Stack Overflow’s Developer Survey — nearly 50,000 respondents across 177 countries — puts AI coding tool adoption at a record 84% for 2025, up from around 70% in 2023. Trust in the output has moved the opposite direction: only 29% of developers say they trust AI-generated code to be accurate, down from 40% a year earlier, and just 3% say they “highly trust” it (2.6% among experienced developers specifically). Forty-six percent actively distrust what the tools produce.
Adoption and confidence have decoupled
This isn’t developers rejecting AI tools — usage keeps climbing every year. It’s that using a tool constantly and trusting its output unconditionally have turned out to be two separate things. The top complaint, cited by 66% of respondents, is code that’s “almost right, but not quite” — plausible-looking, confidently wrong, and expensive to catch late. Forty-five percent say debugging AI-generated code now takes longer than writing the equivalent by hand would have.
Why this matters for anyone commissioning software
If a development team’s pitch is “we use AI, so it’s faster and cheaper,” that’s now table stakes, not a differentiator — 84% of teams can say the same thing. The question worth asking a potential partner isn’t whether they use AI tools, it’s what sits between the AI’s output and what ships: code review discipline, test coverage, a second model or a senior engineer checking the first one’s work. Teams that treat AI output as a first draft rather than a finished product are the ones closing this trust gap in practice, not just in adoption stats.
So what
The trust gap is a buying signal, not just a developer-survey curiosity — it tells you what to ask about process, not just tooling, when you’re choosing who builds your product. We treat every AI-generated change as a draft that needs review before it ships, using a deliberate multi-model check rather than a single tool’s unverified output. More on how that works on the AI-assisted development page, or get in touch to talk through a project.