Analysis
Stack Overflow's latest developer survey, covering 30,903 responses from developers and technologists in 169 countries, found that daily AI users spend heavy time with the tools but remain skeptical of trusting them unsupervised, according to The Register. Of the 10,926 daily AI users who answered a time-use question, about 80% spend at least an hour of their workday with AI tools, and nearly a third spend four or more hours.
High Usage, Low Trust
The trust numbers are the survey's sharpest finding: only about 7% of respondents trust AI output for important work decisions outright, while 48% say they trust it specifically when they can easily verify the result themselves. Stack Overflow CEO Prashanth Chandrasekar frames that gap as a feature, not a problem: "As AI agents become rapidly more autonomous, that skepticism now acts as a critical safeguard."
“Stack Overflow's own relationship with AI training data adds a wrinkle to how cleanly this skepticism finding should be read.”
Tool adoption is already broad: 66% of respondents use coding assistants or agents, 63% use general-purpose chat tools, and 26% already run automated agent workflows -- real numbers for a category still being pitched to some enterprises as early-stage. Only 17% of respondents reported using no AI tools at all. Separately, 79% said source attribution matters when deciding whether to trust an AI-generated technical answer, and 52% said they learned a new coding skill or language this year, with AI code-generation tools ranking among the most common learning resources, just behind technical documentation.
Stack Overflow's own relationship with AI training data adds a wrinkle to how cleanly this skepticism finding should be read. The company launched OverflowAPI in early 2024 to sell paid, attribution-required access to its Q&A content, then struck a May 2024 partnership letting OpenAI train on 15 years of user contributions -- a deal that triggered user protests, with some threatening to delete their posts rather than let them train a competitor's model. A company that monetizes developer content for AI training has some incentive to frame developer trust in that same AI as cautious rather than enthusiastic.
Even accounting for that, the usage-versus-trust gap matches what coding-assistant vendors have been building toward all year: not raw autocomplete, but tools that make verification fast -- inline diffs, generated tests, and citations back to source material. The 48% figure is the more commercially relevant number than the headline 7%, since it says developers aren't rejecting AI judgment outright, they're conditioning it on being able to check the work quickly, which is a product design problem rather than a trust problem that scales with model capability alone.
