AI coding statistics: adoption, trust and agents

How many developers use AI coding tools?

Nearly all of them. 84% of respondents to the Stack Overflow 2025 Developer Survey use or plan to use AI tools in their development process, up from 76% the year before, and 51% of professional developers use them daily. Google's 2025 DORA report, from nearly 5,000 technology professionals, puts AI use at work at 90%.

How many developers trust AI-generated code?

Far fewer than use it. In the Stack Overflow 2025 survey, more developers actively distrust the accuracy of AI tools (46%) than trust it (33%), and 3% say they highly trust the output. In DORA 2025, 30% report little or no trust in AI-generated code. Use is high and trust is not, so people check the work.

How many developers use AI agents?

Most professional developers, as of mid-2026. JetBrains' Developer Ecosystem Survey 2026 (May to July, more than 15,000 professional developers) found 90% using AI coding agents at work at least weekly and 68% daily. A year earlier, Stack Overflow's 2025 survey said agents were not yet mainstream, with 52% not using them.

Do AI tools make developers faster?

It depends on who you ask and how you measure. In METR's randomized study, 16 experienced open-source developers working on 246 real issues in their own repositories took 19% longer with AI tools allowed. They expected a 24% speedup beforehand and still believed afterwards that AI had sped them up by 20%.

What do developers hand to an agent like Claude Code?

More of the whole task. Anthropic's analysis of 500,000 coding interactions found 79% of Claude Code conversations were automation, where the AI performs the task directly, against 49% on Claude.ai. Feedback loops, where Claude works on its own and the person sends errors back, made up 35.8% of Claude Code interactions.

What these numbers mean if you run coding agents

Put the sources side by side and one pattern stands out: people hand agents whole tasks (79% automation on Claude Code, about 47% of code fully agent-written in 2026) while most don't fully trust the output (46% distrust) and can misjudge their own speed (METR's 19% slower, felt as 20% faster). That is our reading, not a finding of any one study.

• Plan and progress you can see: an agent that writes its plan as steps and reports which step it is on gives you something to check other than the final diff. See how to see what your coding agent is doing.
• Questions answered quickly: 66% of developers are frustrated by answers that are almost right. An agent that stops and asks one specific question is cheaper than one that guesses. See answering a blocked coding agent.
• A review step after the run: a summary per finished task and a list of what the agent filed. See reviewing what your agent did.
• Measure your own speed: METR's gap between felt and measured speed is a reason to time a few tasks with and without an agent before deciding.

Taskhold is a task list built around the first three: agents plan, report progress and ask questions over MCP, and you watch and answer on one live board. The agent visibility check is a two-minute way to see which of these you already have.