84% of developers are using or planning to use AI tools. 46% distrust what those tools produce. Both numbers come from the same survey.
Stack Overflow's 2025 Developer Survey has adoption up from 76% last year, and more developers actively distrusting the accuracy of AI output (46%) than trusting it (33%). Only 3% highly trust it, and the most experienced developers are the most cautious. The top frustration, named by 66% of respondents, is code that is almost right but not quite.
I think the interesting number is not adoption. It is the gap. A team can use these tools every day and still not trust the output, and that is a reasonable place to be. What it changes is code review. Almost right code looks finished. The old review habits, skim the diff and trust the author, were built for a different kind of mistake.
The way I think about it: if the author is partly a tool, the review has to carry the trust the author used to carry. That means reviewers reading for intent rather than style, and teams being explicit about which kinds of changes get a closer look.
How has your team's review process changed since AI tools arrived, if it has?
Photo source: https://photos.robertstowe.com/flowers