The Seam in the Pipeline

Stanford’s undergraduate computer science degrees fell by about 70 students this year — almost 14 percent in a single cycle. That’s a sharp drop at a school that, for a long stretch, felt like the factory floor for Silicon Valley.

A single year is not a trend. Mehran Sahami, Stanford’s computer science chair, called it natural variation — cycles the department has seen before. He’s right to be cautious. But the number lands inside a pattern that’s harder to wave off.

Nationally, computer science enrollment at four-year colleges is down roughly 8 percent, after nearly two decades of uninterrupted growth. Entry-level software jobs have thinned. Laid-off mid-career engineers are competing with new graduates. AI tools now write a lot of the code that used to be the first rung on the ladder.

I got curious how MIT looked next to Stanford. The comparison is messy in a useful way.

MIT doesn’t have one “computer science” degree — computing is split across several Course 6 majors. Classic Computer Science and Engineering (6-3) awarded 265 bachelor’s degrees this year. The newer Artificial Intelligence and Decision Making major (6-4) awarded 114. Electrical Engineering and Computer Science (6-2) added 72 more, plus smaller joint programs.

Look at majors instead of diplomas and the shift sharpens. 6-3 slipped from 706 students to 672 — about 5 percent this year, and down 18 percent from its 2022 peak of 823. 6-4, which started in 2022 with 37 students, is now at 372. Intro programming enrollment has been sliding since its 2022–23 high. Course 6 overall fell last year for the first time in a decade — but only modestly.

Stanford’s story is a clean 14 percent drop in one degree. MIT’s is a reallocation: students moving from traditional software construction toward a track built around judgment and decision-making instead of a coding sequence. The footprint is still large. The composition has already changed.

That distinction matters.

After the dot-com bust, CS enrollment fell too. Sahami noted that the students who enrolled then graduated straight into the Facebook and Google growth years. A downturn in the pipeline isn’t automatically a verdict on the field. It can be a lag. It can also be a signal that the old promise — learn to code, collect a golden ticket — no longer maps onto the first job.

I don’t know which this is yet. A Census Bureau paper cited in the Stanford piece found that recent graduates in highly AI-exposed fields have seen employment and starting-job quality deteriorate on a scale comparable to a large recession. That’s not a cycle to dismiss. It’s also too early to call it destiny. Models change. Firms eventually spend the productivity they just captured. What’s left tends to look less like junior ticket-closing and more like judgment, systems design, knowing what the tools should be pointed at.

I’ve watched enough technologies overshoot and settle to be wary of panic and nostalgia alike. The question isn’t whether computer science is “over.” It isn’t. It’s what kind of computing education still compounds once the first-year coding job stops being the obvious on-ramp.

Stanford is still talking in cycles. MIT has already built a major that assumes the on-ramp has moved. Both may be right for a while.

The seam, as usual, is in the middle — between the degree that used to be a ticket and the skill that still is. That’s the part I’ll keep watching.