AI coding tools were supposed to shrink dev teams — so why hasn't that shown up in the job numbers?

This explores why AI coding tools haven't led to fewer programmers overall. The first thing the corpus does is correct the premise a little. Stanford's 2026 AI Index found that headcount did shrink in one place: employment among software developers aged 22 to 25 fell by about 20%. The much larger layoffs people expected haven't appeared in the aggregate data Is AI already shrinking the entry-level job market?. So the better question is why the cuts stayed small and why they fell on the youngest programmers.

Part of the answer is that the speedup is smaller and less reliable than the headlines suggest. In a randomized trial, 96 Google engineers finished a complex task about 21% faster with AI features, though the margin of error was wide Do AI coding features actually speed up engineer productivity?. A different randomized trial, with experienced open-source developers working in codebases they knew well, found early-2025 tools made them 19% slower, even though they had predicted a 24% speedup Do AI coding tools actually speed up experienced developers?. At Anthropic, engineers reported 50% productivity gains and merged 67% more pull requests. Yet most of them could fully hand off only 0–20% of their work Does AI assistance erode the skills needed to oversee it?. More output per person doesn't mean fewer people when most of the job still needs a human in the loop.

What keeps the human in the loop is checking the AI's work. Stack Overflow's 2025 survey found that 80% of developers use AI tools, while trust in their accuracy fell from 40% to 29%. The main complaint was code that looks right but contains subtle errors Why do developers keep using AI tools they don't trust?. Someone has to catch those errors, and the time spent catching them eats into the time saved. Interviews with junior developers found the same pattern: they use AI mainly for work they can already evaluate and avoid it where they can't Do junior developers choose AI based on their ability to verify results?. AI makes people more productive in the areas where they're already competent. That favors experienced engineers over newcomers.

That explains why the cuts landed on the youngest workers, and it points to a problem. The routine work that AI absorbs is how juniors used to build skill. In a randomized trial, developers learning a new library with AI help ended up with weaker understanding and worse debugging ability, unless they engaged actively with what the AI produced Does AI assistance actually harm the way developers learn?. Anthropic's engineers worried that leaning on Claude erodes the hands-on practice they need to catch its mistakes Does AI assistance erode the skills needed to oversee it?. Nielsen counters that tests which take AI away measure a situation that never happens at work. In his view, the real question is what higher-level skills develop when AI is always available Does removing AI tools actually measure real skill loss?. In either case, the industry may be cutting the entry-level roles that produced its future senior engineers.

Finally, faster code writing doesn't remove the other bottlenecks in software work. Evans argues that making tools easier to build doesn't solve two harder problems: most workers don't see their own tasks as automatable, and companies need decisions across departments before anything changes Does easier tool-building actually solve enterprise adoption problems?. Even inside Microsoft, engineers started using Copilot CLI because their peers did, not because of their seniority or tenure. Adoption spread unevenly through social networks rather than all at once Do social networks drive adoption of new coding tools?. The corpus has no direct data on company-wide hiring decisions, so this is an inference: gains that are uneven, partial, and limited by the need to verify are not enough to justify cutting staff on the scale people predicted.