Stack Overflow's 2026 developer survey data, released in its annual retrospective at the end of September, captured a trend that's been building for a couple of years and finally crystallized into a clear contradiction: developers are using AI coding tools more than ever, while trusting them less than ever.

The numbers

AI coding tool adoption has climbed steadily and now sits at a record 84%, up from 79% in 2025, 62% in 2024, and 44% in 2023. At the same time, trust in AI output has fallen sharply: only 29% of developers now say they trust AI-generated code to be accurate, down from 40% in 2024 and from more than 70% expressing positive sentiment toward AI tools back in 2023, according to coverage of the survey by ByteIota.

AI agents are the fastest-growing category

The most striking shift is in AI agent usage specifically, as opposed to simple code-completion tools. Agent usage jumped from 31% of developers in the main 2025 survey to nearly double, 59%, in a smaller pulse survey run in April 2026. Within that, usage of Claude Code rose from 41% to 55% over the same window, while ChatGPT and GitHub Copilot remain the most widely used entry points overall, at 82% and 68% of developers respectively.

Why adoption and trust are moving in opposite directions

The pattern makes more sense once you separate convenience from correctness. Developers have learned that AI tools are reliably useful for speeding up boilerplate, drafting first passes, and handling repetitive refactors — reasons to use them constantly. But the same developers have also accumulated enough firsthand experience with subtle bugs, hallucinated APIs, and confidently wrong explanations to stop assuming the output is correct by default. Heavy use and low trust aren't contradictory once AI tools are treated the way experienced developers treat a fast but occasionally unreliable junior collaborator: useful for throughput, but always reviewed.

What this means if you're job hunting as a developer

• Fluency with AI tools is now a baseline expectation, not a differentiator. With adoption at 84%, not using these tools at all is more likely to raise questions in an interview than using them.
• Code review and verification skills are becoming more valuable, not less. As AI-assisted code volume grows, the ability to quickly spot when AI output is subtly wrong is a skill employers increasingly need and can't assume junior hires already have.
• Agent-specific experience is worth calling out separately from general AI tool use. Given how fast agent adoption is growing, candidates who can speak concretely about using coding agents for multi-step tasks — not just autocomplete — stand out from those who only mention AI tools in passing.
• Don't submit AI output you haven't personally verified, especially in take-home assessments or technical interviews — the same survey context that shows AI use is near-universal also shows experienced developers and interviewers are more skeptical of unreviewed AI output than at any point in the last three years.

The practical read for anyone building a developer career right now is that the bar has moved from "can you use AI tools" to "can you use them critically." That's a more demanding, but ultimately more durable, skill to build.