Stack Overflow says results from its 2026 Developer Survey, which opened on 23 June 2026, are imminent. Its 2025 survey set the baseline, with AI usage among developers rising from 76% to 84% while trust in AI output fell from 40% to 29%.
What the 2025 survey found
Stack Overflow reported that 84% of developers used AI tools in 2025, up from 76% the year before. Over the same period, the share who trusted AI output dropped from 40% to 29%. The pattern was sharper among people learning to code. Stack Overflow found AI tool use in that group rose from 63.0% in 2024 to 73.3% in 2025, while favourable sentiment fell from 72.1% to 52.8%. A separate Stack Overflow survey on AI agents found agent usage had doubled while concerns about AI-generated code grew.
Use and trust are moving apart
Stack Overflow's figures show developers using AI more each year and believing its output less. Most keep using the tools regardless, so a growing share of the code in your repository is written by a model and then checked by a person who doesn't fully trust it. Stack Overflow's July 2026 post on tooling made a related point in its title, arguing that developers are attached to tools because tools encode trust. When the tool can't earn that trust on its own, the engineer reviewing its output has to supply it.
What it means for engineering teams
Your team's output now depends heavily on how well engineers read, test and reject generated code. Writing code got faster, and the slow, expensive part moved to deciding whether that code is correct, secure and maintainable. With agent usage doubling, according to Stack Overflow, engineers are reviewing larger changes they didn't write line by line. Teams that treat review as a formality ship the 71% of output their own developers say they don't trust, using Stack Overflow's 2025 trust figure as a rough guide.
What it means when you hire engineers
Verification is now the skill to test for in interviews. A candidate who produces a lot of code with an assistant tells you little about whether they'd spot a subtle bug in it, question a dependency it pulled in or notice it solved the wrong problem. You learn more by watching how someone works without AI, then how they direct and check their own agents on the same kind of task. The drop in favourable sentiment among learners, reported by Stack Overflow, also suggests newer developers are meeting AI before they've built the judgement to audit it.
Our vetting runs two sixty-minute assessments in a real codebase. Assessment 1 has AI tools off. Assessment 2 has the candidate's own agents on, and both scorecards are shared with the client. We'll update this item when Stack Overflow publishes the 2026 results. The full assessment format is on how we vet.
Sources
