Academia and tech are one system now

For most of the last century there was a workable story about who did what. Universities pursued questions with no obvious buyer; companies took the answers and made them into products. That story has stopped describing reality, and the people harmed most by the fiction are early-career researchers who are told to choose between two things that are already the same thing.

The compute is in companies. So, increasingly, are the datasets, the deployment surfaces, and the salaries. Meanwhile the questions that most need patience — safety, evaluation, long-horizon health outcomes — are exactly the ones a quarterly cycle cannot hold. The result is a system where the capability and the patience live in different buildings.

Capability moved. Patience didn't. That gap is the whole problem.

I don't think the fix is to make universities more like startups, which is the reflex of the last decade and mostly produced worse universities. The fix is to be explicit about what only an academic institution can hold: the long clock, the freedom to publish an unflattering result, and the obligation to teach.

Practically, that means joint appointments treated as normal rather than suspicious; compute access negotiated as infrastructure rather than as a favour; and evaluation work funded as a public good, because nobody's product roadmap will ever fund the thing that might say no.