Teaching AI Fluency.
A working framework for teachers and students who want to use AI without losing their minds, their integrity, or their critical faculties.
Students are already using AI. Most are using it badly, not because they're cynical, but because nobody has shown them what good use looks like. Teachers feel the same gap on their own side of the desk; the tools arrived faster than the training, and the conversation about how to handle them in a classroom has been mostly vibes.
This course is one attempt at a fix. It's built on the AI Fluency Framework, four competencies (Delegation, Description, Discernment, Diligence) that map what people actually do when they work with AI, and what they should do to keep that work honest. We'll move through all four in two parallel tracks: one for faculty, one for students. The teacher track comes first; teachers should be at least a few steps ahead of the kids they're teaching. One big caveat: I'm one guy in a community of very smart people. Please let me know about any insights, ideas, or crazy notions you might have. Also, feel free to let me know when something is unclear or just plain dumb. It wouldn't be the first time I've heard such a thing. Happy learning! --Mark Mitchell
Choose a starting point
Start the faculty track
Six modules. Worked classroom examples, talking points, the honor-code conversation, subject-specific demos. Designed for an in-service or for working through on your own before you teach this to students.
StudentsStart the student track
Six modules calibrated for high schoolers. Same framework, different examples. Interactive exercises throughout. Most modules take ten to fifteen minutes.
If you're new to the framework altogether, read The Framework in Five Minutes first.
The 4Ds.
Four competencies. They're not a sequence so much as a loop you keep walking around; each one feeds the others.
The AI Fluency Framework was developed by Rick Dakan at Ringling College and Joseph Feller at University College Cork, in collaboration with Anthropic. It identifies four interconnected competencies people need to use AI effectively, efficiently, ethically, and safely. It's deliberately agnostic about which tools you use. Claude, ChatGPT, Gemini, Flint: the framework holds.
Should I use AI at all?
Setting goals and deciding whether, when, and how to engage with AI. The decision before the decision.
How do I ask?
Effectively describing your goals so the AI behaves usefully. Prompting, but bigger than prompting.
Is this any good?
Critically assessing what the AI gave back. Catching the hallucinations, the bias, the polished nonsense.
Can I stand behind it?
Taking responsibility for what you do with AI, being transparent, vouching for the final product, owning the consequences.
Three modes of interaction
The framework distinguishes three modalities of human-AI interaction. They sit on a spectrum of how much agency you hand over.
Automation. You tell the AI exactly what to do; it does that specific task. Summarize this article. Format this list. Translate this paragraph. You drive, it executes.
Augmentation. You and the AI work together on something neither of you would produce alone. A draft you bat back and forth. An outline you both contribute to. This is where most thoughtful AI use lives, and it's the mode this course will spend the most time on.
Agency. You configure an AI to act on your behalf, often with other people or systems. A tutor bot, an assistant that handles email, an interactive character. Less common in K-12 classrooms today; worth knowing about anyway.
The framework isn't a checklist or a curriculum. It's a vocabulary. Once teachers and students have shared words for "I delegated badly" or "your discernment was light here," the conversation about AI in school gets dramatically easier. Half of what we're doing in this course is just installing the words.
Welcome, faculty.
A few honest things about why this matters and what this course can and can't do.
You're being asked to teach something the people who built it don't fully understand. That's the unflattering fact of AI in education, and pretending otherwise won't help anyone. The good news is that students don't need their teachers to be AI experts; they need their teachers to have a framework, a few good questions, and the willingness to think out loud. That's what this course is for.
Some of you have been using AI since GPT-3 dropped and you have strong opinions about which model to use for what. Some of you have a deep, principled suspicion of the whole enterprise. Some of you have never typed a prompt in your life. All three of you are welcome, and the framework is designed to meet you where you are.
What success looks like
By the end of this track, you should be able to:
- Name the four competencies and give a classroom example of each.
- Recognize when a student's AI use is a delegation problem, a description problem, a discernment problem, or a diligence problem, because they're different problems and they need different conversations.
- Demonstrate at least one interactive exercise from this course in your own classroom.
- Articulate, in your own words, the difference between using AI as a thought partner and using it as a homework outsourcer.
How to use this course
Each module takes about fifteen minutes to read and another ten to do the exercises. You can do them in order or skip to whichever D you want; the modules cross-reference each other but stand alone. The student track parallels the teacher track and uses the same exercises in many places; you can show your students the student modules directly, or you can pull the exercises out and use them in your own way.
One last thing. The framework comes from Dakan, Feller, and Anthropic; the examples and the snark are mine. The course is licensed CC BY-NC-SA, like the underlying framework, so you can adapt anything in here for your classroom without asking. Just keep the attribution honest. That's the diligence move.
Delegation.
The decision before the decision. Whether to use AI at all is itself a skill, and most failures of AI in school start here.
Delegation is the most boring word in the framework and the most important one. It covers everything that happens before you type a single character into a prompt: knowing what you actually want, knowing what AI can and can't do, and deciding how to divide the work between you and the machine. Skip this step and the other three Ds collapse.
The framework breaks Delegation into three pieces (Goal Awareness, Platform Awareness, and Task Delegation), and each one rewards a teacher's attention, because each one shows up in student work as a different kind of failure.
Goal Awareness
Knowing what you're trying to do; understanding the task before you outsource any of it. A student who asks AI to "write an essay about The Great Gatsby" hasn't done goal awareness. They don't yet know what their argument is, what evidence they want, what audience they're writing for. They've delegated their thinking to the tool, and the tool will hand them an average of every Gatsby essay it ever ingested. That's not collaboration; that's surrender.
Platform Awareness
Knowing what the tool can actually do. Different models have different strengths; the same model behaves differently in different interfaces. Flint is not Claude. Claude is not ChatGPT. A free model from two years ago is not a current model. Platform awareness is the part of delegation where you stop and ask whether you're using the right hammer.
Task Delegation
Splitting the work between human and machine deliberately. Some tasks the AI should do alone. Some you should do alone. Most live in the middle, where AI drafts and you revise, or you draft and AI critiques, or you brainstorm together. The skilled delegator decides this on purpose instead of by default.
Sort these classroom tasks.
Drag each task into the bucket where it most belongs. There aren't always clean answers; the point is to make the call deliberately. Reveal the commentary when you're done.
The most common AI failure in student writing isn't a hallucination; it's a delegation failure. The student didn't know what they wanted to say, so they asked AI to want it for them. The essay reads as flat because the thinking was outsourced before the writing started. The fix is upstream: better assignments, better front-end planning, more time spent on goal awareness before anyone touches a tool.
For your classroom
Before a writing assignment, run a five-minute exercise where students write down (a) what their argument is in one sentence, (b) what kind of help they want from AI, if any, and (c) what kind of help would cross the line for this assignment. That single page does more for AI literacy than a syllabus paragraph ever will.
Description.
Most people call this prompting. Description is bigger than prompting, and the difference matters.
Description is how you tell the AI what you want. It sounds like the easiest of the four Ds, and in some ways it is; you just type words and the AI does things. The complication is that what you type determines almost everything that comes back. The framework breaks Description into three layers, and the trick of teaching it is getting students to think in all three at once.
Product Description
What you want the output to be. A five-paragraph essay. A bulleted list. A first-person diary entry from a Civil War soldier. A counterargument to a thesis. Most students stop here and wonder why their output is mediocre.
Process Description
How you want the AI to get there. "Before you answer, list the three strongest counterarguments." "Walk me through your reasoning step by step." "Ask me three clarifying questions before you start." Process description turns the AI from a slot machine into something closer to a colleague.
Performance Description
How you want the AI to behave during the conversation. "Be skeptical of my claims." "Don't hedge; pick a position." "Write at a tenth-grade level." "Push back on me." This is the layer most students never use, and it's the one that makes the biggest difference.
The same goal, two prompts.
Both of these are real prompts written by real students working on an essay about Beloved. Read both, then notice how the second one uses all three layers.
The first prompt produces a Wikipedia-flavored essay the student will then have to disown. The second produces a working partnership.
Watch a prompt accumulate.
A weak prompt and a strong prompt aren't different in kind. The strong one just has more layers turned on. Toggle each layer to see the prompt grow.
Have students rewrite the worst prompt they've ever used, adding one layer at a time. Watch their faces when the output changes. This is the single fastest way to break the "AI = slot machine" mental model.
Discernment.
The output looks confident. That's the problem. Discernment is the habit of asking whether confidence and correctness are the same thing.
Large language models are professional bullsh*tters in the technical sense Harry Frankfurt meant the word, speech that's indifferent to the truth. The model isn't lying; it doesn't know what lying is. It's producing fluent text that sounds like what a correct answer would sound like. Sometimes the fluent text is also correct. Sometimes it's polished nonsense delivered in the tone of an expert. Discernment is how you tell the difference.
The framework splits Discernment into three pieces, mirroring Description.
Product Discernment
Evaluating the output itself. Is it accurate? Is the source real? Does the date make sense? Did it cite a book that exists? This is the surface layer of discernment and the easiest to teach; it's basically fact-checking with extra steps.
Process Discernment
Evaluating how the AI got there. Did its reasoning hold up, or did it skip a step? Did it make an assumption you didn't authorize? Did it treat one piece of evidence as decisive when it was actually contested? This is the layer that separates novice users from skilled ones.
Performance Discernment
Evaluating how the AI is behaving in the conversation. Is it sycophantic? Is it agreeing with you when it shouldn't? Is it hedging on something you need a real answer about? Students rarely notice this layer; teach them to.
Spot the hallucinations.
This paragraph was generated by an AI in response to the prompt "Summarize Frederick Douglass's contributions to American literature in one paragraph." Click anything you think is wrong. Take your time.
Notice what the AI did not do. It didn't hedge. It didn't say "I'm not sure about this." It delivered three confident factual errors in the same register it used to deliver eight confident factual truths. There is no signal in the prose itself that tells you which is which. Discernment isn't a feeling; it's verification.
Teaching discernment without crushing trust
The risk in teaching discernment is producing students who don't trust anything: the AI, the textbook, you. That's not the goal. The goal is calibrated trust: high for things the AI is reliably good at, low for things it isn't, and verified for anything load-bearing. AI is solid for restructuring, drafting, brainstorming, summarizing what you've already read. It's unreliable for citing sources, dates, anything obscure, anything beyond its training cutoff, and anything emotionally charged where it might just tell you what you want to hear.
Diligence.
Owning what you produce with AI. The least glamorous of the four Ds and the one that holds the rest together. It's the step where the Honor Code and character formation really come into play.
Diligence is the competency that connects AI use to professional and academic integrity. It covers responsibility for the work, transparency about how it was made, and verification before it goes out the door. The framework splits it three ways.
Creation Diligence
Being thoughtful about which tools you use and how. Choosing a model that's appropriate for the task; being aware of its biases; declining to use AI when the assignment is explicitly about your own thinking. This is the upstream piece, ethics during the work, not after.
Transparency Diligence
Being honest about AI's role in your work, to the audience that needs to know. In school that's your teacher. In a job that's your boss or your reader. The standard isn't "always disclose everything," it's "disclose what the context expects." Students often need help calibrating this.
Deployment Diligence
Verifying and vouching for what you submit. If the AI wrote a paragraph and you turned it in, you are responsible for everything in it, every claim, every citation, every implication. The signature on the assignment is yours. The AI cannot share credit for the work; it also cannot share blame for the errors.
Which kind of diligence failure?
Each of these is a real scenario from a high school humanities classroom. Pick which type of diligence got skipped.
If you want a one-line disclosure rule for student work, try this: If AI shaped the thinking or shaped the prose, name it. Tools like spell-check and grammar suggestions don't count. If you'd be uncomfortable if I knew, that discomfort is the signal. Students respond well to honesty norms when they're stated as norms rather than as gotchas. One other note about what to tell students: avoid making grandiose threats about programs that market themselves as AI detectors. For the most part, they do not work. Worse, claiming to have the ability to detect AI and then not detecting AI on an assignment blows your credibility apart.
The Honor Code Conversation.
AI didn't break academic integrity. It exposed how thinly we'd defined it.
Most honor cases involving AI fail not because the policy was unclear but because the policy was unclear about the wrong things. "Don't use AI to cheat" doesn't help a student who genuinely isn't sure whether using AI to brainstorm counts as cheating. The framework gives us a better way in. Most honor-code questions about AI are diligence questions, and most diligence questions reduce to one of three things: whether the task was supposed to be your work, whether you disclosed what wasn't, whether you verified what you used.
The cases that should worry us most aren't the obvious ones. They're the medium ones, the gray-zone uses that students slide into not because they're cheating but because they don't know where the line is. Teaching the framework reduces those cases. Punishing students who never had the framework just generates resentment without changing behavior.
The spectrum.
For each scenario, decide what you'd do as the teacher. Then click to see how the framework reads it.
The policy question
Schools keep asking for the right AI policy. There isn't one. What there is, instead, is a set of conversations that need to happen at the assignment level, not the policy level. The right granularity is the syllabus and the assignment sheet, not the handbook. Tell students for this assignment, AI use looks like this, disclosure looks like this, and the line is here. Repeat for every major assignment. The framework gives them the vocabulary to follow you when you do.
They don't work reliably. They produce false positives at rates that should disqualify them from punitive use. A student's writing can be flagged as AI-generated for being too polished, too consistent in voice, or too organized, all of which can describe a careful human writer, especially a non-native English speaker. If you're going to use detectors at all, treat them as one signal among several, not as proof.
In Your Classroom.
The framework is platform-agnostic and discipline-agnostic. That's a strength when you're learning it and a problem when you have to teach it. Here are starting points by subject.
Pick the subject you teach. Each tab below has three things: a concrete example of an AI-augmented assignment that uses the 4Ds, a list of failure modes specific to that discipline, and a one-paragraph language sample you can lift into your syllabus. Adapt freely.
English
Sample assignment. A close-reading essay on a poem of the student's choice. Students must (a) draft a thesis on their own, (b) workshop the thesis with AI using the prompt-builder approach (product, process, performance) and submit both the original and revised thesis, (c) write the essay themselves, and (d) include a 100-word diligence statement explaining what AI did and didn't do.
Failure modes to watch for. Voice flattening (the AI smooths out the student's actual voice into a generic literary-essay voice); over-reliance on AI's first interpretation of a text instead of the student's own reading; fabricated literary criticism citations.
Syllabus language. "AI can be a useful thought partner for brainstorming and revising in this class. It is not a useful tool for doing your reading or your writing for you. Every assignment will specify what AI use looks like for that assignment; when in doubt, ask. Submitted work must include a brief diligence statement when AI shaped your thinking or your prose."
History
Sample assignment. A document-analysis paper on a primary source. Students may use AI to summarize secondary sources but must verify every factual claim against the original sources, which they cite. The diligence statement names which sources AI summarized and which the student read directly.
Failure modes to watch for. Fabricated dates, names, and document titles (AI is unreliable for specifics that aren't widely circulated); AI's tendency to flatten historical complexity into a moral narrative; uncritical use of AI summaries of sources the student hasn't read.
Syllabus language. "In history, the work is the verification. AI may help you understand context and develop questions. It may not substitute for engagement with primary sources. Every factual claim in your writing must be traceable to a source you have actually read."
Math
Sample assignment. A problem set where students attempt each problem on their own first, then may use AI as a tutor, not as a solver, and turn in both their original attempt and their final solution. The interesting work is in the gap between the two.
Failure modes to watch for. AI confidently producing wrong answers, especially for novel problems or proofs; students treating AI's solution as authoritative when it isn't; loss of productive struggle. The friction of working through a problem is what builds the skill, and AI can short-circuit it.
Syllabus language. "AI can be a useful study partner in this class. Treat it the way you'd treat a friend who's pretty good at math but not reliably right. Verify everything. For exams and quizzes, AI is not permitted; you should be able to do this work yourself."
Science
Sample assignment. A lab report where AI may help with the discussion section (specifically, with situating results in the broader literature) but not with the methods, results, or data analysis. Students submit a diligence statement showing which parts of the discussion they wrote and which AI shaped.
Failure modes to watch for. AI invents studies and citations; AI smooths over uncertainty in ways that misrepresent what the data actually show; students using AI to "fix" anomalous results instead of reporting them.
Syllabus language. "In science writing, AI can help you write about your results. It cannot help you find them. Methods, results, and data analysis are entirely your work. The discussion section may involve AI, with disclosure."
World Languages
Sample assignment. A short essay in the target language. Students draft entirely without AI, then may use AI to identify grammatical errors and suggest revisions. They must explain, in English or the target language, why they accepted or rejected each suggestion. The reasoning is the assessment.
Failure modes to watch for. Translation laundering (writing in English, translating, claiming it as original work); AI flattening idiomatic choices into textbook constructions; students unable to defend any of the language choices in their own paper.
Syllabus language. "In this class, AI is a tutor, not a translator. You may use it to check your work after you've drafted it. You may not use it to produce work you couldn't produce yourself. Every assignment will be paired with a short conversation in the target language so you can demonstrate what you actually understand."
Arts
Sample assignment. A creative project of the student's choice. AI may be used for ideation, reference-gathering, or technical questions. AI-generated content may not be submitted as the student's creative output. The diligence statement names where AI showed up.
Failure modes to watch for. Confusion about what counts as "the work." In a visual arts class, the work is the student's hand; in a creative writing class, the work is the student's prose; in performance, the work is the performance. AI use that displaces the student's own creative labor is the line.
Syllabus language. "In arts classes, the medium is the message and the labor is the lesson. AI may inform your creative process but may not produce work submitted as yours. When in doubt, ask. Disclosure of AI use is expected and never penalized; failure to disclose is."
Where to start tomorrow
You don't need to redesign your curriculum to teach AI fluency. Pick one assignment in the next unit. Add a five-minute delegation exercise at the front and a brief diligence statement at the end. Run it. See what happens. The framework gets internalized through use, not through study.
Show students the 4Ds overview, run the prompt-builder exercise from Module 2, run the hallucination-spotter from Module 3, and end with five minutes on what a diligence statement looks like for their next assignment. That's a fluency lesson in fifty minutes, and it leaves a residue that lasts.
AI is not an oracle.
It is a very fluent guesser that has read more than you have. That's a useful thing to have around. It is also not what you've been told it is.
You've probably heard AI described in one of two ways. Either it's the future and you'd better get on board, or it's the apocalypse and you should stay away. Both of these are kind of dumb. AI is a tool. Like every tool, it has a shape: things it's good at, things it's terrible at, things it does so confidently you might miss the fact that it's wrong. This course is about learning that shape.
The framework you're going to learn was built by professors who teach with AI every day. It has four parts, and each part is something you already do when you work with anyone, AI or human. You decide what kind of help you want. You explain what you need. You check whether the help is actually good. You take responsibility for what you turn in. Those four things have names: Delegation, Description, Discernment, Diligence. We'll spend a module on each.
What you'll be able to do by the end
You'll be able to use AI as a thought partner instead of a homework outsourcer; spot the moments when it's making stuff up; describe what you want clearly enough to get useful answers back; and explain, honestly, what AI did and didn't do in any piece of work you turn in.
You'll also, and this is the bigger one, have a way to talk about AI with your teachers, your parents, and yourself. Right now most of you don't. That's not a flaw in you; it's a flaw in how this technology was introduced. We're trying to close that gap.
AI is wrong sometimes. Not occasionally. Routinely. It says things that sound right because they were assembled from things that were right, and that is not the same as being right. The whole second half of this course is about how to tell the difference. If you take only one thing from this whole experience, take this: never turn in a fact you got from AI without checking it somewhere else first.
Delegation.
Knowing what to ask AI to do, and what not to. The decision before the decision.
Delegation is a fancy word for "deciding who does what." You delegate all the time. When you ask a friend to grab you a snack from the kitchen, you've delegated. When you let your group chat decide where to eat, you've delegated. With AI, delegation means deciding what part of your work you want to do yourself and what part you want help with.
Most people skip this step. They open ChatGPT, type "write me an essay about X," and accept whatever comes back. That is delegation by default. It feels like the AI did your work for you, and in a sense it did, but the work it did wasn't actually what you needed. You needed to think about something. The AI thought for you, and now you don't know what you think.
Three questions before you start
Skilled delegators ask three questions before they ask AI anything.
What am I trying to do? Not "write an essay." What's the actual point of the essay? What are you arguing? What do you want a reader to walk away with? If you can't answer this in one sentence, AI cannot help you. Talk to a friend first, or take a walk.
What can the AI actually do well here? AI is good at: brainstorming, outlining, summarizing things you've read, suggesting alternatives, finding the weak spot in your argument, rewriting at different reading levels. AI is bad at: facts you can't easily check, specific quotes from books, citing sources, anything where being slightly wrong matters a lot.
What part of this should I do alone? Some work has to be yours, full stop. Personal reflections. Creative writing where the point is your voice. Math problems where the practice is what's building the skill. Assignments where your teacher said no AI. Be honest about which part of your work falls in this bucket.
Sort these tasks.
Drag each one into the bucket where you think it best belongs. Then read the explanations.
Before you open ChatGPT or Claude or whatever, take thirty seconds and write down on paper (real paper) what you're trying to accomplish. If you can't write it down, you don't know it well enough to delegate any of it. Your prompt won't fix that. Only thinking will.
Description.
How you talk to AI determines almost everything that comes back. This is the skill most students underrate.
You know how when you text a friend "wanna get food" you might mean Chipotle, you might mean a coffee, you might mean we should plan something for Saturday? Your friend knows you, so they can probably figure it out. AI doesn't know you. It only knows what you typed. So the more you put into the prompt, the more it has to work with.
The framework breaks Description into three layers. Most students use only the first one and wonder why their AI is meh.
Layer 1: What you want the output to be
The shape of the thing. A paragraph. A list. A poem. A three-sentence email. A counterargument. Stop here and you'll get an average version of whatever you asked for.
Layer 2: How you want the AI to get there
The process. "Before you answer, ask me three questions." "Walk me through your reasoning." "Give me three options and tell me which one is best." When you describe the process, you turn the AI from a vending machine into something closer to a tutor.
Layer 3: How you want the AI to behave
The personality. "Be skeptical of my claims." "Don't just agree with me." "Talk to me like I'm in high school, not like I'm a professor." This is the layer almost no one uses, and it's the one that changes the most.
The same question, two different prompts.
The first prompt will give you a generic Gatsby thesis. The second will give you something you can actually use (and might disagree with), which is even more useful.
"Ask me three questions before you start" is the single most useful phrase you can add to a prompt. It forces the AI to figure out what it doesn't know about your situation before it commits to an answer. Try it on your next prompt. Watch what happens.
Discernment.
AI sounds confident even when it's wrong. Discernment is the habit of checking, every time, no exceptions.
Here is the most important thing in this entire course. AI does not know when it's wrong. It cannot tell the difference between a fact and a fact-shaped sentence. When it confidently tells you that Frederick Douglass won a Pulitzer Prize (he didn't; the Pulitzer didn't exist yet), it isn't lying. It's producing fluent text that sounds like what a correct answer would sound like, and it has no internal way of checking whether the text is actually correct.
This is called hallucination, which is a polite word for "making stuff up." It happens all the time. It happens most when you ask AI about specific facts, dates, citations, quotations, or anything obscure. The output looks identical whether it's right or wrong. Same confidence. Same polished prose. Same vibe of authority. That's what makes it dangerous.
Find the three errors.
An AI generated this paragraph in response to the prompt "Summarize Frederick Douglass's contributions to American literature in one paragraph." Three of the things in it are wrong. Click anything that looks suspicious.
How to discern
You don't need to verify everything AI tells you. You do need to verify everything AI tells you that you're going to turn in, repeat, cite, or rely on. Three quick moves:
Check the load-bearing facts. Any date, name, quotation, or citation needs a second source. Google the thing. If you can't find independent confirmation in fifteen seconds, treat it as suspect.
Ask the AI to check itself. "Are any of the facts in your previous answer ones you're uncertain about?" AI can often flag its own weak spots if you ask, not because it has hidden knowledge of its mistakes, but because asking lowers its confidence and surfaces the fuzzy stuff.
Notice the smoothness. AI rarely writes "I'm not sure" without prompting. If everything sounds equally confident, that's a sign that nothing has been weighted properly. Real experts hedge on the right things.
If you turn in a fact you got from AI without checking it, and that fact is wrong, that is your error, not AI's. The AI did not sign the paper. You did. This is the part of the course that gets people in the most trouble, and it gets them in trouble because they trusted the polish. Don't.
Diligence.
Owning what you turn in. The line between using AI as a tool and outsourcing your education to one.
Diligence is about responsibility. Specifically: when you turn something in with your name on it, you are vouching for it. Every word. Every fact. Every idea. That doesn't change because AI helped you write some of it. The AI cannot take responsibility for your work. You can. Or you can pretend you can't, which is what most honor-code cases really boil down to.
This module is short because the rules are short.
Three rules
Be honest about what AI did. If AI helped you brainstorm, say so. If AI helped you outline, say so. If AI rewrote a paragraph, say so. The goal isn't to make yourself look bad; thoughtful AI use isn't a flaw. The goal is to be straight with the person reading your work. Teachers can tell when AI is shaping a piece; pretending it didn't is what crosses the line.
Don't outsource the parts the assignment is actually about. If the assignment is a personal reflection, the reflection has to be yours. If it's a creative piece, the creativity has to be yours. If it's a problem set, the working-through has to be yours. The trick to figuring this out: ask what skill the assignment is supposed to be building. That skill has to come from you.
Verify before you submit. Every fact. Every quote. Every citation. If you can't find an independent source that confirms it, take it out or rewrite it. The five minutes you spend checking is the cheapest insurance you'll ever buy.
Honor code or no honor code?
Read each scenario. Decide what you'd do as the student. Then see how the framework reads it.
If you'd be uncomfortable if your teacher knew exactly what AI did, that discomfort is the signal. Pay attention to it. The line you're looking for is usually right there.
AI as a thought partner.
The whole framework, in one move. What it looks like to actually use AI well.
You've now seen all four pieces. Delegation: knowing what to ask. Description: knowing how to ask. Discernment: checking the answer. Diligence: owning the result. The framework works because the pieces work together. Skilled AI users are running all four loops at once, often without thinking about it.
Here is what that looks like in practice.
A short story about a paragraph
You're writing an English essay on The Bluest Eye. You've read the book. You have a feeling about it, something about how Morrison shows beauty as a weapon, but you can't get the thesis tight. You open Claude.
You don't say "write me a thesis about The Bluest Eye." That would be delegation by default. Instead you say: "I'm writing an essay on how Morrison uses beauty as a tool of harm in The Bluest Eye. I keep getting stuck on the thesis. Ask me three questions about what I actually want to argue. Don't suggest a thesis yet." That's description with all three layers turned on.
The AI asks: Is the harm internal or external? Is it doing it deliberately or just observing it? Does the novel offer any alternative to beauty as a value? You answer. Now you can see your own argument more clearly than you could ten minutes ago. The AI gave you nothing. It just held a mirror up so you could see what you already thought.
You write the essay yourself. When you're done, you ask AI to look for spots where your argument is weakest. It flags two paragraphs. You don't take its rewrites; you write your own. That's discernment.
You include a one-line note at the bottom: "I used Claude to help me clarify my thesis and to flag weak paragraphs. All writing is mine." That's diligence.
The essay you turn in is better than what you would have written without AI, and it's still yours. That's the whole point of the framework. AI is a thought partner that makes your thinking sharper, not a writing partner that makes your essay theirs.
What this gets you
The kids who learn to do this become much better thinkers, much faster, than the kids who don't. They also stay out of trouble. They also, and this is the part that takes a while to notice, develop something like taste. They learn to feel when AI is being helpful and when it's being a crutch. They learn to notice the moment when their own voice starts to disappear into the polish, and they learn to pull it back.
That last part takes practice. You won't get it overnight, and you'll get it wrong sometimes. The goal isn't to use AI perfectly; the goal is to use it on purpose. Everything in this course is just helping you do that.
The next time you reach for AI, stop and ask: what am I trying to do, and what part of this is supposed to be my thinking? Then write the prompt that tells AI which is which. That's the whole framework. Everything else is detail.