Developers are using AI more and trusting it less. That's healthy.

What happened

Stack Overflow's 2025 Developer Survey drew nearly 50,000 responses. Two numbers moved in opposite directions.

Use kept growing: 84% of developers said they use or plan to use AI tools. Trust went the other way. Only 33% said they trust the accuracy of AI output, down from 43% a year earlier. Favourable sentiment toward AI tools slipped from 72% to 60%.

It isn't all gloom. Among developers using AI agents at work, 69% said productivity is improving. And about three in four said that when they don't trust an AI answer, they ask a person.

So what for your business

Falling trust alongside rising use sounds like a contradiction. It's what maturing looks like. Developers have worked out where the tools shine (boilerplate, tests, unfamiliar syntax) and where they quietly make things up (subtle logic, architecture, anything that depends on context the tool can't see).

The risk for a business isn't developers using AI. It's code arriving faster than anyone reviews it. If output doubles and review capacity doesn't, defects get through. Review is where the trust gap gets handled, so that's what to protect.

What to do this month

1. Hold AI-written code to the same review bar. No exceptions for "the model wrote it".
2. Insist on tests. Ask for tests with every generated change, and read the tests first.
3. Keep diffs small. Nobody can properly review a 2,000-line generated change. Ask for smaller pieces.
4. Track escaped defects, not lines of code. If bugs reaching production rise as AI use rises, you'll see it in that number before you feel it.
5. Decide where AI isn't welcome. Authentication, payments and data migrations are good candidates for extra caution.