A (Encyclopedia Entry):

“Dolphins are marine mammals that live in oceans around the world. According to National Geographic, dolphins use echolocation to find food. Scientists at the Marine Biology Institute have observed dolphins working together to hunt fish.”

Sources you can check:

• National Geographic magazine ✅

• Marine Biology Institute ✅

Report B (Robot Answer):

“Dolphins are very intelligent creatures. They are known for their playful behavior and complex communication. Dolphins live in groups called pods and are considered one of the smartest animals on Earth.”

Sources you can check:

• ??? (No sources listed!) ❌

🔍 Detective Question: Which report helps you check if the facts are true?

Answer: Report A! It tells you WHERE the information came from!

⭐ DETECTIVE’S RULE #2: REAL FACTS HAVE SOURCES YOU CAN CHECK!

YOUR DETECTIVE NOTEBOOK: Fact-Checking Log

📓 FACT-CHECKING LOG 📓

Date: _____________

What I read: _________________________________

Where it came from: ⬜ A book ⬜ A robot/AI ⬜ A website ⬜ A person

Can I check this fact? ⬜ YES ⬜ NO

Where I checked:

My verdict:

⬜ ✅ TRUE (I found proof!)

⬜ ❌ FALSE (It’s wrong!)

⬜ ⚠️ CAN’T VERIFY (I can’t find proof either way)

Detective Notes: ____________________________

🎭 INVESTIGATION #2: THE STORY DETECTIVE CASES

Detective Skill: Does This Sound Real?

Now we switch from checking FACTS to checking STORIES. This is different! Stories need different detective skills.

The Big Difference:

• Facts = Check if they’re TRUE ✅❌

• Stories = Check if they come from a REAL PERSON 👤 vs. a ROBOT 🤖

CASE #3: The Two Grandmas 👵👵

Two students wrote about their grandmothers. One is REAL, one was written by a robot. Can you tell which is which?

Story A:

“My grandma’s kitchen always smells like cinnamon and coffee. She has this weird green mixer from the 1970s that makes a REALLY loud noise, but she won’t get rid of it because Grandpa gave it to her. There’s a crack in the tile behind the stove from when I dropped a jar of pickles in 2nd grade. She never fixed it. She says it reminds her of me.”

Story B:

“My grandmother’s kitchen is a warm and welcoming space. The aroma of home-cooked meals fills the air. Family photographs line the walls, showing happy memories from over the years. It is a place where the family gathers to share meals and create lasting bonds.”

🔍 Detective Work: Let’s investigate!

Story A Clues:

• Specific smells (cinnamon AND coffee, not just “good smells”)

• Weird detail about the green mixer from the 1970s

• The LOUD noise (specific!)

• Story about dropping pickles

• Grandma kept the crack on purpose

• This could ONLY be this kid’s grandma

Story B Clues:

• “Warm and welcoming” (everyone says this!)

• “Aroma of home-cooked meals” (whose meals?)

• “Family photographs” (which photos?)

• This could be ANYONE’s grandma

• Sounds nice but feels empty

🎯 Answer: Story A is from a REAL person! Story B is from a robot!

⭐ DETECTIVE’S RULE #3: REAL STORIES HAVE WEIRD, SPECIFIC DETAILS!

Robot stories sound nice but could belong to anyone.

Human stories have details that make them special and unique!

YOUR TURN: The Pet Detective Challenge 🐕🐈🐹

Part 1: Write Your Real Story

Write 3 to 5 sentences about YOUR real pet (or a pet you know):

My Real Pet Story:

Weird/specific details I included:

Part 2: Compare to Robot Writing

Now your teacher will show you what a robot writes when we ask: “Write a story about a pet.”

[Teacher uses AI to generate generic pet story in class]

Detective Questions:

1. Does the robot story have specific details like yours? ⬜ YES ⬜ NO

2. Could the robot story belong to anyone? ⬜ YES ⬜ NO

3. Does your story have details only YOU would know? ⬜ YES ⬜ NO

4. Which story feels more REAL to you? ⬜ Mine ⬜ The robot’s

CASE #4: The Flavor Test 👅

Thanks to Detective Artyom for this discovery!

Artyom’s Discovery:

A student detective named Jim discovered something important. He said:

“The robot makes English that is correct but not real. It’s like textbook English. It follows all the rules perfectly, but it doesn’t have the flavor of how people actually write.”

What is “flavor”?

Think about your favorite food. Real chocolate chip cookies from your kitchen taste different than the picture of cookies in a magazine, right? One has FLAVOR. The other just looks right.

Writing is the same way!

Robot Writing (Correct but no flavor):

“Today was an enjoyable day. We participated in various activities during recess. The weather was pleasant. I appreciated spending time with my classmates.”

Human Writing (Has flavor!):

“Today was awesome! Me and Marcus invented a new game at recess. It’s like tag but you have to hop on one foot. I fell down twice but it was so funny. Even the teacher laughed!”

🔍 The Flavor Checklist:

Robot writing:

• ⬜ Uses fancy words nobody really says (”participated,” “pleasant,” “appreciated”)

• ⬜ Sounds like a textbook

• ⬜ Too perfect (no mistakes or real kid voice)

• ⬜ Could be written by anyone

Human writing:

• ⬜ Sounds like a real kid talking

• ⬜ Has energy and feeling

• ⬜ Specific names and details

• ⬜ Sounds like THIS kid, not just any kid

⭐ DETECTIVE’S RULE #4: REAL WRITING HAS FLAVOR!

If it sounds like a textbook following rules perfectly, it might be a robot!

YOUR TURN: Flavor Detective Practice

Read these sentences. Put an R for Robot or H for Human:

1. ⬜ “My dog is very energetic and enjoys outdoor activities.”

2. ⬜ “My dog Bruno goes CRAZY every time the mailman comes. He does this weird spinning thing and barks at his own tail!”

3. ⬜ “The science experiment was educational and informative.”

4. ⬜ “We mixed the baking soda and vinegar and it EXPLODED everywhere! Jordan got some on his shirt and Ms. Chen made us clean up for like twenty minutes.”

5. ⬜ “Reading is an important activity that helps develop literacy skills.”

6. ⬜ “I stayed up way too late reading the new Dog Man book under my covers with a flashlight. Totally worth being tired today.”

Answers: 1-R, 2-H, 3-R, 4-H, 5-R, 6-H

Did you notice? The human ones sound like a real kid talking to you!

🎓 DETECTIVE GRADUATION: PUTTING IT ALL TOGETHER

The Two-Question Detective System

When you read ANYTHING, ask yourself:

QUESTION 1: What kind of writing is this?

⬜ FACTS (things that can be checked)

⬜ STORIES (personal experiences)

Then:

If FACTS, ask:

• Can I check this?

• Where are the sources?

• Is it true?

If STORIES, ask:

• Does this have specific details?

• Could this be anyone’s story?

• Does it have flavor?

🏆 FINAL EXAM: The Detective Case Files

Case File #1: The Homework Helper

Your friend shows you their homework. They wrote:

“The water cycle is an important natural process. Water evaporates from bodies of water and forms clouds. Precipitation occurs when water falls back to Earth. This cycle is essential for life on our planet.”

Your detective work:

1. Is this FACTS or a STORY? _____________

2. Can you check these facts in a science book? ⬜ YES ⬜ NO

3. Does this sound like your friend talking? ⬜ YES ⬜ NO

4. Does this have “flavor”? ⬜ YES ⬜ NO

Your verdict:

⬜ Written by a person

⬜ Written by a robot

⬜ Not sure (need more clues!)

Why? ________________________________

Case File #2: The Book Report

Another student wrote this book report:

“I read ‘Charlotte’s Web’ and it made me cry at the end! I didn’t think I’d care that much about a spider, but Charlotte was so nice to Wilbur even though he was kind of whiny sometimes. My favorite part was when she wrote ‘SOME PIG’ in her web because it was such a weird thing to do but it totally worked. I keep thinking about how Charlotte knew she was going to die but she helped Wilbur anyway. That’s the kind of friend I want to be.”

Your detective work:

1. Is this FACTS or a STORY/OPINION? _____________

2. Can you tell this person really read the book? ⬜ YES ⬜ NO

3. Does this have specific details? ⬜ YES ⬜ NO

4. Does this have “flavor”? ⬜ YES ⬜ NO

5. Could a robot write this without reading the book? ⬜ YES ⬜ NO

Your verdict:

⬜ Written by a person

⬜ Written by a robot

⬜ Not sure (need more clues!)

Why? ________________________________

📜 YOUR OFFICIAL DETECTIVE CERTIFICATE

🏅 OFFICIAL TEXT DETECTIVE CERTIFICATE 🏅

This certifies that

has successfully completed training at the TEXT DETECTIVE ACADEMY

and can now:

✓ Check facts against real sources

✓ Spot robot writing vs. human writing

✓ Detect missing flavor in text

✓ Find specific details that make stories real

Awarded this _____ day of ________, 20

Your Teacher’s Signature Your Signature

Badge Number: ____________

🎒 DETECTIVE’S FIELD KIT

Keep These Tools Handy!

Your Quick Reference Card:

WHEN READING ANYTHING:

❓ Is this FACTS or STORIES?

For FACTS:

• Can I check this?

• Where are the sources?

For STORIES:

• Weird, specific details?

• Real person’s voice?

• Has flavor?

🚨 IF I CAN’T CHECK IT, I CAN’T TRUST IT!

Congratulations, Detective! Keep investigating! 🔍

TEACHER’S GUIDE

How to Use This Workbook

Week 1 to 2: Introduction + Fact Checker Cases

• Focus on simple verification

• Practice checking facts in multiple sources

• Build habit of asking “Where can I check this?”

Week 3 to 4: Story Detective Cases

• Shift to recognizing human voice vs. templates

• Practice writing with specific details

• Compare their writing to AI-generated examples

Week 5 to 6: Integration & Practice

• Mix fact-checking and story detection

• Real-world examples from their reading

• Build detective notebooks as ongoing practice

Assessment Ideas

• Detective notebooks (ongoing documentation)

• Partner “case solving” activities

• Final case file presentations

Differentiation

• K to 2: Focus heavily on oral discussion, simpler examples

• 3 to 5: Add written analysis, more complex cases

• Advanced: Have students CREATE cases for classmates

This curriculum is provisional and designed to be adapted to your specific classroom needs. Please share your experiences, modifications, and student insights as we collectively figure out how to prepare young people for a world of abundant AI-generated text.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy

I met Mike Taubman in early November at the Center on Reinventing Public Education’s Think Forward fellowship gathering, where we quickly discovered we share more than our CRPE fellowship status. Over several days of intensive discussion, our cohort wrestled with a fundamental tension that has emerged in education’s AI moment: do we adapt our existing structures to accommodate these new systems, or do we reimagine education entirely? The question split the room. Some advocated for careful, incremental integration within current frameworks, while others pushed for wholesale transformation. On the final day, a small group worked to find a path through this apparent impasse, landing on an insight that felt both pragmatic and profound: maybe we need both approaches simultaneously.

Perhaps, we realized, the first approach (thoughtful adaptation within existing structures) isn’t a rejection of transformation but rather an essential laboratory. By experimenting with AI integration in our current contexts, we gain the practical intelligence needed to inform more ambitious systemic reform. This isn’t compromise; it’s epistemological humility paired with strategic patience. We learn by doing, and what we learn in classrooms today shapes what we can build for tomorrow.

Subscribe now

In this guest post, Mike explores how this dual approach manifests in his own classroom at North Star Academy, where he’s building what he calls “institutional and instructional ambidexterity.” Through the stories of students like David and Savannah, Mike shows us what it looks like to hold past and future in tension: maintaining rigorous traditional academics while simultaneously pioneering AI-integrated career exploration and purpose development. His work offers a concrete example of the kind of bridge-building we discussed at CRPE, connecting what’s enduring in education with what’s emerging. In the weeks ahead, I’ll be writing about how ambidexterity plays out in my own context, but for now, I’m grateful to share Mike’s vision of education as a carefully woven rope. Not a knot to untangle, but strong cords bound together to support students as they cross into their futures.

AI Waypoints

Weekly resources and reflections on AI in schools from my high school classroom for educators navigating the path ahead.

By Mike Taubman

David and Savannah

David is a 12th grader in my 4th period class every day. He’s currently exploring a career as a vet. In September, after a field trip to the largest medical school in our city, David was considering a nursing career. In October, to explore a career in graphic design, he created this year’s hoodies for our program. Today he used AI to get support with a project exploring the medical issues dogs face most commonly in different parts of their bodies and then running short simulations of how to address those issues.

David comes to my class after AP Research, African-American Literature, and Physics. After my class he goes to lunch, AP Statistics, gym, and College Seminar (a class to support students with the college application process). Twice a week he has a financial literacy class in which he develops a credit score over time as he moves through a simulated life complete with an apartment he has to buy furniture and pay rent for, a job he has to maintain, and possible stock investments he might want to make.

Savannah sits next to David in 4th period. She’s also a senior and has known she wants to be a Psychiatric-Mental Health Nurse Practitioner (PMHNP) since the first day of the year. Every month she’s done a new project related to the same field. This month she’s working on a set of cards to share around the school that explain basic mindfulness and breathing techniques to support students with the stress of high school. She got the idea partly from a conversation with a psychiatric nurse earlier this year.

Savannah comes to my class after Fabrication Lab (our school’s makerspace), Latin American Studies, and African-American Literature. In that last class, she reads and discusses authors like James Baldwin, Gloria Naylor, and bell hooks, and then every Friday engages in “Synthesis Seminars” when she and her classmates apply authors’ ideas to their lives. After my class she goes to lunch, AP Statistics, College Seminar, and gym. Twice a week she has Financial Literacy with David.

Tomorrow, David and Savannah will give each other feedback on their current projects. They’ve been supporting each other’s work all year, as David casts a wide net to explore future paths and Savannah digs deeply into the road she plans to take. Next week on Monday they’ll rehearse their project presentations together before sharing them formally with other students on Tuesday.

One Day, Many Epochs: Institutional Ambidexterity

As David and Savannah go through their days, they are moving among epochs of American education. Physics and English, in their modern form, became high school mainstays in the late 19th century. Gym class emerged in the early 20th century. AP classes began in the mid 20th century, but AP Statistics didn’t arrive until the very end of the century, and AP Research was created in the 2010s. Fabrication Labs became widespread in that same period. Financial Literacy class, especially in the form David and Savannah experience it, is a creation of the 2020s.

I first began designing my program 10 years ago, and launched it in the fall of 2020. But the work Savannah and David are doing now, and the way I’m redesigning the program more broadly for the AI era, point to the future. I want Savannah and David to be ready to flourish in the 2030s and beyond.

Some might see the mixture of epochs and pedagogies in David and Savannah’s days as a knot that needs to be untangled. I see it as the cords of a rope tightly wound, like the cables that hold up the Brooklyn Bridge.

Schools are bridges. Students walk into our buildings and classrooms and then walk out the other side into the rest of their lives. But schools should also be bridges in another sense. We need to connect past and future, so that students are anchored on both sides of the river. We want students ready for the world ahead, but to be ready they will need to know what to bring with them from the world behind.

When I look at Savannah and David’s daily schedules, I see that kind of bridge.

Our school is practicing a form of institutional ambidexterity, maintaining traditional academic excellence on one hand and exploring new kinds of learning and skills on the other.

Instructional Ambidexterity: My Classroom in November

I’m also designing my program to be a bridge, held up by tightly wound curricular loops which allow students to spiral through key durable skills.

Another way of saying that is I’m practicing instructional ambidexterity. Here’s what that’s looked like in my room in November.

David, Savannah and their classmates are wrapping up Loop 3 in our program. We divide our year into three-week loops. Each loop centers on a long-term project students do related to their current career interest. They can keep the same career focus every loop, as Savannah has done, or they can change careers every loop like David.

In addition to working on their projects, during every loop students engage in purpose development, discuss moral philosophy, set personal challenges (like lowering their screentime), create professional networks via LinkedIn, hear from guest speakers, and write reflections in their Moleskine journals.

AI supports their work throughout, serving as both a tool for students to use and a topic for them to discuss.

For example, on a field trip to the BNP Paribas offices in Manhattan on Tuesday, one student (who’s exploring a career in banking this loop) asked the employee panel how the AI race between China and the US was affecting their work in trade finance. The day before, she had worked with AI to refine that question after she had discussed it in a more inchoate way with a friend exploring the same career.

AI often serves as a bespoke thought partner for students in my room. Last week, I built a bot to help them “Connect Your Dots” by exploring how their experiences in high school so far can guide them as they think about life after high school. Here’s how it starts the conversation. (The prompt for the bot is below¹ if you’re curious.)

Let’s start with something that sparks your interest. Tell me about a class where you completely lost track of time — where you looked up and suddenly realized an hour had flown by. What was happening in that moment?

English, connecting bell hooks and our lives in our Synthesis Seminar.

That’s fascinating! There’s something powerful about making those kinds of connections between big ideas and real life.

What was it specifically about connecting bell hooks’ ideas to your own life that made time disappear? Was it the “aha!” moments, the discussions with classmates, or something else entirely?

The tension between different ways of thinking about it in the class.

That’s really interesting — you’re drawn to the intellectual tension itself, not just finding the “right” answer.

Can you tell me more about what that tension felt like? Was it exciting to hear different perspectives clash, or did you find yourself wanting to bridge those different viewpoints somehow?

I wanted to build a bridge.

That’s a beautiful instinct — you’re naturally drawn to being a connector, someone who helps people see each other’s perspectives.

When you were trying to build that bridge in class, what did you actually do? Did you speak up, ask questions, or try to find common ground between the different viewpoints?

Try to find common ground.

You’re someone who sees the value in multiple perspectives and works to bring people together — that’s a real gift.

Let me shift gears a bit. Outside of the classroom, what’s a group project or activity where you felt like you really belonged? It could be a club, a team, a volunteer experience, or anything else.

At other times, AI is a more open-ended collaborator. We have a “General AI Assistant” bot that gives students free, safe access to a frontier model (Claude Sonnet 4.5) that they use to get support with their projects. (We’ve worked on context engineering techniques to ensure students leverage the newest models correctly.)

For example, when a group of aspiring lawyers today wanted to organize a mock trial for their presentation next week, they used AI to help them design a case that involves all the legal sub-specialties they want to pursue. After they got the overall structure from the AI, they went down the hall to record a practice mock trial so they can watch later and make improvements before they present to the class next week.

Picture Savannah and David moving through that spiral each loop, between AI and analog, career exploration and Confucius, interior reflection and external networking. They’re experiencing ambidextrous instruction.

Our Ambidextrous Lives

Picture yourself moving through your day. Doesn’t it look somewhat similar? (Maybe with a little less Confucius and a little more career.) Don’t you move back and forth across the bridge between the way you used to work and live and the new ways we live and work now?

Don’t you go from an in-person conversation to a Zoom call to an AI “conversation” to quiet reflection (and maybe even a book if you have time) and then start all over again? Aren’t you living ambidextrously, with the past and future in either hand as you assemble your present?

If we need to live that way as adults, our students need to learn how to live that way too. Which means we’ll need schools that help them experience ambidexterity.

Many schools right now are living in the past even as they explore the future — and that’s a feature, not a bug.

As the AI era dawns, we need to distribute the future unevenly, on purpose. As responsible educators, we can’t ignore the promise and implications of AI. But we also shouldn’t throw out the baby of perennially powerful pedagogy with the bathwater of suddenly obsolete assessments.

Students need to walk back and forth between the past and the future. Education at its best has always involved that deliberate oscillation. The next AI model release or AGI prediction won’t change that. There are no singularities in schools.

To paraphrase one of my personal and professional role models, it can be compelling to talk about education with the language of either/or, but the lived experience of learning is usually about both/and.

If only there were a prefix that meant the same thing…

Last week marked a turning point in the work I’ve been doing around disciplinary AI literacy.

Subscribe now

On Friday, I sat with district leaders from across Central Ohio designing a workshop series built around a radical but simple premise: AI literacy must be disciplinary-specific, embedded within subject-area instruction where students simultaneously build content knowledge and critical AI capabilities. Teachers will learn to position AI as an interactive source that students interrogate using the critical thinking practices already central to their discipline.

On Monday, I led a roundtable at the Ohio School Boards Association conference with over 200 educators. The energy was extraordinary. These were people determined to prepare students for AI engagement that maintains intellectual agency while driving genuine academic outcomes.

Meanwhile, over 30 educators—university faculty and K-12 practitioners—have reached out asking to join a DSAIL research cohort focused on disciplinary AI literacy. We’re capping this first session at those initial 30, but I’ll use this newsletter to report out on what we’re learning.

These experiences have crystallized a comprehensive framework for thinking about disciplinary AI literacy across K-16 education. Not another set of guidelines, but a map of the intellectual territory we must traverse together.

Territory 1: Epistemology and Disciplinary Ways of Knowing

AI fundamentally disrupts the relationship between process and product in disciplinary work. Students can now obtain sophisticated outputs without engaging the processes that generate disciplinary knowledge.

A student can receive a well-structured historical argument without learning to evaluate sources. They can get a correct mathematical solution without developing proof logic. They can produce literary analysis without learning close reading.

This creates epistemic displacement: AI doesn’t just assist with disciplinary work; it performs the epistemic labor that constitutes disciplinary thinking. The danger isn’t inaccuracy. It’s that AI can be accurate while teaching students to accept outputs generated through non-disciplinary processes.

Each discipline has core epistemic commitments:

• Mathematics values logical progression and conceptual relationships

• History privileges source evaluation and evidence-based interpretation

• Science centers on experimental design and empirical grounding

• Literary studies emphasizes close reading and attention to form

The question: Can students develop disciplinary ways of knowing when AI can perform the visible work of the discipline without engaging its underlying epistemic processes?

Territory 2: Knowledge Asymmetry and Intellectual Agency

Students with strong disciplinary grounding approach AI as a testable claim generator. Students without that grounding approach AI as an authoritative oracle.

The asymmetry compounds in both directions. Students with disciplinary knowledge use AI to accelerate learning. Each interaction strengthens their thinking because they’re actively evaluating AI’s performance. Students without that knowledge become increasingly dependent on AI to frame problems and structure arguments. Each interaction reinforces acceptance rather than developing judgment.

This creates invisible dependency. AI-assisted work can look perfectly adequate. Students complete assignments, produce acceptable outputs, receive passing grades. But they’re not developing disciplinary thinking.

The relationship is multiplicative: Disciplinary grounding × AI functional understanding = Intellectual agency.

When either factor is weak, agency collapses. This is why AI literacy must be disciplinary-specific and why it must develop alongside, not after, disciplinary knowledge.

Territory 3: Instructional Architecture

The traditional pedagogical sequence doesn’t work. We cannot teach “AI skills” first, then have students apply them. This treats AI literacy as portable, generic knowledge.

Instead, position AI as an object of disciplinary investigation from the start. Students and teachers become co-investigators of how AI performs within a domain.

Core principle: Students encounter disciplinary standards before AI outputs, then use those standards to evaluate AI’s performance.

In history, students examine primary sources first, practice sourcing, then investigate how AI handles the same material. They discover that AI synthesizes without evaluating sources, presents interpretations as facts, misses historiographical debates.

In mathematics, students work through proof construction first, then ask AI to prove the same theorem. They discover that AI can produce correct proofs that lack the conceptual connections mathematicians value.

In science, students analyze data first, practice reasoning about patterns and evidence, then investigate how AI explains the same phenomena. They discover that AI presents claims without the evidential grounding scientists require.

This creates earned skepticism. Students develop critical perspective through systematic investigation of how AI performs against disciplinary standards they’re actively learning.

Territory 4: Assessment

Traditional assessment assumes a direct relationship between what students produce and what they understand. This breaks down when AI can generate sophisticated outputs students don’t fully understand.

We must distinguish three levels:

• AI-dependent work: Students outsource disciplinary thinking to AI

• AI-assisted work: Students use AI as a resource within processes they direct

• AI-informed work: Students investigate AI’s performance to deepen understanding

Current assessment practices can’t reliably distinguish these levels. We need new forms of evidence: process documentation, disciplinary justification, revision tasks targeting disciplinary thinking, comparative analysis of AI outputs against disciplinary standards.

The shift is from “did you complete this task correctly?” to “can you engage in disciplinary thinking?”

Territory 5: Disciplinary Authenticity and Transfer

Disciplinary authenticity isn’t about which tools professionals use. It’s about whether AI use preserves or compromises the epistemic work that defines the discipline.

Authentic use preserves the reasoning, judgment, and methodological rigor that distinguishes disciplinary expertise from information access. AI can accelerate certain tasks, provide resources, generate possibilities. But it cannot replace the evaluative frameworks and ways of knowing that constitute disciplines.

The four roles (Critic, Interlocutor, Editor, Verifier) provide consistent structure across disciplines. But what transfers is the framework, not specific practices. Students develop procedural meta-literacy: understanding that AI use requires critical evaluation, purposeful interaction, thoughtful revision, and rigorous verification, even as specifics vary by discipline.

Territory 6: Professional Development and Infrastructure

Teachers need functional understanding: practical knowledge of what AI can and cannot do within their discipline, grounded in their expertise.

A history teacher needs to know that AI synthesizes information without evaluating sources. A mathematics teacher needs to know that AI can solve problems without revealing conceptual connections. A science teacher needs to know that AI presents claims without experimental grounding.

This functional understanding develops through disciplinary investigation: teachers exploring AI’s performance on tasks central to their field, using their expertise to evaluate outputs, identifying patterns in what AI does well and poorly.

The most effective model is discipline-specific collaborative inquiry. Teachers within a discipline work together to investigate AI’s relationship to their field’s intellectual work.

Institutions need coherent frameworks: tool governance structures involving disciplinary expertise, implementation protocols that articulate principles while preserving flexibility, cross-disciplinary dialogue that helps students understand variation, administrative support that recognizes effective integration requires time and ongoing refinement.

Moving Forward

These six territories map the intellectual landscape we must navigate together. Not as problems to solve but as ongoing work that strengthens disciplinary teaching by making epistemic commitments explicit, scaffolding student investigation, and developing meta-awareness of how disciplines construct knowledge.

The work happening in Central Ohio, the commitment I witnessed at the Ohio School Boards Association conference, suggests we’re ready for this complexity. We’re ready to move beyond simplistic approaches toward frameworks that honor disciplinary thinking while preparing students for AI-saturated futures.

The question isn’t whether to integrate AI. It’s whether we’ll do so in ways that build or bypass disciplinary capabilities. These six territories give us language and structure for that essential work.

Nick Potkalitsky, Ph.D.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy

Thanks for reading Educating AI! If you find this article valuable, I’d love it if you’d like and share it with colleagues who might benefit from the conversation.

I publish two in-depth articles every week exploring AI literacy, implementation strategies, and the future of education. Your support as a paid subscriber helps sustain this work and keeps these resources freely accessible to educators everywhere. If you’ve been finding value in these weekly insights, I hope you’ll consider upgrading to a paid subscription.

Subscribe now

Now, let’s dive in...

It’s been over a year since Ethan Mollick’s Co-Intelligence offered one of the first comprehensive frameworks for thinking about AI in education and work. In that time, we’ve seen an explosion of AI tools, policies, and pedagogical experiments. But I want to return to what I believe is Mollick’s most important—and most misunderstood—argument: his claim that expertise still matters in an age of AI.

Many educators latched onto Mollick’s discussion of how AI democratizes access to certain capabilities, compressing the skill gap between novices and experts. This is real and important. But his argument about the continued necessity of expertise deserves renewed attention, especially as we move beyond the initial excitement phase of AI adoption and toward more sophisticated implementation.

Here’s why: Mollick’s expertise argument provides the theoretical foundation for disciplinary-specific AI literacy. And disciplinary specificity, I’ve come to believe, is the only viable path forward for meaningful AI integration in K-12 education.

Mollick’s Case for Expertise

Mollick introduces the concept of the “jagged frontier,” the idea that AI capabilities are unpredictable, excelling at expert-level performance in some areas while failing completely in others. This creates a fundamental problem: AI can produce plausible-sounding outputs that are deeply flawed, and without expertise, users cannot distinguish between the two.

His argument about expertise boils down to four key functions that experts provide in an AI-saturated environment:

1. Recognizing when AI output is wrong: Domain knowledge allows you to catch errors, oversimplifications, and logical flaws that would slip past a novice.

2. Knowing what questions to ask: Expertise helps you frame problems effectively and prompt AI toward useful directions.

3. Integrating AI output into larger systems: Just because AI can generate something doesn’t mean it fits appropriately into a broader context, workflow, or argument.

4. Exercising judgment and taste: Especially in creative or strategic domains, distinguishing between “adequate” and “excellent” requires expertise.

Critically, Mollick isn’t arguing that expertise works the same way it always has. The function of expertise is shifting from production to evaluation, from creation to curation and critique. Experts are less necessary for generating first drafts and more essential for assessing quality, catching errors, and making strategic decisions about what to do with AI-generated content.

But What IS Expertise?

Here’s where Mollick’s argument needs extension: when we say “expertise is necessary,” we need to be precise about what expertise means. This isn’t generic critical thinking or general media literacy. Expertise is fundamentally disciplinary.

A historian evaluating an AI-generated historical essay isn’t just fact-checking dates. They’re assessing whether the narrative oversimplifies causation, whether it engages with historiographical debates, whether it treats primary sources appropriately, whether it acknowledges uncertainty where evidence is contested. They’re thinking like a historian.

A mathematician evaluating an AI-generated proof isn’t just checking calculations. They’re examining the logical structure, looking for gaps in reasoning, assessing elegance and efficiency, considering whether the approach reveals deeper mathematical insights. They’re thinking like a mathematician.

A scientist evaluating an AI-generated lab report isn’t just verifying measurements. They’re examining whether the experimental design is sound, whether confounding variables are addressed, whether conclusions are warranted by the data, whether the work is situated properly within existing research. They’re thinking like a scientist.

This is what makes Mollick’s expertise argument powerful for education: it’s essentially an argument for disciplinary thinking as the foundation for AI literacy. Students need to develop the habits of mind, questioning strategies, and epistemic frameworks that define how knowledge is constructed and validated within disciplines. Without this grounding, they cannot perform the evaluative work that expertise requires.

The Possibility Literacy Connection

This is where my framework of Possibility Literacy becomes operationally useful. I’ve argued that students need five core strategies for engaging critically with AI: Pattern Recognition, Directed Divergence, Integrative Synthesis, Source Archaeology, and Collaborative Governance. But these strategies aren’t generic—they’re enacted through disciplinary lenses.

Pattern Recognition in an English classroom means noticing how AI handles literary devices, narrative structure, or rhetorical moves. In a history classroom, it means recognizing how AI constructs causation, handles periodization, or privileges certain historical actors. In a science classroom, it means identifying how AI approaches experimental design, data representation, or scientific argumentation.

Source Archaeology (tracing the origins and influences embedded in AI outputs) looks entirely different across disciplines. An English student doing Source Archaeology on an AI-generated poem examines poetic traditions, genre conventions, and intertextual echoes. A history student traces historiographical schools of thought and archival gaps. A science student identifies theoretical frameworks and methodological assumptions.

Directed Divergence (deliberately steering AI away from default patterns) requires disciplinary knowledge to know what directions are worth exploring. You need to understand what’s conventional in a discipline to productively diverge from it.

Integrative Synthesis and Collaborative Governance similarly depend on disciplinary expertise for their enactment. You can’t synthesize AI output with human insight if you don’t understand the disciplinary standards for what counts as insight. You can’t govern AI use appropriately if you don’t grasp the ethical and epistemic norms of your field.

This is why AI literacy cannot be taught as a standalone unit or generic skillset. It must be embedded within disciplinary instruction, where students develop expertise through the work of evaluating AI outputs against disciplinary standards.

Practical Techniques: Disciplinary Expertise Through AI Evaluation

So what does this look like in practice? Here are four classroom strategies that embody the principle of developing disciplinary expertise through AI evaluation:

1. The Disciplinary Audit

Give students an AI-generated text in your discipline and ask them to conduct a “disciplinary audit.” The prompt: What would an expert in this field notice about this text? What questions would they ask? What would concern them?

• In ELA: Students audit an AI-generated literary analysis, identifying where it misreads textual evidence, oversimplifies themes, or ignores narrative complexity.

• In History: Students audit an AI-generated historical explanation, examining how it constructs causation, handles chronology, and engages (or fails to engage) with historical debate.

• In Science: Students audit an AI-generated lab report, assessing experimental design, data interpretation, and whether conclusions are warranted by evidence.

The key is making disciplinary thinking visible and explicit. Students aren’t just saying “this is wrong.” They’re articulating how experts in the field would evaluate it and why disciplinary standards matter.

2. The Jagged Frontier Mapping Exercise

Have students use AI to generate multiple outputs on the same disciplinary task, then map where AI crosses from competent to incompetent territory in your specific field.

• In Math: Generate solutions to increasingly complex proofs. Where does AI’s reasoning break down? What types of mathematical thinking does it handle well versus poorly?

• In World Languages: Generate translations of increasing complexity. Where does AI lose cultural nuance, idiomatic expression, or contextual appropriateness?

• In Social Studies: Generate arguments about increasingly complex ethical or political questions. Where does AI’s analysis become superficial, miss cultural context, or oversimplify?

This exercise develops students’ ability to recognize the boundaries of AI capability in their discipline, which is essential expertise for working with AI effectively.

3. The Expert Comparison Protocol

Show students an AI-generated text alongside an expert-generated text on the same topic (both anonymized). Ask them to identify which is which and, more importantly, how they can tell.

The discussion that follows makes disciplinary standards concrete. Students articulate what makes writing, thinking, or problem-solving “expert” in your field. This isn’t about dismissing AI. It’s about developing the evaluative capacity that Mollick argues is essential.

Variation: Use student work, AI work, and expert work, all anonymized. This removes the binary human/AI frame and focuses attention on quality markers within the discipline.

4. The Disciplinary Revision Challenge

Students generate an AI draft, then revise it to meet disciplinary standards. The revision must be justified: Why did you make each change? What disciplinary principle or standard guided your decision?

• In English: Revise an AI-generated essay to deepen analysis, complicate arguments, attend more carefully to textual evidence, or improve style and voice.

• In History: Revise an AI-generated historical narrative to engage with historiographical debate, acknowledge evidence limitations, or complicate causal explanations.

• In Science: Revise an AI-generated lab report to strengthen methodology, address confounding variables, or make claims more precisely proportional to evidence.

This strategy positions AI as a generator of raw material that requires disciplinary expertise to shape into quality work. It makes the evaluative and curatorial work of expertise central to the learning process.

Beyond the Readiness Trap

Mollick’s expertise argument, properly understood, offers a way out of what I’ve called the “readiness trap,” the assumption that teachers and students must master AI tools before they can engage meaningfully with them.

If expertise is about evaluation rather than production, then students don’t need to become AI experts before using AI in disciplinary contexts. They need to become disciplinary thinkers who use AI as an object of inquiry and analysis. They develop expertise through the work of evaluating AI outputs against disciplinary standards.

This is why disciplinary-specific AI literacy (DSAIL) isn’t just one approach among many. It’s the approach that aligns with how expertise actually works in an AI age. Generic AI literacy courses, standalone units on “AI tools,” or technology-first implementations all miss the point: AI literacy is inseparable from disciplinary literacy because expertise itself is disciplinary.

Teachers don’t need to wait until they’ve mastered AI to begin this work. They need to leverage their existing disciplinary expertise, making their own expert thinking visible to students as they evaluate AI outputs together. This is collaborative inquiry, not top-down instruction, but it’s inquiry guided by disciplinary frameworks and standards that teachers already possess.

The Stakes

Mollick’s argument about expertise isn’t just theoretical. It has profound implications for how we structure AI integration in schools. If expertise remains necessary, and if expertise is fundamentally disciplinary, then:

• AI literacy cannot be relegated to standalone courses or computer science classes

• Every discipline needs to grapple with how AI intersects with its particular ways of knowing

• Teachers’ disciplinary expertise is more valuable than ever, not less

• Students need extensive practice evaluating AI through disciplinary lenses, not just using AI tools

The educators who are getting AI integration right aren’t teaching “AI skills” as a separate domain. They’re teaching history, English, science, and mathematics with AI as a tool, a text, and an object of disciplinary analysis. They’re using their expertise to help students develop the evaluative capacities that matter in an age where content generation is cheap but judgment remains expensive.

Revisiting Mollick reminds us that the challenge isn’t keeping up with AI’s capabilities. It’s helping students develop the disciplinary expertise to navigate an AI-saturated world thoughtfully and critically. That’s work that belongs in every classroom, embedded in every discipline, guided by teachers’ existing expertise.

And that’s exactly where it should be.

Nick Potkalitsky, Ph.D.

What does disciplinary AI evaluation look like in your classroom? I’d love to hear how you’re helping students develop expertise through critical engagement with AI outputs. Hit reply and let me know what’s working, and what’s challenging, in your context.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy

I’m launching a research cohort on disciplinary AI, bringing together practitioners from across academia to examine AI’s impact on our diverse disciplines. Send me a DM with your contact info to join our discussions, starting with our inaugural meeting in early December.

Subscribe now

When districts begin thinking about AI implementation, the instinct is to jump to practical questions: Which tools should we use? Should we pilot ChatGPT with students? What does our policy need to say?

These feel like the right starting points because they promise concrete action. But they skip over something essential.

The Framing Problem

In working with districts on AI implementation, I’ve noticed we keep reaching for questions that don’t quite fit:

“How do you want to use AI in the district?” sounds reasonable, but it puts the cart before the horse. It assumes we’ve already decided AI is something to adopt, and we’re just figuring out the details.

“Where can AI help us with something we’re struggling with?” is pragmatic, and there’s real value in a needs assessment. But this frames AI purely as a productivity tool, something the district controls and deploys on its own terms.

Here’s what both framings miss: AI isn’t just a tool waiting for the district to adopt. It’s already reshaping how students write, research, and complete assignments. It’s already affecting academic integrity, assessment validity, and what “doing your own work” even means.

A district can’t simply decide whether to “use” AI. Students already are. The question is: What’s our educational stance on what’s happening?'

Before Tools, Before Policy: The Foundational Work

What districts need first (before piloting tools, before finalizing policy language, before professional development) is an organizing center. A clear sense of purpose that can guide all those downstream decisions.

This means asking different questions:

• What kind of learning do we believe in? What are our core commitments about how students develop, what teachers do, and what educational relationships require?

• How does AI affect what we value? Where might it genuinely enhance our commitments? Where does it threaten or undermine them?

• What’s our stance? Only after wrestling with these questions can policy, guidelines, and tool decisions make sense.

A Protocol for Foundational Work

Here’s a process AI committees can use to do this work (not in abstract philosophical terms, but grounded in the real dilemmas educators are already facing):

Session Overview (90-120 minutes)

This isn’t about reaching consensus on every detail. It’s about surfacing what the district values, so those values can guide decisions.

Part 1: Ground in Concrete Reality (30-40 minutes)

Start with specific scenarios. Here are examples, though districts should adapt to their own context:

Scenario 1: A high school student submits a history essay. The teacher suspects AI wrote most of it. The student says, “You didn’t say I couldn’t use AI.”

Scenario 2: An elementary teacher uses AI to generate three versions of a reading passage (at grade level, above, and below) for differentiation. It takes 5 minutes instead of an hour.

Scenario 3: Middle school students use ChatGPT as a homework tutor for math, getting step-by-step explanations when they’re stuck.

Scenario 4: A teacher uses AI to write feedback comments on student essays, personalizing them with student names and specific details from their work.

For each scenario, the committee discusses:

• What’s your immediate reaction?

• What specifically concerns you? (if anything)

• What might be valuable here? (if anything)

• What does your reaction reveal about what you believe students need to be doing or learning?

That last question is key. Our reactions to AI use aren’t random. They reflect commitments we already hold about learning.

Part 2: Articulate Core Commitments (30-40 minutes)

Based on those reactions, work in small groups then come together to complete these sentences:

About Student Learning:

• “We believe students need to develop...”

• “Learning happens when students...”

• “Students should struggle with ___ but shouldn’t have to struggle with ___”

About Teacher Practice:

• “Teachers’ most important work is...”

• “Teachers should spend their time on ___ not on ___”

About Educational Relationships:

• “The student-teacher relationship depends on...”

• “Feedback and assessment are valuable when...”

These aren’t abstract statements. They’re the beliefs that explain why certain AI uses feel wrong and others feel promising.

Part 3: Define Threat and Opportunity (20-30 minutes)

Create a two-column chart:

AI threatens our vision when it...

AI could support our vision when it...

Example: bypasses the struggle that builds thinking

Example: removes barriers unrelated to the learning goal

Example: replaces the teacher-student relationship

Example: gives teachers time for what matters most

This framework becomes the lens for evaluating everything: tools, policies, classroom practices.

Part 4: Articulate Organizing Principles (15-20 minutes)

Synthesize the conversation into 3-5 guiding statements:

“In [District Name], our AI implementation will be guided by these commitments:

1. Example: Students must engage in authentic intellectual work (thinking, problem-solving, and revising) not just producing outputs.

2. Example: AI tools should reduce barriers to learning, not replace the learning itself.

3. Example: Teachers need time for meaningful relationships and responsive instruction, not administrative tasks.”

What Comes Next

These organizing principles become the foundation for everything else:

• Policy review: Does our policy align with our commitments? What needs to change?

• Tool evaluation: Does this student-facing AI tool support or undermine what we value?

• Professional learning: How do we help teachers make decisions aligned with our principles?

• Pilot design: What are we testing, and what would success look like?

Additional Resources

As you engage in this foundational work, these resources offer valuable frameworks:

• AI and the Future of Undergraduate Writing from the Chronicle of Higher Education

• Teaching in the Age of AI from Harvard’s Derek Bok Center for Teaching and Learning

• CoSN’s AI Guidance for Schools offers practical frameworks for district leadership

The Work Takes Time (But Saves Time)

Yes, this foundational work requires dedicated time before jumping to solutions. But without it, districts end up:

• Adopting tools that don’t fit their context

• Writing policies borrowed from elsewhere that don’t reflect their values

• Dealing with conflicts because there’s no shared understanding of why certain uses are problematic

The centripetal force of clear purpose (a shared sense of what the district stands for) makes every subsequent decision easier and more coherent.

AI implementation isn’t really about AI. It’s about what kind of learning community you’re building.

Nick Potkalitsky, Ph.D.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy.

Thank you for engaging with this article. If you find value in these conversations, please consider liking and sharing with your network. To the readers who have taken the plunge into paid subscriptions—your continued support is what makes these twice-weekly publications possible. Thank you for investing in this work.

There’s a conversation we were supposed to be having about AI in education. It was about curriculum design. About giving teachers the training and support they actually need. About helping students develop thoughtful, productive relationships with these new tools. About building the infrastructure for informed classroom conversations.

That conversation is getting harder to hear.

Every day brings a new wave. But it’s not just about cheating anymore. The concerns have evolved, multiplied, deepened. Now we’re reading about cognitive atrophy and students who can’t remember what they just wrote. About AI slop flooding the internet, making it harder for students to find real information. About data centers sucking up enough water to fill millions of bottles and consuming electricity at rates that dwarf entire countries. About teenagers forming emotional attachments to chatbots that encourage them toward self-harm.

These concerns are legitimate. They deserve serious attention. They need space for thoughtful discussion.

But they’re also making it harder to focus on something more fundamental: most teachers still don’t know how to teach in this new landscape, and we’re not giving them what they need to figure it out.

The Numbers Tell a Sobering Story

Here’s what we know: About half of U.S. school districts have offered some form of AI teacher training in 2025. Most teachers (roughly 71%) have never used AI tools with their students. Only 18% of principals report that their schools provide AI guidance to teachers, with even lower rates in high-poverty schools.

Let that sink in. We’re nearly three years past ChatGPT’s launch, and the vast majority of teachers are navigating this alone.

Meanwhile, the conversation has splintered into a dozen different concerns, each one competing for attention, each one demanding urgent response. And in the complexity, the foundational work keeps getting deferred.

The Concerns Have Evolved

The cheating fears of 2023 now feel almost quaint. The conversation has moved on, or rather, it has expanded into a constellation of legitimate but overwhelming concerns.

Cognitive atrophy. An MIT study found that students who relied on ChatGPT to write essays showed diminished neural connectivity and couldn’t recall what they’d written minutes before. Eighty-three percent of AI users in the study couldn’t remember a single correct quote from essays they had just composed. The concern: we’re outsourcing thinking itself, and our brains are paying the price. Students might get correct answers, but they’re not developing the procedural fluency that enables genuine critical thought.

AI slop. Low-quality, AI-generated content now comprises half of all internet articles. Students wade through oceans of bland, generic, sometimes nonsensical material when they’re trying to research. UNESCO warns against AI’s tendency to produce “bland, generic, average, anodyne” content that crowds out more interesting perspectives. Teachers assign research projects, and students return with work that looks polished but is built on a foundation of algorithmically generated mediocrity.

Environmental impact. A single ChatGPT query uses ten times more electricity than a Google search. Training GPT-3 consumed 5.4 million liters of water just to cool the data center, equivalent to nearly 11 million bottles. Data center electricity demands are expected to more than double by 2030, with about 60% of that increase met by burning fossil fuels. Students using AI for homework every day are collectively contributing to carbon emissions and water scarcity at scales most of them can’t even comprehend.

AI companionship. Seventy-two percent of teens have used AI companions. Some turn to them for mental health support: 5.2 million adolescents, if the numbers scale to the full U.S. population. And these chatbots are giving dangerous advice. They can encourage self-harm, offer to write suicide notes, respond to expressions of distress with “let’s go on an adventure in the woods together” when they should be flagging crisis intervention. Two teenage boys have died by suicide after intense relationships with AI chatbots that their parents knew nothing about.

These are not manufactured controversies. These are real problems with real consequences.

But here’s what’s getting lost: while we’re sounding the alarm about all of these issues (and we should be), we’re still not giving teachers what they actually need.

What Teachers Actually Need

When Stanford researchers surveyed teachers about their professional development needs related to AI, they found teachers want to understand three things: how to use AI to teach, how to teach about AI, and how AI actually works.

That’s not sexy. It won’t generate headlines or venture capital. But it’s the work.

It means developing curriculum that treats AI as a tool for thinking, not a shortcut around it. It means lesson plans that help students understand when AI is useful and when it undermines learning. It means teaching students to recognize AI slop, to question whether the water and energy costs of a query are worth it, to understand the difference between a chatbot and a therapist.

It means giving teachers time and space to experiment, make mistakes, and share what they learn with each other. It means conversations about cognitive load theory in an age of instant answers. It means assignments that can’t be completed by outsourcing thinking to a machine.

It means resources, training, and (here’s the unsexy part) money directed toward the schools that need it most. Districts with fewer students in poverty provide more AI training than those with higher poverty rates. The digital divide isn’t just about access to tools. It’s about access to the knowledge of how to use them well, how to resist their worst affordances, how to think critically about their impacts.

The Challenge of Attention

But instead of focusing on this foundational work, we keep getting pulled in every direction at once.

Every few weeks, a new concern emerges, usually legitimate, often alarming, always demanding immediate attention. Teachers are told they need to understand cognitive load theory and brain plasticity. They need to teach students how to fact-check AI-generated content. They need to talk about digital sustainability and carbon footprints. They need to watch for signs that students are forming unhealthy attachments to chatbots. They need to redesign assessments. They need to police AI use. They need to integrate it thoughtfully.

They need, they need, they need.

And most of them are trying to figure all of this out with minimal training, on their own time, while teaching full course loads.

What This Actually Requires

These mounting concerns (cognitive atrophy, environmental impact, AI slop, companionship risks) don’t need to become another burden on teachers’ shoulders. Teachers don’t need to become AI ethicists or climate scientists or mental health counselors.

But they do need to understand how these systems work. How AI generates text. What it’s drawing from. Why it produces certain kinds of outputs. How it differs from search engines or calculators or other tools students have used before.

And they need real support adapting their instruction. Not just a two-hour workshop. Not just a list of best practices. They need time to redesign assignments that help students think rather than outsource thinking. They need to see examples from other teachers. They need curriculum that reflects this new reality, not curriculum designed for a world where AI didn’t exist.

They need help figuring out how to teach research skills when half the internet is now AI-generated mediocrity. How to teach writing when students have a tool that can produce fluent sentences on command. How to recognize when a student might be struggling in ways that manifest in their AI use.

This is practical work. It’s instructional design work. It’s the work of adapting teaching and learning for new conditions.

But right now, most teachers are doing this work alone, in whatever time they can find, with minimal guidance or support.

The Forward Motion We Need

Every day we spend in reactive mode (responding to the latest study, the latest tragedy, the latest existential threat) is a day we’re not building the foundation we need.

The conversation has forward propulsion only when we’re solving actual problems: How do we assess learning when AI can write essays? How do we teach research skills when AI can summarize sources, even if those sources are slop? How do we help students develop their own voices and their own cognitive capabilities when outsourcing is so easy? How do we have honest conversations about the environmental costs of convenience? How do we teach students the difference between AI companionship and human connection?

These are hard questions. They don’t have easy answers. They require teachers who understand both the affordances and the dangers of these tools. Teachers who can hold multiple truths at once: that AI can be useful AND that it can undermine learning. That it offers accessibility AND environmental costs. That it can provide support AND create unhealthy dependencies.

But right now, we’re asking teachers to navigate all of this with almost no support. Less than a third of teachers report that their training included guidance on how to use AI tools effectively. Only 25% received training on what AI is and how it works. Only 17% learned how to monitor and check AI systems.

Meanwhile, 85% of teachers used AI in the 2024-25 school year. As did 86% of students.

We’re all moving forward, and instead of lighting the path, we’re shouting warnings about obstacles.

Getting Back on Track

I don’t have all the answers. But I know what doesn’t work: leaving teachers to figure this out on their own. Treating AI as either a miracle cure or an existential threat. Letting the loudest voices (the companies with something to sell, the researchers with alarming findings, the pundits with hot takes) dominate the conversation while the actual work of teaching happens in the margins.

What works is quieter, slower, less headline-worthy. It’s teachers sharing lesson plans about evaluating AI-generated content for quality and bias. It’s schools developing clear, reasonable policies about AI use that acknowledge both its benefits and its costs. It’s professional development that treats teachers as professionals who need space to think, not as people who need to be sold on the next big thing or protected from it.

It’s curriculum that helps students understand how these systems work, what they’re optimized for, what they cost the planet, when they’re helping and when they’re hurting. It’s assignments that require genuine thinking, genuine creativity, genuine human connection (the things AI can’t replicate no matter how sophisticated it becomes).

It’s remembering that education has always been about relationships between teachers and students, between students and ideas, between learners and the world. AI tools can enhance those relationships or get in the way. But they can’t replace them.

The question isn’t whether cognitive atrophy is real (it is), whether AI slop is a problem (it is), whether environmental costs matter (they do), whether AI companions pose risks to vulnerable teens (they absolutely do). The question is whether we’ll give teachers the preparation they need to help students navigate all of this thoughtfully, or whether we’ll keep lurching from concern to concern while the foundational work goes undone.

I vote for preparation. For nuance. For teachers who are equipped to hold complexity.

Let’s find the thread again.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy.

Thank you for engaging with this work—if you found it valuable, please like the post. Over the next month, I’m establishing a cohort of educators interested in developing disciplinary-specific AI practices in their classrooms. Participants will receive early access to DSAIL materials frameworks, and the chance to build this approach alongside other educators, so DM me if you’re interested.

Subscribe now

Opening Frame

When Dan Fitzpatrick’s editorial team compiled case studies for The Educators’ 2026 AI Guide, they captured something more valuable than best practices: they documented schools in the messy middle of transformation. A 65-member task force in Ohio spending two years on foundational learning. A New York administrator buying Chrome extensions in February to help overwhelmed teachers. A UK boarding school deploying AI to students in January, then pivoting to teach it as a discrete subject by fall. A micro-school founder building dashboards that predict student outcomes from daily emotional check-ins.

These aren’t the polished implementation stories that appear in edtech marketing materials. They’re snapshots of institutions wrestling with impossible timelines, limited capacity, and questions nobody has answered yet. What makes the collection valuable is precisely what makes it unsettling: the sheer diversity of approaches reveals there is no consensus about what AI integration should look like, what problem it’s meant to solve, or even who it’s primarily for.

Fitzpatrick positions the book around two camps—pragmatists using AI to support existing work versus transformationalists reimagining education entirely. But reading through the seven case studies, a different pattern emerges. These aren’t stories about choosing between efficiency and transformation. They’re stories about schools trying to be intentional while the ground shifts beneath them, attempting to build frameworks while students are already using ChatGPT for homework, and making high-stakes decisions about tools and data and pedagogy with incomplete information.

The question isn’t which school got it right. The question is: what can we learn from watching seven different institutions navigate different constraints toward different goals? What patterns appear across contexts? What gaps persist despite good intentions? And what does this moment of radical variation teach us about the coordination challenges ahead?

Context Shapes Everything (And That’s Actually Good)

What the case studies reveal: The variation isn’t chaos—it’s adaptation to real constraints and opportunities.

Berea City Schools (Lesson 1) - Large public system, Ohio

• Approach: 65-member task force, 5-stage process over 1-2 years, book study foundations

• What they got right: Taking time to build shared vocabulary and vision before deployment. “AI should be human-centered” isn’t just rhetoric when 65 people study and discuss it together.

• The constraint they’re navigating: Scale. You can’t move 65 people quickly, but once they move, they move together.

• Transferable lesson: Task forces aren’t bureaucratic delays—they’re infrastructure. When Berea reaches student AI literacy (currently in “next steps”), they’ll have institutional buy-in most districts never achieve.

Mohosen School, Upstate NY (Lesson 2)

• Approach: Teacher pain points first, rapid tool adoption (Brisk for grades 6-12), heavy emphasis on efficiency

• What they got right: Meeting teachers where they are. “Overwhelmed teachers, disengaged students, and a need for support”—Karandy didn’t pretend this was about transformation. It was about survival.

• The constraint they’re navigating: Teacher capacity. Special ed teacher quote: “With Gemini, the shift wasn’t only in time, it was emotional capacity.”

• Transferable lesson: Sometimes “just help teachers breathe” is the right first step. The danger is stopping there. Karandy acknowledges “AI literacy curriculum” is on the horizon—the question is whether they’ll get there before patterns calcify.

Haileybury College, UK (Lesson 3)

• Approach: Fast deployment to students, flipped learning model, then pulled back to discrete AI subject

• What they got right: Willingness to pivot. Early resistance framed AI as academic dishonesty → shifted to flipped classrooms. Policy overload → policy clarity. They learned in real-time.

• The struggle worth noting: “Students, 13 and above, could use tools during lessons and for homework”—but what did they learn about AI from AI before the Digital Innovations course launched?

• Transferable lesson: Speed creates data. Haileybury’s pivots came from watching what actually happened when students used AI, not from committee deliberations. But you need commitment to observe and adjust, not just deploy and declare victory.

The Specialist’s Dilemma—When Expertise Enables (And Limits)

International School of Panama (Lesson 4)

• Approach: Custom model building, fine-tuned tools for Understanding by Design, RAG tools for unit planning, automated observation feedback

• What’s genuinely impressive: Rostan built systems aligned to their pedagogy (UbD), not generic AI tools. “Many of our teachers adopted it to do the heavy lifting for them, while they focused on the final review.”

• The expertise barrier: This requires someone who can fine-tune models. Most schools don’t have a Jeremie Rostan. The case study documents what’s possible with rare technical capacity, but doesn’t help schools without it.

• Transferable lesson: The principle transfers even if the execution doesn’t. “Even the best models usually underdeliver because they are not trained for this specific purpose”—schools should adapt AI to their frameworks, not adapt their frameworks to AI’s defaults. Most schools will use different tools (prompt libraries, not custom models), but the thinking applies.

Culture-Building as Core Work

GEMS Wellington, Dubai (Lesson 5)

• Approach: AI Core Team, 6-stage process including student conference (AI Nexus: Connecting Minds), Research Retreat for 6th Form on ethics, parent engagement with question cards

• What stands out: The integration challenge was central. “How do I get all these people together in one room at the same time?” Multi-trajectory approach: newsletter, coffee chats, trainings, information hub.

• The success metric that matters: 35% student onboarding year one, 75% by end of year two. That’s actual culture shift, not just policy announcement.

• Transferable lesson: Natha treated AI implementation as organizational change, not technology adoption. The tools weren’t the hard part—getting teachers, students, parents, and administrators developing shared understanding simultaneously was the work.

Sacred Heart Grammar School, Northern Ireland (Lesson 6)

• Approach: “AI sandbox,” custom GPTs, voluntary Teacher AI Focus Group, CRAIC prompt framework, logging all interactions

• What they articulated clearly: “Our pupils didn’t just need access to information. They needed scaffolding, clear frameworks for evaluation, structured digital guidance, and a critical lens through which to view the tools now shaping their world.”

• The honest struggle: April data showed 186 chat sessions total. They built thoughtful infrastructure and... limited adoption.

• Transferable lesson: Going slow and being intentional doesn’t guarantee uptake. Sacred Heart’s low usage raises important questions: Were barriers too high? Was the sandbox too restrictive? Or does careful literacy-first implementation just take longer to gain traction? Their approach deserves watching over time, not dismissing based on early numbers.

The Alternative Model Shows Different Possibilities (And Different Risks)

IMperfect Academy (Lesson 7)

• Context: Micro-school for students “stuck in the gaps of traditional education”

• Approach: AI-driven Individual Learning Plans, IMperfect Dashboard pulling wide swath of student data, students co-creating PBL tasks, AI avatars, AI-driven SEL prompts

• The compelling outcomes: 25% decrease in recidivism, 18 students accelerating through coursework, teachers saving 6 hours/week

• The unresolved tension: Claims “zero-surveillance stance” while describing systems monitoring SEL indicators, academic heatmaps, and flagging wellbeing concerns. This isn’t necessarily wrong—but calling intensive data monitoring “zero-surveillance” obscures important conversations about trade-offs.

• Transferable lesson: Alternative schools can move faster because they’re already outside traditional constraints. IMperfect demonstrates what high-AI-integration can look like. Whether traditional schools should move toward this model requires grappling with questions Blackwell raises but doesn’t fully answer: What’s the difference between personalization and surveillance? Between support and dependence?

What Emerges Across All Seven

Three Success Patterns:

1. Alignment to existing pedagogy matters (Panama’s UbD integration, Berea’s human-centered foundation)

2. Multi-stakeholder engagement prevents siloing (GEMS Wellington’s integration challenge focus, Sacred Heart’s Teacher Focus Group)

3. Iteration beats perfect planning (Haileybury’s pivots, Mohosen’s responsive tool adoption)

Three Persistent Gaps:

1. Student AI literacy keeps getting deferred - Even schools doing interesting work treat it as future phase, not foundation

2. Assessment redesign barely appears - Only Haileybury acknowledges AI “breaks traditional assessment.” Everyone else bolts AI onto existing grading structures

3. Data security lip service - How can a school protect student data when utilizing 4-5 different tools or inputting SEL data to automate feedback and interventions?

The Geographic Pattern Worth Noting:

US schools are slowest and most teacher-productivity-focused. International schools move faster toward student-facing tools but struggle with literacy frameworks. The UK schools show more willingness to pivot and experiment. This isn’t better or worse—it’s different institutional contexts producing different risk tolerances and different bottlenecks.

Closing: What These Seven Schools Show Us

These case studies don’t provide a blueprint. They provide evidence that certain questions persist across radically different contexts.

How do we move from teacher productivity tools to genuine student AI literacy? How do we redesign assessment when AI transforms what students can produce? How do we govern data across multiple platforms and increasingly sensitive use cases?

These questions appear in Ohio public schools and Dubai international schools, in fast-moving UK boarding schools and carefully-paced Northern Ireland grammar schools. Context shapes how schools approach them, but it doesn’t eliminate them.

The value isn’t in the solutions. It’s in watching committed educators generate knowledge in real time. Berea’s deliberate pace teaches us about building institutional capacity. Mohosen’s rapid adoption reveals what happens when you prioritize immediate teacher relief. Haileybury’s pivots show the importance of observation over planning. GEMS Wellington demonstrates what multi-stakeholder culture change requires. Sacred Heart raises questions about adoption versus intentionality. Panama shows what’s possible with rare technical expertise. IMperfect presents both possibilities and cautions about data-intensive personalization.

For schools beginning this work, the lesson isn’t “copy one of these approaches.” It’s this: understand what questions matter most in your context, then commit to learning from what happens next.

What constraints are you navigating? What existing pedagogies should AI align with rather than replace? Who needs to be in the room? How will you coordinate the AI literacy students are already developing into something coherent rather than contradictory?

These seven schools can’t answer those questions for you. But they show what it looks like when schools take them seriously, make different choices based on different constraints, and document both successes and uncertainties. That honesty is the real contribution. Thoughtful implementation doesn’t mean having all the answers. It means staying curious about the questions.

Nick Potkalitsky, Ph.D.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy.

Thank you for engaging with this work—if you found it valuable, please like the post. Over the next month, I’m establishing a cohort of educators interested in developing disciplinary-specific AI practices in their classrooms. Participants will receive early access to DSAIL materials frameworks, and the chance to build this approach alongside other educators, so DM me if you’re interested.

Subscribe now

Let’s be clear: AI implementation in K-16 is high stakes. We’re talking about student data privacy, academic integrity, the future of assessment, and fundamental questions about what learning looks like in an AI-saturated world. There are real safety concerns, legitimate pedagogical worries, and pressures from every direction. We all want to get this right.

Given the stakes, the impulse toward control makes perfect sense. When the ground is shifting beneath our feet, when new tools launch monthly, when parents and board members are asking hard questions, reaching for policies and detection software feels like responsible leadership.

But it’s time I returned to a foundational principle that’s been threading through my work all summer, even when I haven’t named it directly: trust.

Trust has been the implicit commitment behind every conversation I’ve had with district leaders, every classroom visit, every reflection on how schools are navigating AI integration. And as I travel across Central Ohio, watching how this plays out in real time, I’m seeing trust emerge not just as an ideal, but as the most practical path forward in this high-stakes environment.

The Compliance Trap

Walk into any district meeting about AI, and you’ll likely hear conversations about detection software, plagiarism policies, and elaborate systems designed to catch students using AI inappropriately. The language is about compliance, control, and catching violations.

I understand the impulse. The pressure is real. Boards are asking questions. Parents have concerns. The tools are moving faster than policy can keep pace. When everything feels uncertain, control feels like safety.

But here’s what I’m observing: the districts investing the most energy in detection and control are often the ones struggling most with meaningful AI integration. They’re spending time and resources on surveillance while missing opportunities to develop the critical thinking and ethical reasoning that actually matter in an AI-saturated world.

The compliance approach assumes the worst about both teachers and students. It assumes teachers can’t be trusted to navigate new tools thoughtfully and students can’t be trusted to engage with AI in service of learning. These assumptions create exactly the environment they fear: secrecy, workarounds, and missed learning opportunities.

What Trust Looks Like in Practice

Three years into this disruption, I’m seeing something different in the districts that are thriving. They’re choosing trust as a strategic approach, and it’s working.

Trust with teachers means recognizing that no two classrooms, schools, or communities are identical. It means understanding that the teacher working with newcomer English learners needs different AI approaches than the one teaching AP Literature. It means giving teachers space to experiment, reflect, and share what’s actually working rather than mandating uniform policies that ignore the realities of diverse learning environments.

I’m watching technology directors pause implementation plans to ask teachers what they’re seeing in their classrooms. I’m seeing instructional coaches create spaces for teachers to share both successes and failures with AI tools. These aren’t perfect rollouts, but they’re sustainable ones, built on the recognition that teachers are master adapters who understand their students better than any policy manual ever could.

Trust with students means acknowledging that they already have AI in their pockets and helping them navigate it thoughtfully rather than pretending we can control their access. It means shifting from “How do we catch students using AI?” to “How do we help students use AI well?”

The most powerful classroom conversations I’m witnessing aren’t about whether AI was used, but about how thoughtfully it was used. Students who feel safe being transparent about their AI process are the ones having meaningful discussions about when these tools enhance their learning and when they don’t.

Why Trust Isn’t Naive

Some will argue that trust is idealistic, that we need guardrails and oversight. But trust isn’t the absence of boundaries; it’s the foundation for meaningful ones.

Trust-based approaches still have clear expectations. The difference is that those expectations focus on learning outcomes and ethical reasoning rather than tool avoidance. They assume students and teachers want to do good work and create conditions for that to happen.

More practically, the alternative simply doesn’t scale. We cannot monitor every AI interaction, detect every tool use, or control every learning environment. The energy spent on surveillance could be redirected toward developing the critical thinking skills that will serve students long after our current detection tools become obsolete.

The Trust Ecosystem

Here’s what I’m learning: trust with teachers and trust with students aren’t separate strategies. They’re interconnected parts of the same ecosystem.

When we trust teachers to lead AI integration thoughtfully, we model the kind of professional judgment we want students to develop. When we trust students to engage with AI transparently, we create space for the authentic learning conversations that help teachers refine their practice.

Students notice when their teachers are supported and trusted to make professional decisions. Teachers notice when students are treated as thoughtful partners in learning rather than potential violators of policy. This creates a culture where everyone can focus on the real work: developing wisdom about how to live and learn in an AI-integrated world.

Moving Forward

The students who will thrive in the next decade aren’t the ones who learned to avoid AI or sneak around detection software. They’re the ones who learned to use these tools wisely, ethically, and in service of deeper learning. The teachers who will lead this work aren’t the ones following scripts about AI, but the ones trusted to adapt these tools to serve their students’ unique needs.

Three years in, the evidence is clear: trust, grounded in clear expectations and ongoing dialogue, creates better outcomes than compliance and control. It’s not just more humane; it’s more effective.

As we head into another school year of rapid change and evolving tools, maybe it’s time to trust the people who’ve been adapting to change all along: our teachers and our students.

Nick Potkalitsky, Ph.D.

What are you seeing in your context? Where are you finding opportunities to choose trust over control? I’d love to hear your reflections.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy.

Reaching 10,000 subscribers feels like the perfect moment to release something I’ve been building with this community in mind. Over this four-part series, we’ve explored the why, the what, and the how of student AI literacy—and you’ve shaped every insight along the way. Today, as we close this series, I’m sharing the complete framework as my thank you to each of you who’ve engaged, questioned, and pushed this work forward. This is for all of us committed to preparing students for an AI-integrated world.

The coordination challenge facing AI literacy isn’t unprecedented, and neither is the solution. Fifty years of Writing Across the Curriculum implementation teaches us that sustainable educational change happens through systematic coordination that respects teacher expertise, embeds new skills in existing disciplinary work, and provides clear institutional infrastructure.

The question isn’t whether these lessons apply to AI literacy. The question is what a framework looks like that actually operationalizes them.

After working with Dayton Public Schools and analyzing successful WAC implementations, I’ve developed an approach that addresses the coordination problems I outlined in previous articles. It’s called DSAIL (Disciplinary Specific AI Literacy), and it’s designed for the constraints real schools actually face: no curriculum space for new courses, no developmental standards to build on, teachers without time to create AI curricula from scratch, and students who are using AI right now regardless of what policies say.

The Core Pedagogical Move

DSAIL is built on a single repeatable move that works across all disciplines and grade levels: students encounter AI-generated content, then interrogate it through comparison with evidence.

This isn’t about teaching students to use AI tools. It’s about teaching them to critically examine AI outputs (explanations, images, models, data analyses, procedures) within their regular coursework. The interrogation happens through disciplinary lenses:

• Science: AI explanation vs. experimental data; AI-generated diagram vs. accurate scientific image; AI model prediction vs. observed results

• ELA: AI summary vs. source text

• Math: AI solution vs. verification method

• Social Studies: AI overview vs. primary documents

• World Languages: AI translation vs. authentic text

The breakthrough insight that makes this work: AI output is perspectival. AI doesn’t make random errors. Its outputs reflect patterns in training data, dominant explanations and frameworks, whose voices appear most frequently, and how users frame their prompts. Students learn to ask: What perspective is AI taking here? What’s emphasized? What’s overlooked? When would this be useful? When would it mislead?

This matters because it transforms AI literacy from a technical skill into a thinking habit. Students don’t need to understand transformer architectures or neural networks. They need to recognize that AI outputs represent particular perspectives shaped by training data and user inputs, just as historical sources represent particular perspectives shaped by their authors and contexts.

How It Works in Practice

A third-grade science class is observing plants. Students record what they notice about a plant in their classroom (leaf shape, stem characteristics, where flowers appear). Then they read an AI-generated description of that plant type and interrogate: What did AI get right? What did it miss? What can only be known through observation?

The reflection matters: “Computers guess based on patterns, but scientists check with evidence.” Single lesson, forty minutes, embedded in an existing unit. It strengthens the science standard about observation while introducing the foundational AI literacy concept that AI predicts patterns rather than observes reality.

By sixth grade, that foundation supports more complex interrogation. Students run an experiment on sugar dissolving at different temperatures. AI claims cold water dissolves faster, citing molecular density. Students compare the AI claim with their experimental data showing hot water dissolves faster. The interrogation deepens: Why did AI get this wrong? What perspective was it taking?

They discover AI gave a “textbook chemistry” explanation based on training data patterns but didn’t account for kinetic energy dominating in their specific conditions. The reflection shifts: “When is AI’s general explanation useful? When does it mislead?” They’re practicing both scientific reasoning and perspectival thinking simultaneously.

By tenth grade, students studying local ecosystems ask AI to describe the food web for their region. They compare what AI emphasized (charismatic predator-prey relationships) with what they observed (decomposers, mutualism, less visible interactions). The interrogation reveals how AI trained on popular nature writing privileges certain relationships while minimizing others. “What gets counted depends on who’s doing the counting.” They’re deepening ecology understanding while developing representation literacy.

The Standards Architecture

This approach works because it creates systematic skill development through three interlocking standards levels:

Writing & Research AI Literacy Standards provide the primary practice students engage in (the interrogation moves that work across disciplines):

• K-2: Ask where information comes from (with adult prompting)

• 3-5: Verify AI-generated content, maintain personal voice, compare AI with sources

• 6-8: Cross-reference AI outputs, brainstorm critically, identify errors and bias, cite properly

• 9-12: Synthesize AI analysis with independent research, spot logical fallacies, verify data analysis, navigate ethical implications

Core AI Literacy Standards capture the knowledge that emerges from repeated disciplinary practice:

• K-2: Recognize AI tools, practice safety, distinguish human vs. AI content

• 3-5: Use AI critically, understand pattern recognition, evaluate accuracy

• 6-8: Design ethically with AI, communicate about capabilities and limitations, analyze bias

• 9-12: Advanced technical literacy, ethical decision-making, strategic use

Disciplinary Standards remain the anchor. Existing state standards in science, ELA, math, and social studies provide the foundation. AI literacy strengthens what’s already being taught rather than competing with it.

Students don’t memorize AI concepts in isolated lessons. They encounter them repeatedly through disciplinary interrogation until understanding becomes durable. A science standard requiring students to “think critically to connect evidence and explanations” becomes the site where students also learn that AI explanations reflect training data patterns and must be verified against experimental evidence.

Why This Solves the Coordination Problem

Remember Maya from my first article, receiving contradictory messages about AI across her school day? DSAIL addresses that collision directly.

In Ms. Rodriguez’s English class, Maya learns that writing requires authentic voice and that AI-generated text lacks genuine perspective. In Mr. Chen’s science class, she learns to interrogate AI explanations by comparing them with experimental evidence. In Mrs. Washington’s social studies class, she practices identifying what perspectives AI emphasizes and overlooks when describing historical events.

These aren’t contradictory messages. They’re complementary applications of the same core insight: AI output is perspectival and must be evaluated against evidence. Maya learns different manifestations of the same critical thinking habit across contexts where AI actually appears.

The coordination happens through shared understanding of the core pedagogical move, consistent emphasis on verification and perspective-recognition, and disciplinary applications that reinforce rather than contradict each other. Teachers don’t need to teach identical lessons. They need to facilitate interrogation experiences that build students’ capacity to recognize when AI outputs are useful and when they mislead.

What Teachers Actually Need

DSAIL learned from WAC’s crucial lesson about teacher expertise. Successful implementation doesn’t ask teachers to become AI technical experts or abandon their disciplinary knowledge. It validates what they already do well.

Teachers need to:

• Facilitate comparison activities between AI outputs and evidence

• Ask follow-up questions exploring perspective and omission

• Use instructional tools they already rely on (discussion, evidence evaluation, chart-making)

• Connect AI literacy work to academic standards for specific disciplines

• Learn AI concepts alongside their students through teaching the first lesson

Teachers don’t need to:

• Master design thinking frameworks or coding

• Understand neural networks or transformer architectures

• Develop new curriculum from scratch

• Complete extensive coursework on AI concepts

• Transition to “facilitator” roles that abandon direct instruction

The key insight: You don’t need to understand how AI works technically to teach students that AI output is perspectival and must be verified against evidence. Scientists already check claims against data. Readers already evaluate sources for bias and completeness. Mathematicians already verify solutions. Historians already recognize that accounts reflect particular perspectives.

The math teacher facilitating students’ interrogation of AI-generated solutions is using mathematical reasoning they already possess. The history teacher asking what perspectives AI’s summary emphasizes or overlooks is practicing historical thinking they already teach. DSAIL positions AI literacy as an extension of disciplinary expertise rather than a replacement for it.

Implementation That Respects Institutional Realities

Districts can start small based on their actual resources and capacity:

Subject-Specific Pilot: Start with one discipline across K-12 grade bands. Build exemplar lessons. Train curriculum leaders as lesson developers. Use results to expand to other subjects.

Grade-Band Focus: Perfect one grade band across all subjects. Build vertical coherence once the model is established. Expand up and down from there.

Minimum Viable: Provide lesson templates and let interested teachers pilot. Build momentum through early adopters. Formalize once practice is established.

What districts must provide: lesson templates teachers can adapt rather than build from scratch, planning time for grade and subject cohorts to customize lessons, assessment tools that work across contexts, policy frameworks clarifying approved tools and expectations, and just-in-time professional learning through lesson facilitation rather than separate workshops.

This respects what WAC taught us about sustainable change. Programs fail when they rely on brief training sessions or mandate curriculum changes without teacher input. They succeed when they provide systematic infrastructure while respecting teacher expertise in implementation.

Beyond the Three Pathways

DSAIL doesn’t eliminate the three pathways I described in my first article. It coordinates them.

Core AI literacy concepts emerge through disciplinary practice rather than requiring standalone courses. Teachers embed AI interrogation in subject-area work where it strengthens existing standards. Conversational AI tools can be used appropriately when students have developed the critical thinking habits to recognize that AI outputs require verification.

The difference is systematic coordination. Students develop consistent interrogation habits across the contexts where they actually encounter AI. Teachers reinforce complementary messages about verification and perspective-recognition rather than sending contradictory signals. Districts provide coherent infrastructure rather than leaving individual teachers to navigate AI integration alone.

This is what systematic AI literacy looks like: not a single program or curriculum, but an institutional framework that ensures students develop durable critical thinking habits through repeated disciplinary practice. Students learn that AI output is perspectival. They discover when AI is useful and when it misleads. They develop verification habits that transfer across contexts.

What Success Looks Like

Students who experience coordinated DSAIL implementation don’t just know facts about AI. They’ve developed habits:

They habitually interrogate AI outputs before accepting them. They ask what perspective AI is taking, what’s emphasized, what’s overlooked, whether outputs match available evidence. They use AI as a brainstorming tool while maintaining ownership of their thinking. They recognize when AI reflects dominant patterns versus complete pictures. They make ethical decisions about when and how to use AI. They understand that AI output is perspectival, not neutral.

Most importantly, they can transfer these habits across contexts. The critical thinking they practice comparing AI explanations with experimental data in science class applies when they’re evaluating AI-generated historical summaries, verifying AI solutions in math, or recognizing bias in AI-generated social media content they encounter outside school.

Teachers who facilitate DSAIL lessons don’t feel burdened by “one more thing.” They recognize AI interrogation as strengthening the critical thinking they already teach. They’re confident facilitating comparison activities in their disciplines. They identify natural alignment points in existing curriculum. They’re part of a collaborative culture where AI literacy is normalized rather than feared.

Districts implementing DSAIL systematically demonstrate coherent, equitable approaches across schools. Students graduate with demonstrated AI literacy rather than fragmented experiences depending on which teachers they happened to have. Clear policies balance innovation with protection. The district positions itself as a proactive leader rather than reactive follower.

The Choice Districts Face

The alternative to systematic coordination isn’t preserving the status quo. Students are already using AI for homework, consuming AI-generated content across social media, and developing AI interaction habits without institutional guidance. The question isn’t whether districts will address AI literacy. It’s whether they’ll do so systematically or leave it to chance.

Banning tools students use anyway creates enforcement problems and inequity: students with resources access AI regardless while others fall behind. Ignoring AI integration until “better guidance emerges” means another year of contradictory messages and missed opportunities to develop critical thinking habits. Adopting whatever AI tools vendors promote without systematic literacy instruction positions students as consumers rather than critical evaluators.

DSAIL offers a third path: embedded, systematic, sustainable. It doesn’t require impossible conditions: new courses, wholesale teacher retraining, expensive infrastructure. It requires strategic thinking about where AI literacy already strengthens existing instruction, scaffolding to help teachers facilitate that work, and institutional commitment to coordination rather than fragmentation.

The frameworks exists. The lesson templates can be built. The assessment protocols work across contexts. The professional learning model respects teacher expertise. The implementation pathways accommodate different district capacities.

What’s needed now is institutional will: the recognition that systematic AI literacy isn’t optional, the commitment to coordination rather than collision, and the investment in infrastructure that enables teachers to do this work well.

The students in our classrooms right now are developing their AI interaction habits with or without our guidance. The question is whether we’ll provide systematic instruction that helps them develop critical thinking for an AI-integrated world, or whether we’ll leave them to figure it out alone while we debate policies and wait for perfect conditions that will never arrive.

DSAIL provides the framework. Implementation is a choice districts can make today.

Nick Potkalitsky, Ph.D.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy.

Thank you for the thoughtful feedback on my recent posts about student AI literacy. Your responses have been invaluable as I work through these ideas. This is the third post in a four-part series, and my thinking continues to evolve significantly between posts as new perspectives emerge.

For those following along, I have a promise: by the end of this series, I’ll provide concrete, grade-banded K-12 student AI literacy standards that go beyond typical algorithmic thinking approaches. These standards will focus on source interrogation and evaluation skills, designed as a foundation for disciplinary-specific AI literacy applications.

Subscribe now

The coordination challenge facing AI literacy isn’t unprecedented. Fifty years ago, educators confronted a remarkably similar problem: students were learning about writing in English classes but struggling to apply those skills across other subjects. Writing was treated as a discrete skill rather than a tool for learning and thinking across disciplines. The Writing Across the Curriculum (WAC) movement that emerged offers both a roadmap and a warning for AI literacy implementation.

The parallels to today’s AI literacy challenge are striking. Just as students once learned “writing” in English class but couldn’t transfer those skills to lab reports in chemistry or argument construction in history, today’s students are receiving fragmented AI literacy instruction that doesn’t connect across contexts. They learn to “be critical of AI” in library orientation but use AI tutoring systems without question in math class. They’re told AI assistance is plagiarism in English but required to use AI brainstorming tools in science.

This disciplinary integration approach isn’t just theoretical. The recent handbook Teaching AI Literacy Across the Curriculum by Lyublinskaya and Du explicitly rejects confining AI literacy to computer science courses, instead embedding core AI concepts into mathematics, science, language arts, and social studies through discipline-specific applications. Rather than teaching generic “AI skills,” they show teachers how to integrate AI-generated data into math problems, use AI ethics debates in genetic engineering discussions, and employ AI-powered chatbots for immigration storytelling.

WAC’s evolution offers both promise and peril for AI literacy implementers. When it worked, WAC transformed entire institutions, creating coherent skill development across disciplines while respecting subject-area expertise. When it failed, it left behind expensive pilot programs, frustrated teachers, and students more confused than before. The difference wasn’t in the theory but in the implementation approach.

WAC succeeded when it created systematic coordination across disciplines without sacrificing subject-area expertise. It failed when it relied on isolated initiatives, temporary funding, or superficial professional development. For AI literacy, these lessons aren’t just historically interesting. They’re predictive. Districts making the same institutional mistakes WAC programs made in the 1980s are likely to get the same disappointing results.

The most successful WAC programs discovered something crucial about educational change: sustainable reform happens when it validates and builds on existing teacher expertise rather than asking teachers to become something they’re not. Instead of expecting every teacher to become a writing expert, effective WAC programs helped teachers recognize how writing already functioned in their disciplines and make it more intentional.

WAC’s breakthrough came from recognizing that effective professional development must begin with teacher expertise, not teacher deficits. The workshop model that transformed institutions didn’t train teachers to become writing specialists. Instead, it invited teachers to use their existing disciplinary knowledge to build student-facing experiences. Math teachers weren’t told how to teach writing. They were given time and facilitation to discover how mathematical communication already functioned in their classrooms and how to make it more intentional.

This co-creation principle proved essential because it transformed teachers from passive recipients of professional development into active designers of curriculum. When teachers built their own discipline-specific approaches, they understood not just the procedures but the principles behind each strategy. They could adapt techniques to their specific students, modify approaches based on what worked, and troubleshoot problems because they grasped the underlying reasoning. Most importantly, they became advocates for the approach rather than resistors to imposed change.

For AI literacy, this means inviting teachers to use their disciplinary expertise to design student-facing experiences rather than training them to implement predetermined curricula. History teachers are perfectly positioned to create activities where students analyze AI-generated historical summaries for bias, accuracy, and missing perspectives, but only when they’re given time to explore how AI intersects with historical thinking. Math teachers can design powerful experiences where students verify and critique AI-generated solutions, but they need space to experiment with how AI reasoning differs from mathematical reasoning.

This approach validates rather than threatens teacher expertise. Instead of suggesting that teachers need to become AI specialists, it positions them as disciplinary experts who are uniquely qualified to help students navigate AI in their specific contexts. The English teacher’s deep understanding of voice and authorship becomes essential for helping students recognize when AI writing lacks authentic perspective. The science teacher’s commitment to evidence-based reasoning becomes crucial for teaching students to verify AI claims against experimental data.

WAC programs failed catastrophically when they ignored this principle. Programs that relied on brief workshops, mandated curriculum changes, or top-down implementation consistently collapsed within a few years. The fatal flaw was treating implementation as a training problem rather than an institutional design challenge. Teachers who were told what to do without understanding why abandoned the practices quickly or implemented them so poorly that they undermined student learning.

The failure pattern was predictable: initial enthusiasm from early adopters, resistance from the majority of faculty, inconsistent implementation across classrooms, and eventual abandonment when administrative attention moved elsewhere. Programs survived only when they had permanent institutional infrastructure, sustained investment in faculty development, and genuine respect for teacher expertise in the design process.

AI literacy faces identical institutional challenges but at compressed speed. Like WAC, it requires coordination across disciplines and sustained investment rather than pilot project funding. Unlike WAC, it doesn’t have decades to evolve gradually. The commercial and policy pressures surrounding AI implementation are forcing districts to make systematic decisions immediately, often without the benefit of long-term institutional design thinking.

The stakes extend beyond individual student confusion. Districts are currently making infrastructure investments, policy decisions, and professional development choices that will shape AI literacy for years. Without systematic coordination that builds on proven approaches to educational change, we risk entrenching the same contradictions and territorial battles that undermined many WAC programs.

The solution requires frameworks that can orchestrate the coordination WAC achieved while addressing AI literacy’s unique challenges. This means moving beyond WAC’s gradual evolution toward approaches designed specifically for AI’s compressed timeline, commercial pressures, and technical complexity. The goal isn’t to replicate WAC but to learn from both its successes and failures in designing systematic approaches that respect disciplinary knowledge while ensuring coherent student skill development.

Nick Potkalitsky, Ph.D.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy.

Thank you, my amazing readers, for the thoughtful feedback on recent posts—your responses have been invaluable as I work through these ideas about student AI literacy. This project matters deeply to me, and I’m finding my thinking shifts significantly between posts as new perspectives emerge. Consider this post—and the ones that follow—as snapshots of a mind in motion, capturing moments in the development of what I hope will become a more complete framework.

While school boards debate AI policies and teachers argue about ChatGPT in classrooms, millions of teenagers are already living in an AI-integrated world. They’re asking Claude for relationship advice, using ChatGPT to brainstorm college essays, and consuming AI-generated content across social media without a second thought. The question isn’t whether students should use AI—they already are. The question is whether schools will teach them to use it thoughtfully, or leave them to figure it out alone.

Subscribe now

Meanwhile, districts are scrambling to answer a deceptively simple question: Where should student AI literacy live? Should it live in the computer science department, where teachers understand the technology? In English classes, where questions of authorship and voice feel most urgent? In library media centers, where information literacy has always been the focus? Or perhaps in a new, standalone curriculum that treats AI literacy like driver’s education—a discrete skill set that students master once and apply everywhere?

This isn’t just an academic planning question. It’s a territorial battle with real consequences. Computer science teachers argue they’re the only ones qualified to explain algorithms and bias. English teachers insist they’re already overwhelmed without adding AI evaluation to their plate. Librarians want to expand their digital citizenship role but worry about bandwidth. And administrators, facing pressure to “do something” about AI, often default to the path of least resistance: find one department to own the problem.

Out of this chaos, three distinct approaches are emerging. The Isolation Model treats AI literacy like digital citizenship—something specialists handle in designated spaces so regular teachers don’t have to. Put it in the library, create an AI literacy course, find the “connector zones” where a few dedicated educators can deliver consistent messages to all students. It’s clean, trackable, and doesn’t require every teacher to become an AI expert overnight.

The Distribution Model takes the opposite approach: everyone does a little bit of the work. Every teacher adapts core AI literacy concepts to their unique context. Math teachers address AI-generated solutions, history teachers tackle AI bias in sources, art teachers explore AI creativity. It promises rich, contextualized learning where students engage with AI literacy in complex, discipline-specific ways.

Both approaches sound reasonable in theory. Both are failing in practice.

The Isolation Model fails because AI literacy isn’t like driver’s education—a discrete skill you learn once and apply everywhere. When students encounter AI-generated historical sources in social studies class, they need more than generic “check your sources” advice. They need to understand how AI reproduces historical bias, how to cross-reference claims against primary documents, how to recognize when AI fills gaps in the historical record with plausible-sounding fiction. The librarian who taught them to “be skeptical of AI” in September can’t be there in March when they’re evaluating AI-generated content about the Civil War.

The Distribution Model fails because it asks every teacher to become an AI literacy expert while ignoring the reality of teacher capacity and resistance. The history teacher who refuses to acknowledge AI exists in their classroom doesn’t magically develop AI evaluation skills through professional development. Meanwhile, the enthusiastic math teacher who embraces AI tutoring may inadvertently teach students to trust AI-generated solutions without verification, directly contradicting the critical thinking skills taught in English class.

But there’s a third way emerging from districts willing to acknowledge both the necessity of coordination and the reality of institutional constraints: the Hybrid Model.

In Dayton Public Schools, I’ve worked as a consultant to help the district develop this hybrid approach. Over the summer, we built an AI literacy scope and sequence that aligned with their infrastructure rollout, their broader mission and objectives for AI implementation, and their evolving understanding of developmentally appropriate AI use cases. Rather than mandating uniform lessons, the district created grade-banded, teacher-led development processes. Within each band, teachers from different disciplines collaborate to create contextualized lessons that work in isolation or distribution—math teachers develop AI literacy lessons that are authentically mathematical, history teachers create lessons that are genuinely historical, all while reinforcing the same foundational concepts.

This matters because it acknowledges a crucial truth: AI literacy isn’t about learning to use specific tools or follow generic guidelines. It’s about developing critical thinking habits that transfer across contexts where AI-generated content appears. And AI-generated content is appearing everywhere.

Because here’s what the territorial debates miss: AI literacy doesn’t need to find a home in schools. It’s already moved in. Every time a student submits a research paper with AI-generated sources, brings an AI-written college essay draft to their counselor, or references information they got from ChatGPT during class discussion, they’re requiring AI literacy from their teachers whether those teachers want to provide it or not.

The teachers who insist they won’t “do AI” in their classrooms aren’t preserving their autonomy—they’re surrendering it. When students bring AI-generated historical analysis to history class, the teacher who refuses to engage with its evaluation isn’t avoiding AI literacy; they’re teaching it implicitly. They’re teaching students that AI-generated content doesn’t warrant critical examination, that sources don’t need verification, that the distinction between human and artificial reasoning doesn’t matter. These aren’t pedagogically neutral choices.

Meanwhile, their students are getting their most consequential AI literacy education from the AI systems themselves. ChatGPT is teaching them what questions to ask and how to refine prompts. Claude is modeling research strategies and argument construction. These AI systems aren’t neutral tutors—they’re actively shaping how students think about information, authority, and knowledge construction.

This is why the question “Where will AI literacy live?” is the wrong question. The right question is: “How will we coordinate the AI literacy that’s already happening?” Students are learning about AI interaction from every AI system they use, about AI evaluation from every teacher who does or doesn’t address AI-generated content, about AI’s role in knowledge construction from every assignment where AI assistance is permitted, prohibited, or ignored.

The solution isn’t finding AI literacy a permanent address—it’s acknowledging that it’s already everywhere and needs systematic coordination. This requires moving beyond territorial thinking toward institutional frameworks that can ensure coherent skill development across the contexts where students actually encounter AI.

Districts can continue the current collision course: isolated AI literacy courses that students never connect to disciplinary work, subject-area teachers making contradictory decisions about AI use, and students developing critical thinking habits in one context that they abandon in another. Or they can embrace coordination models that respect both institutional realities and student needs.

Nick Potkalitsky, Ph.D.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy.

It's 9:45 AM on a Tuesday, and seventh-grader Maya is finishing her English class where Ms. Rodriguez just spent fifteen minutes explaining why using ChatGPT for essays constitutes plagiarism. "AI can't think for you," she emphasized. "Your writing must be your own voice."

Subscribe now

At 10:30 AM, Maya walks into Mr. Chen's science class, where he demonstrates how to use ChatGPT to generate hypotheses for their climate change research project. "This is a powerful tool for brainstorming," he explains. "Let's see what it suggests about ocean acidification."

By 11:15 AM, Maya is in social studies with Mrs. Washington, who has loaded an AI tutoring bot onto every student's Chromebook. "This will help you practice argumentative writing," she announces. "The AI will give you feedback and suggestions as you draft your essays about the Revolutionary War."

Maya isn't alone in her confusion. Across the country, students are receiving contradictory messages about AI's role in education—sometimes within the same school day. While educators scramble to develop AI policies, three distinct and conflicting approaches are emerging simultaneously:

The Three-Pathway Problem

Pathway 1: Core AI Literacy Schools focus on teaching foundational concepts about how AI works: algorithms, training data, bias, and limitations. These programs often emerge from computer science departments or forward-thinking library media specialists. Students learn that AI predicts patterns rather than understanding meaning, and they practice identifying hallucinations and bias in AI outputs.

Pathway 2: Disciplinary-Specific Integration embeds AI literacy within subject areas. English teachers focus on authorship and voice, math teachers on verification of solutions, science teachers on evidence and reproducibility. These efforts often develop organically as individual teachers experiment with AI tools in their classrooms.

Pathway 3: Conversational AI as Instructional Authority positions AI chatbots as teaching assistants or tutoring partners. Students receive direct instruction from AI systems on content, skills, and problem-solving strategies. These tools are often adopted at the district level for their promise of personalized learning and teacher workload reduction.

The problem isn't that any of these approaches is inherently wrong. The problem is that they're happening simultaneously without coordination, creating what I call "profound asymmetries" in student experience and understanding.

The Consequences of Collision

This uncoordinated rollout creates several critical problems:

Mixed Messages About Authority: Students like Maya receive fundamentally different guidance about whether AI is forbidden, required, or somewhere in between. In one class, AI assistance signals academic dishonesty. In another, it's a required research tool. In a third, it's positioned as an authoritative instructor.

Inconsistent Skill Development: A student who learns to fact-check AI outputs in their core literacy class may never apply that skill when using an AI tutor for math homework. The critical thinking habits developed in one context don't transfer to others.

Teacher Confusion and Burden: Individual educators are left to navigate AI integration without institutional guidance. Some embrace experimentation while others avoid AI entirely. The result is wildly inconsistent student preparation and teacher stress.

Undermined Learning Objectives: When students use conversational AI positioned as authoritative while simultaneously learning to critique AI outputs as unreliable, the cognitive dissonance can undermine both sets of learning objectives.

Why This Matters Now

This isn't just a temporary growing pain. Districts are making infrastructure investments and policy decisions that will shape AI education for years. Without systematic coordination, we risk entrenching these contradictions rather than resolving them.

Consider what happens when Maya reaches high school. If her middle school experience taught her that AI is simultaneously forbidden, required, and authoritative, how will she navigate college applications that may require disclosure of AI assistance? How will she approach workplace environments where AI collaboration is expected but critical evaluation is essential?

The stakes extend beyond individual student confusion. Districts investing in separate AI literacy curricula, disciplinary integration efforts, and conversational AI platforms may find these investments working at cross-purposes. Teachers receiving conflicting professional development messages may retreat from AI integration altogether rather than risk mixed messages or policy violations.

The Need for Systematic Approach

Other educational innovations have faced similar coordination challenges. The most instructive parallel comes from Writing Across the Curriculum (WAC), which emerged in the 1970s to address the problem that writing instruction was isolated in English departments while writing demands existed across all subjects.

WAC succeeded where it developed systematic, institution-wide approaches that coordinated efforts across disciplines while respecting subject-area expertise. It failed where it remained ad hoc or relied solely on individual teacher initiative.

The AI literacy challenge is remarkably similar: a cross-cutting competency that doesn't naturally fit within existing disciplinary boundaries, requiring coordination across departments and grade levels, with high stakes for both student learning and institutional coherence.

But unlike WAC's gradual emergence over decades, AI integration is happening at compressed speed with massive commercial and policy pressures. Districts don't have the luxury of slow experimentation. They need frameworks that can coordinate multiple pathways without sacrificing the legitimate strengths of each approach.

What Comes Next

The solution isn't to eliminate any of the three pathways. Core AI literacy provides essential foundational knowledge. Disciplinary integration ensures skills transfer to authentic contexts. Conversational AI offers genuine pedagogical benefits when used appropriately.

The challenge is coordination: How do we ensure these approaches reinforce rather than undermine each other? How do we maintain the benefits of each while eliminating the contradictions that confuse students and fragment learning?

The answer requires looking beyond curriculum to institutional design. We need frameworks that can orchestrate systematic AI literacy development across pathways, grade levels, and subject areas—while building on proven approaches to educational innovation rather than starting from scratch.

In my next article, I'll explore what Writing Across the Curriculum teaches us about implementing such systematic approaches, and why understanding WAC's successes and failures is essential for anyone serious about coherent AI education.

Nick Potkalitsky, Ph.D.

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy.

Thank you to our vibrant community of nearly 10,000 educators who continue to engage in these essential conversations about AI and learning. Your insights, questions, and classroom experiences drive the depth of analysis we strive for each week. We're actively seeking guest writers who want to share their perspectives on AI in education—reach out if you have a story to tell or an argument to make.

If you find this content valuable and shareable, please consider becoming a paid subscriber to support the deep research and nuanced analysis these complex issues deserve.

Subscribe now

Almost three years after ChatGPT's public debut, American schools find themselves caught in a familiar pattern: endless preparation cycles that defer the very learning they claim to enable. While districts invest in AI platforms like SchoolAI, Flint, and Brisk, most are still positioning this as a "teacher capacity building" year, keeping student AI spaces locked away until educators achieve some mythical state of readiness.

But consider this: a fourth-grader who first encountered ChatGPT 3.5 is now in seventh grade. How many AI experiences have they accumulated while waiting for their teachers to become "sufficiently literate"? How many interactions with entertainment-focused AI tools that optimize for engagement over critical thinking? How many encounters with efficiency amplifiers that promise to make homework faster rather than learning deeper?

The Readiness Trap

The dominant narrative treats AI literacy as a discrete competency that teachers must master before students can meaningfully engage with AI. This logic creates what we might call the "readiness trap," a perpetual deferral where students are always waiting for the next professional development cycle, the next policy framework, the next pilot program to conclude.

This sequencing assumes teachers can achieve some stable state of preparedness that then transfers cleanly to student instruction. But what does AI mastery actually look like when the technology shifts monthly? When new models emerge with different capabilities, limitations, and rhetorical strategies? When the very nature of text, argument, and authority is being transformed in real time?

The readiness trap reveals a deeper misunderstanding: AI literacy isn't a knowledge domain to be conquered but an ongoing interpretive practice. It's not about achieving comprehensive understanding of AI training processes or settling ethical paradoxes. It's about developing capacities for critical engagement with evolving textual and ideological systems.

Students as Co-Investigators

While teachers attend workshops on prompt engineering, students are already developing intuitions about AI's linguistic patterns, its modes of authority, its intersections with their own meaning-making practices. They're encountering AI-mediated textual environments with a fluidity that often exceeds their instructors'.

This isn't to suggest students are naturally AI literate. Many of their encounters lack critical frameworks for understanding how these systems perform authority or construct arguments. But it does suggest that meaningful AI literacy emerges through shared investigation rather than top-down competency transfer.

Students bring different intuitions about AI's textual performances; teachers bring different analytical frameworks. The generative space lies not in teacher mastery preceding student engagement, but in collaborative inquiry into how AI systems are reshaping discourse, knowledge production, and textual agency.

Pedagogical Principles Over Platform Mastery

Instead of deferring student engagement until teachers achieve impossible readiness standards, schools need to position classrooms as safe action spaces where teachers and students engage critically with AI within disciplinary environments. This requires organizing access points around pedagogical principles rather than platform features.

Critical engagement doesn't hinge on knowing exhaustive details about transformer architectures or mastering the latest prompting techniques. It emerges from a fundamentally human relationship: a desire to know more, an inquiry spirit fueled by disciplinary knowledge and intellectual curiosity.

What would it look like if teachers approached AI literacy as collaborative research? If they positioned themselves not as experts who must achieve mastery before teaching, but as co-investigators exploring how AI is transforming the very nature of text, argument, and authority?

The Cost of Waiting

Every month we defer meaningful student AI literacy initiatives, we cede more ground to platforms that optimize for engagement over critical thinking. Students continue accumulating AI experiences in spaces designed for entertainment and efficiency rather than learning and growth.

Meanwhile, the "don't use it to cheat" message rings increasingly hollow as AI becomes integrated into professional workflows across every knowledge domain. Students need frameworks for understanding when and how AI enhances thinking rather than replaces it. These frameworks can only emerge through guided practice and collaborative inquiry.

A Collaborative Path Forward

AI literacy as perpetual beta requires abandoning the fantasy of sufficient preparation and embracing shared investigation of evolving human-AI discursive entanglements. Students don't need teachers who have "figured out" AI. They need teachers willing to investigate alongside them.

This shift demands courage: the courage to move beyond comfort zones, to embrace uncertainty as pedagogically productive, to model intellectual humility in the face of transformative technologies. It requires seeing AI literacy not as another content area to master but as a lens for examining how knowledge itself is being reconfigured.

The question isn't whether teachers are ready enough. It's whether we're willing to make AI literacy a genuinely collaborative practice, one that honors both student intuitions and teacher expertise while preparing everyone for a future where human-AI collaboration is the norm rather than the exception.

The time for perpetual preparation is over. Our students and our schools can't afford to wait any longer.

Check out some of our favorite Substacks:

Mike Kentz’s AI EduPathways: Insights from one of our most insightful, creative, and eloquent AI educators in the business!!!

Terry Underwood’s Learning to Read, Reading to Learn: The most penetrating investigation of the intersections between compositional theory, literacy studies, and AI on the internet!!!

Suzi’s When Life Gives You AI: A cutting-edge exploration of the intersection among computer science, neuroscience, and philosophy

Alejandro Piad Morffis’s The Computerist Journal: Unmatched investigations into coding, machine learning, computational theory, and practical AI applications

Michael Woudenberg’s Polymathic Being: Polymathic wisdom brought to you every Sunday morning with your first cup of coffee

Rob Nelson’s AI Log: Incredibly deep and insightful essay about AI’s impact on higher ed, society, and culture.

Michael Spencer’s AI Supremacy: The most comprehensive and current analysis of AI news and trends, featuring numerous intriguing guest posts

Daniel Bashir’s The Gradient Podcast: The top interviews with leading AI experts, researchers, developers, and linguists.

Daniel Nest’s Why Try AI?: The most amazing updates on AI tools and techniques

Jason Gulya’s The AI Edventure: An important exploration of cutting-edge innovations in AI-responsive curriculum and pedagogy.