David Laszczkowski
Co-Founder, EdPilot
Right now, most universities' AI strategy is "allow students to use whatever they find." That's not a strategy. It's an absence of one — and the costs of that absence are accumulating.
Right now, most universities' AI strategy consists of two things: a policy document describing what students are and aren't allowed to do with AI, and implicit permission to use whatever tools they find on their own.
That's not a strategy. It's an absence of one. And the costs of that absence are accumulating in ways that are hard to see because they're distributed across millions of individual student interactions that no institution has any visibility into.
Saying universities need their own AI infrastructure sounds expensive and technically daunting. The practical version is more tractable than it sounds.
It doesn't mean every university builds its own language model. It means every university has a deployed AI environment that it governs — where the knowledge the AI draws from is defined by faculty, where the behavior of the AI is configured by people who understand the academic context, where interactions are visible to the institution, and where the platform is accountable to institutional policies rather than to a consumer product roadmap.
The difference between this and "use ChatGPT" is significant. Consumer AI products are built to be useful to everyone. Institutional AI infrastructure is built to be right for this context — this curriculum, these standards, this population of students.
There's a strategic dimension to this beyond the immediate educational context.
When universities become dependent on external AI platforms for a core part of their educational delivery, they've outsourced a significant amount of institutional knowledge to a third party. The course materials that get fed into these systems. The interaction patterns that reveal where students struggle. The aggregate data about what's working and what isn't in a given course.
All of that currently disappears into platforms that universities don't own, can't inspect, and can't guarantee will retain the same terms of service next year that they have today.
This isn't hypothetical risk management. It's the kind of dependency that becomes obvious in retrospect and expensive to unwind.
The most immediate argument for institutional AI infrastructure is the visibility it creates.
Right now, most faculty don't know how their students are using AI. They don't know what questions students are asking, what confusions they're carrying into class, or where the course materials are failing to communicate what they're supposed to communicate.
An institutional AI environment that's grounded in course materials and visible to faculty generates all of that. Not as a surveillance mechanism — as a feedback loop. Which concepts are generating consistent confusion? Which readings are students treating as authoritative? Where are the gaps between what the course assumes students know and what they actually know going in?
That feedback loop is valuable independently of any integrity concern. It makes courses better.
Building this doesn't require a massive upfront investment or a years-long implementation project. The practical starting point is a course-level deployment — one faculty member, one course, one semester.
Give a professor a tool to upload their course materials and configure how the AI responds to students. See whether it changes the quality of student questions in office hours, whether it surfaces confusions that the professor can address in class, whether students who use it perform differently than students who don't.
Universities that run those pilots now will have real data and real experience to draw on when the infrastructure question becomes unavoidable. That point is coming regardless of what institutions decide in the meantime.