Nearly a third of developers who use artificial intelligence tools every day spend four hours or more of their working day with them, according to Stack Overflow's latest survey of the profession.
The survey drew 30,903 responses from developers and technologists in 169 countries. Of the 10,926 daily AI users who answered the question about time, about 80 per cent said they used the tools for at least an hour a day.
Adoption is broad. Coding assistants or coding agents are used by 66 per cent of respondents and general-purpose chat tools by 63 per cent, while 26 per cent use automated agent workflows. Seventeen per cent said they did not use AI tools at all.
Trust is a different matter. About 7 per cent of developers said they trusted AI output for important work decisions. Forty-eight per cent said they trusted it when they could easily verify the output.
Seventy-nine per cent considered source attribution important or very important when deciding whether to trust an AI-generated technical answer.
Stack Overflow's chief executive, Prashanth Chandrasekar, said developers were highly sceptical as a group, and presented that as a feature rather than a problem. "As AI agents become rapidly more autonomous, that skepticism now acts as a critical safeguard," he said.
"As agents take on more work, developers need to know where an answer came from, what context informed the information that was produced or a decision that was made, and the most crucial piece - that they can actually verify its accuracy and relevance," he added. "Trust is not a given when it comes to developers; it is earned through transparency, source attribution, and rich technical context."
The company's own position in this market is not a neutral one. In early 2024 it introduced OverflowAPI, which sells access to its question-and-answer archive and requires participating AI providers to attribute answers to the relevant posts. A later partnership with OpenAI, which allowed the model builder to train on fifteen years of users' contributions, prompted a revolt among some contributors, a number of whom said they would rather delete their posts and lose their reputation scores.
The survey also found that 52 per cent of respondents had learned a new coding skill or programming language this year, with AI code-generation tools among the most common learning resources, behind technical documentation.
The two trust figures are the survey's real finding, and they are worth setting side by side. Seven per cent of developers trust the output for important decisions. Forty-eight per cent trust it when the output is easy to check.
The second figure therefore measures something other than the tool. A developer who can easily verify an answer is a developer who can already tell a right answer from a wrong one, which is the entire service being purchased.
This newspaper accepts the implication without complaint. The profession has established, across 169 countries, that an AI answer is trustworthy exactly when it is unnecessary, and untrustworthy exactly when it is not.
Our analysts note the time budget that follows. A third of daily users spend four hours or more with a tool they trust for important decisions seven times in a hundred, which makes it comfortably the most closely supervised member of the team.
The Department of Unnecessary Research supports the 79 per cent who want every answer to carry its source. Traced far enough, each one arrives at a person who typed it into Stack Overflow for nothing.