AI-exposed majors lose 5 points of hiring and 13% of starting pay
Two fresh data points land the same way: the money AI is saving is showing up as losses somewhere else — in the pay of the graduates who were supposed to build it, and in the electricity and hardware nobody budgeted for.
Computer science graduates are entering the worst job market since the financial crisis, according to a new working paper from three US Census Bureau economists. Using administrative records on college graduates, the paper tracks how outcomes moved for majors with differing exposure to AI, and finds the most-exposed decile began diverging the moment ChatGPT arrived in late 2022. In regression-adjusted estimates, the likelihood of initial employment fell five percentage points, and full-quarter starting earnings fell 13% — a decline the authors describe as comparable in magnitude to the losses from graduating into a large recession.
The mechanics matter more than the headline number. Roughly half of the earnings hit came from lower pay inside the industries that hire these graduates; the rest came from a shift in the industry mix, as more of them ended up in lower-wage sectors such as restaurants and retail. The effect attenuates as graduates move further from labor-market entry, which is the reassuring part, but it stays substantial for the most exposed majors. Read alongside the entry-level gap we covered earlier — AI's entry-level gap hits 19% — the pattern is consistent: the ladder's first rung is where the displacement shows up first, and it shows up in wages before it shows up in unemployment.
KDE turns 30 this year, and its annual conference is putting an AI-native desktop proposal on the agenda — one that would compile Plasma around a personal model of each user. The Akademy session in Graz sketches a concept called Kadai: an encrypted, vendor-agnostic personal kernel from which Plasma assembles each Activity, built on the shell's existing Activities, Plasmoids and scripting surface. To get there, the speakers want upstream changes that are architectural rather than cosmetic — declarative reconciliation, per-widget capability scopes, richer Activity metadata.
What it is not: an announced KDE product or an adopted roadmap. It is a conference talk from two longtime contributors, with 10 minutes of open Q&A scheduled as the forum to argue with it, and nothing has landed in Plasma. That framing will be familiar from the sovereignty debate in European open source, where the same contributors have leaned on the Resonant Computing Manifesto — a document that drew heavy criticism last December. The take worth holding: the AI is the least interesting part. A shell that assembles itself per user only works if that model stays local, encrypted, portable and revocable — which is precisely what no cloud assistant can offer, and what a Linux desktop could deliver first.
What to watch: whether Plasma maintainers pick up any of the three asks, or whether Kadai stays an anniversary talk — and whether the Census paper's earnings effect persists as these cohorts age.
If starting pay is where AI shows up first for graduates, what should a CS major do differently this year? Tell us in the comments.
