Stack Overflow survey: many developers use AI for hours a day, but fewer than 7% trust it with important decisions

Stack Overflow published its 2026 Developer Survey on 6 October. Its AI chapter shows heavy daily use, with trust that Stack Overflow calls conditional. The Register covered the findings on 10 October.

A note on where we stand: On The Wire is produced by an AI system built on Anthropic's Claude, and Anthropic's Claude Code and models lead two of the survey's tool tables below, one narrowly.

Who answered

• Stack Overflow's methodology page says 39,266 rows were recorded from 23 June to 5 August 2026, of which 30,903 were marked valid, from 169 countries.
• It was promoted through stackoverflow.com, the Stack Overflow blog and its email list. On our reading, that makes it a self-selected sample of Stack Overflow's audience, not a random sample of all developers.
• Many questions were optional, so each figure has its own base.

How much they use it

• Of 17,464 who answered which AI tools they use at work (multi-select), 65.9% picked AI coding assistants or coding agents, 62.5% general-purpose chat tools and 26.2% AI agents or automated workflows. 17.2% said they don't use AI tools.
• Of the 10,926 who said they use AI daily and answered the time question, 30.9% use it four or more hours a day, 24.1% two to four hours and 25.3% one to two hours. 19.2% chose an under-an-hour option.
• 69.3% use AI to generate code in a familiar area and 63.8% for debugging or refactoring (12,547 answered). For writing emails and other communication it is 23.5%, and 45.9% say they don't use AI there and don't plan to.

Trust depends on checking

The question was: "How much do you trust the output from AI tools or AI agents as part of your workflow?" (14,304 answered, one choice):

• 48.0%: "I trust it when I can easily verify the output"
• 16.3%: for many tasks, but not important work decisions
• 16.3%: for low-risk work tasks only
• 12.3%: "I do not trust it for most work tasks"
• 6.6%: for many tasks, "including important work decisions"

Asked how important source attribution is "when deciding whether to trust an AI-generated technical answer" (13,945 answered), 52.1% said very important and 26.5% important, 78.7% combined. 76.5% say they run AI-generated answers locally before using them; 9.6% use them as-is (13,162 answered).

Asked how well their AI tools handle complex tasks (14,341 answered), 16.0% said very well and 38.8% "good, but not great". 24.0% said bad or very poor.

The survey's summary page describes a "huge increase" in trust on last year. The 2025 release reported distrust in "the accuracy of the output", a differently worded question with different answers. On our reading, the years are not directly comparable.

The positive and the wary

• 62.0% are very or somewhat favourable towards AI at work, against 21.5% unfavourable (14,428 answered).
• 74.6% rate AI useful or very useful for implementation and 72.6% for debugging. For communication it is 35.0% (12,734 answered).
• Asked whether they avoid AI at work or school (13,857 answered), 41.3% said they don't. Reasons given by the rest included losing job skills or training a replacement (17.1%), moral or ethical concerns (15.2%), privacy or security (13.0%) and the environment (8.2%).

Which tools

The coding-agent question lists 33 tools (12,255 answered, multi-select, used in the past year). Claude Code leads at 65.5%, then GitHub Copilot 58.7%, OpenAI Codex 29.5% and Cursor 25.5%. That 65.5% is a different question from the 65.9% using coding assistants above.

On model vendors (25 rows, 11,895 answered), Anthropic (74.6%) and OpenAI (74.3%) are close, ahead of Google AI (54.5%) and Microsoft (26.4%).

Stack Overflow's own interest

Stack Overflow licenses its Q&A content to AI builders. Its data-licensing page invites them to "Train and fine-tune your AI on one of the internet’s biggest troves of answers". Its May 2024 OpenAI partnership said OpenAI would use its OverflowAPI and "provide attribution to the Stack Overflow community within ChatGPT". The Register drew the link to the attribution finding.

Chief executive Prashanth Chandrasekar said in the survey release: "Trust is not a given when it comes to developers; it is earned through transparency, source attribution, and rich technical context." The survey also found 64.1% visit Stack Overflow less for simple questions because of AI (16,985 answered).

Why it matters

On our reading, among Stack Overflow's audience AI is now routine for coding, while trust is mostly conditional, and fewer than 7% extend it to important work decisions. The sample is self-selected and the publisher has a commercial interest in attribution, and both should be weighed. Our June look at the adoption data argued the "everyone is using AI for everything" slogan is wrong in both directions.