AI Exhaustion

What I've Learned During the 2025 - 2026 School Year

It’s mid-May and there is light at the end of the tunnel. Our school year ends in three weeks. Instead of anticipation, what I feel is something closer to exhaustion. The exhaustion is partially due to an especially brutal workload, but it also comes from spending nine months watching the gap between what AI actually is and how schools engage with it grow wider, not narrower. I feel like the conversation itself is stuck.

Below I’ll share a few observations as we wind down the academic calendar. None are especially surprising or groundbreaking. Taken together, though, they suggest significantly more work ahead for teachers, students, and institutions. My key takeaway is that the 2025 - 2026 school year will be remembered as the one where AI became ambient, unavoidable, and still mostly unaddressed.

Ambient AI

At the risk of stating the obvious, AI is everywhere. It’s still mostly associated with ChatGPT, but as of May 2026, unless you live under a rock, almost everyone is aware that several major companies have powerful AI models. More importantly, regardless of whether you know the names, understand the difference between Claude Code and Codex, or have experimented with any agentic capabilities - everyone on the internet has encountered AI images, audio, and sophisticated AI-generated content.

This was not true a year ago. The deep reasoning models had only emerged at the end of 2024 and I recall shocking teachers with demonstrations of what they could do last spring. The speed with which add-ons, plug-ins, and upgrades have rolled out means we are well past the ability to shock.

The implication for educators is simple. Unless we are prepared to keep our students from accessing Google, they will always be able to use a powerful frontier AI model in the form of Gemini through AI mode. I’ve argued since last June that teachers who claim they “forbid” AI are engaging in semantic theatre once a student is on the internet - we all encounter AI overviews whether we like it or not. Google’s AI mode allows for the critical cognitive off-loading educators fear, with the ability to have an LLM write, analyze, and work with uploaded documents baked into search. Literally the same capabilities students can access through ChatGPT. They do not need paid models.

Any notion that AI will not have a significant impact on teaching and learning is pure nostalgia.¹

Schools Are Finally Reacting, But The Gap Is Still Growing

The good news is that schools are finally taking AI seriously beyond writing policies. More administrators and district leaders are grappling with what’s actually happening rather than merely drafting documents that are increasingly opaque, difficult to enforce, and not designed to withstand the speed with which the technology is moving.

The bad news is there is no consensus on what the reaction should be. The recent trend is to zero in on devices themselves - a movement grounded in legitimate concerns but one that also conveniently feels like it “solves” the AI problem. If students don’t have their phones or access to laptops, there’s no technology to turn to. Essentially, it asks us to return to an idealized past where kids all wrote by hand and played outside. As a father of a 6th grader, I see the appeal. I really do.

But the majority of school districts already have technology so embedded in their infrastructure that an unraveling is not nearly as easy as it looks. Many are going in the opposite direction, having invested heavily in a future where AI is a significant part of their mission. With over 13,000 districts in the U.S., we will see almost every permutation along the technology and AI adoption spectrum. At least these experiments will give us more data and more opportunities to evaluate what works.²

The deeper problem is that even the conversations finally taking place are still mostly happening in the context of 2023 AI. Few schools that I know of are seriously engaging with where the models are now and where they are heading. Over the past year, AI has moved well beyond the text generation that prompted most school policies - it now builds functional applications, conducts multi-step research, and produces professional-quality documents and presentations, capabilities that most school conversations haven’t begun to address. And these are the paid frontier models - what the average teacher or administrator is actually encountering is likely already a generation behind.

I agree with those who argue that mastering current models shouldn’t be the goal of AI literacy - they will continue to evolve and what we use in May 2026 will not be the same as September 2026, let alone May 2030. But I depart from those who claim that learning AI is “easy” and that our responsibilities as educators do not include preparing students for the world they are actually entering.

The reason I disagree is the same reason I would have pushed back in the early 2000s when internet access arrived at scale. We had responsibilities then to demonstrate best practices, explain how to evaluate information, and model effective use. Some of those early digital literacy strategies didn’t hold up over time - and that will likely be the case for some of what we teach about AI today. But talking about AI in the abstract without showing students what these tools do, how they work, and what kinds of output they produce is not serving them well. Especially when we know almost all of them are using it outside the classroom.

The Big Questions Are Finally Landing

Perhaps the most meaningful development of the year is that AI has forced educators to confront questions about purpose. More people - even those most viscerally resistant to AI - are recognizing that these systems will likely be a significant part of our lives going forward. Skeptics are also wrestling with something harder to dismiss: that AI can organize, synthesize, and generate information in ways that people engaged in knowledge work find genuinely useful whether or not those opposed to it use AI themselves.

These capabilities are forcing hard, but necessary conversations. Not simply about cheating, which dominated the past three years, but about purpose. What are schools actually for when obtaining and presenting information - in virtually any format we require - is approaching zero cost? What does a research assignment mean when AI can search, synthesize, and organize information faster than the student writing it? These are existential questions, no longer the province of the futurists and technologists who have been raising them for years. They are finally landing in faculty meetings and department conversations - perhaps under duress, but landing nonetheless.

Maybe it was never about AI in the first place. It’s always been about the value of what we assign and what we ask students to do - questions we could avoid as long as the products students handed in were plausibly their own. It’s dawning on most of us, if we hadn’t realized it already, that the skills hardest to automate are the ones the humanities have always tried to teach: empathy, curiosity, the ability to ask good questions, and the patience to sit with ones that don't have clear answers. It's worth noting that many in the industry are arriving at the same conclusion.³

What It Cost Me This Year

I want to add one observation that is entirely anecdotal but may resonate with other teachers. I covered less content this year. The constant vigilance about AI, the reduced number of assignments I gave outside of class, the time spent on research projects completed under supervision rather than at home - it all added up. I simply could not cover as much material as I have in previous years.

I don’t know if this was specific to my circumstances or something more widespread.⁴ But it strikes me as one of the hidden costs of this transition that we are not yet accounting for. AI has added more monitoring, more in-class assessment, and more rethinking of assignments to a curriculum that was never conceived with it in mind. If schools don't change their approach, something has to give - and right now, for me, it's content. Maybe that’s not the worst thing when virtually unlimited access to information and the ability for a motivated student to learn on their own are higher than they’ve ever been. But Sal Khan himself recently acknowledged that Khanmigo was a 'non-event' for most students - they simply didn't use it.⁵ Students are constantly using AI, just not the way adults design it for them.

As I Head Into Summer …

2025 - 2026 is ending more or less where we started, but with even more uncertainty - and significantly more work ahead within our schools, with our students, our parents, and our colleagues, to catch up to the reality of what AI has done to the academic project. I can’t imagine 2026 - 2027 won’t bring more of the same.

I’m exhausted by it and I suspect many teachers are as well. But the questions won’t disappear just because we’re afraid of asking them. If anything, we’re just getting started.

---

Connect With Me

Beyond this newsletter, I work directly with schools, educators, and other institutions about pedagogical questions raised by AI. Take a look at my website and reach out - I’d love to hear what you’re working on.

1. If you want just one account of what AI has done to this year’s graduating college class, read this piece from yesterday’s Times - sure, it’s Stanford, but I suspect any student-journalist from a major college or university could write something similar about the devastating impact AI has had on higher ed.

2. It would be great if there was a more coordinated way for districts to share successes and failures, but the educational system in the United States is so diverse and fragmented that it makes this extremely difficult - not to mention the state of political polarization, which has turned AI into yet another partisan issue, making productive conversation even harder.

3. See, for example, Daniela Amodei's February 2026 interview with ABC News, reported by Fortune, The Future of Work.

4. If you taught this year and noticed something similar - or found a path forward - I’d love to hear about it.

5. See the recent piece in Chalkbeat.