84% of Developers Now Use AI Tools. Only 29% Trust Them. That Gap Is Where the Work Is

The 2025 Stack Overflow Developer Survey collected more than 49,000 responses from 177 countries. Two of its findings sit awkwardly together.

The first: 84% of respondents are using or planning to use AI tools in their development process, up from 76% the year before. Among professional developers, 51% use them every day.

The second: trust has fallen. Positive sentiment toward AI tools dropped from over 70% in 2023 and 2024 to 60% in 2025. Only 29% of respondents said they trust AI output, down 11 percentage points in a single year.

Adoption up, trust down. That is not a contradiction, and it is not a scandal. It is what happens when a technology moves from novelty to tool.

What falling trust actually indicates

When something is new, people either dismiss it or overrate it. Neither position requires much experience. Trust scores from that period measure enthusiasm, not evidence.

After a year of daily use, developers have watched an assistant produce a function that looks correct, passes a quick read, and fails on an edge case nobody thought to check. They have also watched it write a week of boilerplate in an afternoon. The result is a more accurate mental model: useful for some things, unreliable for others, and requiring review either way.

Declining trust alongside rising usage is a sign of a workforce that has stopped guessing and started measuring.

What this means for anyone commissioning software

If you are paying for a product, three consequences follow.

Review capacity is now the constraint, not typing speed. When generating a plausible-looking implementation costs almost nothing, the bottleneck moves to the person who decides whether it is correct. A team that produces more code than it can meaningfully review has not become faster. It has moved its work into the future, where it will be more expensive.

"We use AI" says nothing about quality. Since 84% of developers use these tools, a supplier telling you they use AI is telling you they are ordinary. The meaningful question is what they do about the 71% of developers who do not trust the output: what gets reviewed, what gets tested, and who is accountable when something reaches production and fails.

Some tasks have genuinely changed. Test scaffolding, data transformation, migration between similar frameworks, documentation drafts, and unfamiliar-API exploration are materially faster. Domain logic where correctness is not obvious from reading is not. The distinction is whether a human can verify the output quickly. If verifying takes as long as writing, the assistance was illusory.

How we use it

We use AI assistance daily across mobile, web and AI work, and we hold three rules.

Generated code is reviewed by a person who can explain it. If nobody on the team can explain why it works, it does not ship, regardless of whether the tests pass.

Money, authentication, and personal data get written by hand and reviewed twice. The cost of being wrong is asymmetric, and "the assistant suggested it" has never been an acceptable explanation to a client.

We do not bill for time we did not spend. If a tool saved a week, the estimate reflects that. Charging for effort the tool removed is a short-lived business.

The trust gap is not a problem to be solved. It is the correct professional response to a tool that is right most of the time. The teams that will do well are the ones that build their process around "most", not around "all".

Source: 2025 Stack Overflow Developer Survey (49,000+ respondents, 177 countries).