The Infrastructure Was the Easy Part

On June 2, President Paul Alivisatos wrote to the University of Chicago community with the shape of a decision that had, quietly, been coming for a year.¹ The university has partnered with Anthropic. Every student, every faculty member, every staff member will receive free enterprise access to Claude. Faculty and staff begin in July. Students complete the rollout before the start of the 2026-27 academic year in September.¹ Roughly thirty-seven thousand people, from postdocs to first-years to registrars, will open a browser next quarter and find a working model waiting for them.²

Twenty-eight days later, on the last day of June, a chemistry professor named Mark Levin wrote back. His op-ed in the Chicago Maroon carries a title from an older grammar. Welcoming Claude Into the Classroom Risks the Soul of Education.³ Levin is careful and specific. He teaches with chalk. His subject, organic chemistry, is one of the few corners of an undergraduate degree where the model still fumbles, and his students still take notes on paper.³ What alarms him is not what a first-year in his section might do with the tool. It is what his colleagues might do with it. Students at UChicago, he writes, are paying large sums of money for a world-class, artisanal educational experience, and they expect a product that cannot be replicated by prompting a bot.³

His deeper worry, spoken plainly, is about muscles. The models, he says, promise productivity while subverting the process of grappling with ideas, and with overuse they weaken the very muscles necessary to use them responsibly.³ It is the oldest anxiety in the pedagogy of any new medium. It is also, this time, the exact right one.

What the Announcement Was Really About

Read carefully, the June 2 letter does not solve the pedagogical question. It cannot. What it does is make a quieter decision that had already, in practice, been made. The model is no longer a private choice a student makes on her own dime, subject to no institutional visibility. It is now something the institution has provided. Claude Code and Claude Cowork, both included in the plan, can read and modify the folders on a user's desktop.¹ The distinction between the writing of a paper and the writing with a model has, for practical purposes, dissolved.

That acknowledgment is not an endorsement of Levin's colleagues integrating the model badly. It is closer to the opposite. It says, out loud, that the model is now infrastructure. The question is not whether it is present. The question is what happens next.

The Muscles Levin is Worried About

The muscles Levin names get built inside a moment. Sitting with a paragraph until it collapses under its own weight, then choosing what to keep. Recognizing, mid-sentence, that the argument was wrong and going back. Discovering, on the fourth attempt at an opening, that the essay is really about something else. Every one of those moments is a repetition. Every repetition is a strength added. Overuse of a model that finishes the sentence for you is not, in Levin's frame, a moral failure. It is a missed rep.

The instrument most classrooms use to measure whether the reps happened is the finished paper. And a polished draft is silent about its own history. It looks the same when a student wrestled with three unsatisfactory openings as when she took the model's first suggestion and moved on. The muscle either grew or it did not, and by the time the paragraph is printed, the record of which is gone.

What the Room Would Have to See

A serious response to Levin's worry, on a campus where the tool is now infrastructure, is not a policy about acceptable use. It is a design question about what the workspace holds onto. If the paragraph is the exit, the reps are the corridor, and the corridor is where the muscle either grows or atrophies. A room that can see the corridor can coach it. A room that only ever grades the exit will lose the argument to Levin's future without ever knowing it did.

The corridor is not exotic material. It is the sequence of revisions, kept. The pauses, long enough to notice which sentences were held longest. The exchange with the model, preserved as evidence of thinking rather than confiscated as evidence of cheating. A teaching assistant who can open a document at the end of a week and read the interior alongside the exterior is a teaching assistant who can tell Levin's story with actual detail. She saw the fourth attempt. She saw the moment the student pushed back on the model's suggestion. She saw the paragraph the student kept, and the three the student refused.

None of that is any longer technically hard. What has kept the interior of writing invisible was never engineering. It was a paper-era habit of throwing away everything except the final page. In a paper era, that was reasonable. In an era where the tool sits inside the working life of every student in a top-thirty university, and in the working life of a professor teaching organic chemistry with chalk, the habit is no longer holding up its side of the bargain.

What the Fall Quarter Will Reveal

By September, thirty-seven thousand people at UChicago will have a model.² Some, in Levin's phrase, will skip the reps. Some will use it exactly the way any careful description of an AI-fluent learner would ask, pushing back, asking for feedback on the specific place their understanding thinned, letting themselves be wrong on the way to being right. The finished papers will look, from ten paces, the same. The question is whether the university, or any of the others watching from a distance, builds any way to tell the two apart.

The infrastructure was the easy part. The room is still the hard part.

If Levin's colleagues, next fall, wanted to see whether their students were building muscle or losing it, what would their classrooms actually have to change?