What If the AI Was Right?

A colleague recently shared a curricular menu with me. It paired digital library databases with AI tools across a set of student tasks. The work was thoughtful. It was designed with care. It would be useful in a lot of classrooms tomorrow.

But one pattern kept surfacing as I read through the rows that I have noticed in so many other avenues of work lately and was having a hard time articulating the signal for a newsletter, but it finally hit me.

The AI was always the thing to check. The database was always the corrective. The verbs across the menu told a story the designer probably didn’t mean to tell of find the hallucination, verify against the source, locate the error, demonstrate why the database is the one you should trust. That’s not curriculum about AI literacy. It’s curriculum that teaches a stance and calls it a skill.

If every AI task in your curriculum points toward catching a mistake, what are you actually teaching students AI is for?

This issue is built around that question. The hidden curriculum nobody named out loud. Two resources doing the open-stance work well. What the macro picture looks like right now from Benedict Evans, Google, Microsoft, and McKinsey. And what a year of real district AI policy work looks like when the stance is named explicitly. Grab your coffee.

1) The stance was in every row

The pattern I see in so many educator posts, lessons, suggestions, etc. is not that what people are processing are wrong, but the perspective needs to shift. It does not have to be so AI negative. The pattern isn’t about any individual designer’s choices, it’s about what’s happening across the field right now, in good faith, in lessons being written.

This is a typical structure repeated over and over in classrooms everywhere if they are even navigating AI at all. Step one: have students get a quick AI summary on a topic. Step two: have students consult the vetted research database. Step three: have students identify what the AI got wrong. Step four: have students explain why the database is the reliable source. Cite any AI use whatsoever and not work on ourselves as adults to train brain to not have unintentional bias towards the work.

Each step, read on its own, sounds reasonable. The cumulative move is something else. The lesson assumes a verdict before the student begins. The activity is closed. The “skill” the student is practicing is performing skepticism toward a tool that has already been decided to be the inferior source.

You can build careful curriculum around a stance you’ve never said out loud.

Personal Tie-in: I’ve been on the team writing Iowa’s K-12 AI standards this year, and every meeting pushes me back to this idea of are we asking students to discover something about AI, or to confirm a verdict we’ve already reached? It is much easier to spot in someone else’s curriculum than in your own.

So What? The frame is taught before the content is.

Try This: Open one AI-integrated curriculum document you’ve authored in the last six months. Highlight every verb tied to AI. Read just the verbs. What story do they tell?

2) Fact-check is a stance

The verbs are the lesson.

“Find the hallucination.” “Catch the error.” “Verify against the vetted source.” Each of those is a pedagogical move that commits the lesson to an outcome before the student starts. The student knows what they’re supposed to find. The student knows which source they’re supposed to trust. The student knows the verdict the assignment is asking them to deliver.

That’s not evaluation. That’s a multiple choice test with one answer pre-selected.

Compare those verbs to a different set: compare, triangulate, evaluate, defend, decide. These verbs open the task. They invite the student to read both artifacts as evidence, not as defendant and prosecutor. They put the student in the position of the source rather than the position of the auditor.

The verbs tell you the answer the lesson has already decided on.

Personal Tie-in: I’m in the middle of a chapter draft of something I am writing that lives at exactly this seam. A lesson built around fidelity doesn’t need open verbs. A lesson built around integrity can’t survive without them.

So What? Closed tasks teach closed thinking about AI, about sources, about everything.

Try This: Take one AI-integrated lesson plan in your district. Sort the action verbs into two columns: verbs that invite evaluation, and verbs that confirm a predetermined verdict. Count the ratio.

3) Same resources, different question

The fix isn’t to throw out the tools. The fix isn’t even to throw out the comparison. The fix is in the question.

Here’s an original task, lightly anonymized but structurally honest:

Fact-Checking the AI — Use an AI tool to get a quick summary. Then, find the same topic in a database. Locate one hallucination the AI invented and explain why the database is the vetted source you should use when accuracy is your top priority.

Here’s a version that uses the same resources, the same workflow, and asks a different question:

Source Triangulation — Generate a summary on your topic with an AI tool and pull a related entry from a research database. Compare the two using a three-column chart: What appears in both? What is unique to each? Where do they disagree? Decide which source or which combination best serves your purpose, and defend that choice with evidence.

Same materials. Same time on task. Same workflow. But the task no longer assumes the answer. Sometimes the AI synthesis wins on clarity. Sometimes the database wins on depth. Sometimes the strongest move is to pull from both and cite the seams. That’s the lesson. The student does the evaluating. The teacher does the coaching.

Sometimes the AI synthesis is the stronger artifact and the task has to be open enough to discover that.

Personal Tie-in: I’ve been writing publicly about my garden and AI for a while now and every season turns into a new round of triangulation. The AI gives me companion planting guidance, sun exposure plans, succession schedules. I sit with seed packets, last year’s notes, and the local extension info. Sometimes the model is right. Sometimes the seed packet is right. Sometimes the AI surfaces a question I wouldn’t have thought to ask. The honest lesson hasn’t been “the AI is unreliable.” It’s been: fact-check everything, hold the verdict open, let the strongest answer win whichever source it came from.

So What? Open the question, get real evaluation. Close the question, get students performing the answer.

Try This: Take one comparison task in a current lesson plan. Rewrite the prompt so the outcome isn’t predetermined. Read both versions back to back.

4) Two resources already naming the move

I want to point at two resources doing this work in public, at very different scales.

ITEC just launched their latest newsletter which contains Instructional Coach Corner, a monthly Problem of Practice series for instructional coaches absorbing what used to be the tech coach role. The first installment is on AI-driven personalization and differentiation, and the design is worth looking at closely. The Problem is named honestly: teachers struggle to differentiate a single lesson for IEPs, ELLs, and varying reading levels without building three versions of the same resource. The Pivot is Low Floor, High Ceiling. Don’t change the objective. Change the entry point. Use Brisk Teaching or MagicSchool to level a single article into three reading levels. Pair it with an analog choice board. The bonus pointer is to Eric Curts’ edugems.ai which specifically the Learner Scaffolds Gem and the Re-level Text Gem.

Notice the verbs. Adjust, level, scaffold, offer, demonstrate. Every one of them is open. None of them say catch or verify. The tool isn’t on trial. The tool is in service of the work. And the Prompts ITEC closes with are genuine coaching questions: Which students are most likely to check out because the material feels out of reach? If you could instantly create three versions of a complex text, how would that change small-group instruction? Those are coaching questions, not gotcha questions.

At the higher ed level, Elon University and AAC&U just published the third volume in their AI-U Guide series: Human Wisdom for the Age of AI: A Field Guide to Cultivating Essential Skills, in partnership with The Princeton Review. The whole 16-page guide is built around 10 human capacities anchored in classical thinkers like Descartes for curiosity, Seneca for deep thinking, Phillis Wheatley for creativity, Mary Parker Follett for social intelligence, Aristotle for storytelling, Cicero for decision-making. Each capacity comes with a mini-tool. The Idea Compass for curiosity. The Persuasion Triangle for storytelling. Panning for Gold for decision-making. The Signal Bars for social intelligence that start at one bar with just you and the AI, move to two bars with one human peer, move to three bars with a small group, and notice how the answer changes. The companion self-assessment tool is worth ten minutes before you assign anything from the guide to anyone else.

Worth naming: not every page of the Elon guide stays out of the fact-check stance. The Human Validator section frames AI as “a confident but unreliable witness who is known to fabricate details to please the interrogator.” That’s exactly the language Section 2 of this issue is questioning. The guide is a substantive resource AND a place where the stance question is visibly being worked out in real time. Both things are true at the same time.

Two resources at two different levels of the system are doing the open-stance work and even the strongest of them shows the stance question hasn’t been settled inside any one frame.

Personal Tie-in: I work in this exact space facilitating AI conversations and tools and sessions across our region try to land the same pivot ITEC is naming. The hardest part isn’t finding the tools. It’s writing the prompt that doesn’t pre-decide the answer.

So What? Good AI integration design names the problem first, the pivot second, and the tool third and admits where the stance is still being figured out.

Try This: Take the Elon Field Guide’s self-assessment yourself before you assign it to anyone. Then read the Human Validator page next to the Idea Compass page. Notice which one opens the question and which one closes it.

5) Noise, hype, anti-hype

The stance question is just as visible at the macro level. Pull up Benedict Evans’ AI Eats the World deck that he updates it twice a year, and the Autumn 2025 edition came out of his Slush keynote in Helsinki. One of the slides is literally titled Noise, hype, anti-hype. That’s the whole 2025–2026 conversation in three words.

Evans builds a deck with by his own framing more questions than answers. The chart that stuck with me longest compares what enterprises expected their cloud adoption to look like after a decade and what it actually looks like (around 30% of workloads moved over). The future arrived, slowly and unevenly. The honest framing isn’t “AI will transform everything by 2027” or “AI is mostly hype.” It’s that the future will arrive on a longer curve than the loudest voices on either side are claiming, and the work right now is to build with that timeline in mind.

The same week, Google published an update on AI Overviews and the substance of the update matters. They’re rolling out five changes that all push AI Search toward original sources rather than away from them: a “further exploration” panel at the end of AI responses pointing to in-depth articles, news subscription highlights so your trusted sources surface, firsthand advice quoted from public forums with the community named, more inline links right next to the bullet point they support, and hover previews so you know what site you’re about to click into. A lot of the K-12 conversation about AI Overviews is still anchored to the early-2024 hallucination examples. The product has been changing every quarter since.

The macro picture keeps revising itself. A curriculum stance that pre-decided in 2024 is teaching from a snapshot that doesn’t match what’s on the screen this morning.

Browse: AI Eats the World — Autumn 2025 (Benedict Evans) / 5 new ways to explore the web with generative AI in Search (Google)

Personal Tie-in: I’ve been writing a long-form Chaos Navigators essay on the hype/panic swing for months, and Evans named the move in three words I couldn’t land in three paragraphs. Noise, hype, anti-hype. That’s going on a slide. And the Google update is a useful reminder for me too as I’ve stayed attached to my own first impressions of AI Overviews longer than the product warrants.

So What? Anything you decided about AI in 2024 deserves a recheck in 2026. The tools keep moving. The stance has to be able to move with them.

Try This: Pull one belief you hold about AI in education that you formed in the last 18 months. Search for the most recent evidence for and against it. Decide if you’d still hold it the same way today. I have been doing this with my workshops from 3 years ago and connecting dots to what still resonates and what doesn’t and also have little we have progressed in systems thinking and design.

6) The new agency equation

This week Microsoft published its 2026 Work Trend Index, surveying 20,000 workers across 10 countries plus a privacy-preserving analysis of 100,000 Microsoft 365 Copilot conversations. The headline is the one your students are going to live inside: as AI and agents take on execution, our own agency expands. Forty-nine percent of all Copilot conversations support cognitive work of analysis, problem-solving, evaluation, creative thinking. Eighty-six percent of AI users in the survey said they treat AI output as a starting point, not a final answer, and that they “stay responsible for the thinking.” The two skills users said matter more as AI takes on more work: quality control of AI output and critical thinking.

The finding I am chewing on: organizational factors of culture, manager support, talent practices account for more than 2x the reported AI impact of individual mindset and behavior. The leader’s stance shapes the worker’s outcomes more than the worker’s own orientation does. Microsoft calls this the Transformation Paradox: 65% of workers fear falling behind if they don’t use AI to adapt quickly, but 45% say it feels safer to focus on current goals than to redesign work with AI. Only 13% say their org rewards reinvention even if results aren’t yet there.

Read that paragraph again and replace “worker” with “teacher” and “organization” with “district.” The whole thing maps.

McKinsey’s Shopping in the Age of AI report adds a parallel from another industry. As agentic AI takes over routine purchase decisions, the role of the physical store has to be named on purpose as a convenience hub, discovery destination, fulfillment node. The retailers losing are the ones managing every store the same and “fully optimized for none.” McKinsey’s first imperative is define the role of each store with precision, including what the store will not do.

Define the role of each lesson with precision, including what the lesson will not do.

Personal Tie-in: The Microsoft finding that organizational factors carry twice the weight of individual mindset is a useful number for the conversations I’m having with curriculum directors. The teacher writing the lesson plan matters. The system the teacher is writing inside of matters more.

So What? The world your students are entering is naming the stance question explicitly. Curriculum that doesn’t is mis-preparing them.

Try This: Bring three questions to your next district leadership meeting. Who reviews AI-assisted work in our system? Who has the authority to redesign workflows around AI? How does a local win get captured and scaled? (Microsoft frames these as the three Frontier Firm questions. They’re also the three district AI governance questions.)

7) What naming the stance looks like in a real district

I’ve been working with a district for the last year on exactly this and “for the last year” is the part that matters most.

I can’t name the district. But I can name the shape of the work, because the shape is the lesson. A cross-role AI team of building principals, classroom teachers, instructional coaches, tech directors, curriculum leads. Topics covered explicitly, in this order: Policy. Definitions. Acceptable Use for Students. Acceptable Use for Staff. Vetting. Data Privacy. Year 1 Implementation. None of those is a tool conversation. All of them are stance conversations.

The staff platform decision is a case study by itself. The team evaluated Gemini, ChatGPT, Claude, and Copilot against the same criteria of enterprise data protection, data sovereignty (does the vendor train on our data, yes or no), FERPA and COPPA compliance, integration with the tools staff already use every day. They landed on Gemini for staff, with explicit reasoning. They didn’t land on a tool and back-fill the reasoning. They built the reasoning and chose the tool downstream of it.

The AI Tool Vetting Workflow is the part I keep coming back to. Three stages: Building Level review by the principal, Tech Review against safety guidelines (FERPA, COPPA, data training), and Final Approval by the District AI Team. Approved tools land on a central document visible to every staff member. The verbs across that workflow are submit, review, audit, approve, deploy and notify the requesting staff member if guidelines aren’t met. No tool gets in the building until somebody named has taken responsibility for it being there.

Two new board policies live underneath all of it one on internet appropriate use, one specifically on AI in the educational environment. Both reach the board. Both get communicated to families. Both get revisited.

The detail I think about most: on the final slide of the board deck, the team disclosed that they had used Gemini’s “Enhance this slide” feature to format their content. They cited their own AI use on the deck about AI use. That’s stance-naming behavior. Not “AI is the suspect.” Not “AI is the hero.” AI is a tool that was used; here is where; here is why we’re telling you.

The stance gets named or it gets inherited. There is no third option.

Personal Tie-in: I’ve spent a year inside this work, and the thing that has surprised me most is how often the conversation that looked like a tool conversation turned out to be a stance conversation in disguise. Picking the platform was actually about data sovereignty. Building the vetting workflow was actually about whose name goes on what. Drafting the AUPs was actually about what we believe a student is allowed to do, and a staff member is allowed to do, in a system that hasn’t fully decided yet. None of that work shows up in a curriculum menu. It shows up underneath it.

So What? The stance question is a year of work, not a paragraph in a policy.

Try This: Pull your district’s current AI policy (or your one-paragraph placeholder if that’s what you have). Ask: which of the seven topics above does it actually address? Which does it gesture at? Which does it leave to whoever happens to be in the room?

---

DIGITAL CHALLENGE

This week’s challenge is the verbs audit.

1. Open the most recent AI-integrated lesson plan, curricular document, or PD task you have authored or co-authored.

2. Highlight every verb attached to AI as an object like use, generate, find, verify, compare, fact-check, evaluate, decide, defend, identify, locate, summarize, and so on.

3. Sort them into two columns: verbs that open the task to genuine evaluation, and verbs that close the task around a predetermined verdict.

4. Count the ratio.

5. Pick the one closed verb doing the most pedagogical work, and rewrite that task with an open verb.

You don’t have to overhaul the document. You have to notice the stance.

ANALOG CHALLENGE

Pair with a colleague. Trade one AI task you are each proud of. Read each other’s task without comment for sixty seconds. Then each of you names out loud, to the other person the stance the other person’s task is teaching.

It is much easier to see in someone else’s work than in your own. That’s the point.

ONE SMALL HUMAN THING

Started the seeds this week.

Some are up. Some aren’t. The model didn’t know which would be which.

Neither did I.

CLOSING REFLECTION

The stance gets written in before the standards do. By the time a designer is sequencing their verbs, the frame is already chosen and usually unconsciously, usually inherited from whichever AI conversation that designer happened to be in last. Districts that have spent a year naming the stance explicitly are building governance the rest of us are still pretending we can skip.

If you assumed, for one task this month, that the AI might get it right what would change about what you ask the student to do?

— A-A-Ron