The Stack Overflow 2026 Developer Survey, published on October 6, 2026 with 30,903 responses from 169 countries, found that two thirds of developers now use an AI coding assistant or agent, that Claude Code (65.5%) and GitHub Copilot (58.7%) are the two tools most of them reach for, and that only about a quarter have moved on to automated agent workflows. The headline from Stack Overflow’s own press release is the one worth remembering: sentiment is positive, but trust remains conditional.
Here are the numbers, what they do and do not tell you, and how to use them when picking tools. New to the vocabulary? Start with what an AI agent is.
The headline numbers
| Figure | Value | Source |
|---|---|---|
| Responses | 30,903 from 169 countries | Stack Overflow blog |
| Use coding assistants or agents | 66% of all respondents | Survey AI section |
| Use general-purpose chat tools for work | 63% | Survey AI section |
| Use automated agent workflows | 26.2% | Survey AI section |
| Claude Code, among assistant/agent users (n=12,255) | 65.5% | Survey AI section |
| GitHub Copilot, among assistant/agent users | 58.7% | Survey AI section |
| Kept the same tool for the past year | 34% | Survey AI section |
| Top reason to switch tools | Superior output quality | Press release |
Two things about the table. First, the tool percentages are from the 12,255 people who answered the tool question, not from all 30,903, so “66% use Claude Code” is only true of people already using these tools. Second, respondents could pick several tools, which is why the two leaders together exceed 100%.
Usage is not trust
Developers have adopted AI assistants at a pace few enterprise tools ever manage, and they still do not take the output on faith. The Register put it as developers being “hooked on AI but not totally sold on its judgment.” In practice: the tool writes the first draft, the developer reviews it like a junior colleague’s pull request, and the review step is not going away.
For teams, the metric that matters is not adoption. It is how much review time the tool saves net of the review time it creates.
Multi-tool stacks are normal
Only 34% of respondents stuck with one tool over the year. Everyone else switched or added. The stated top reason for switching was output quality, not price, not integration, not vendor loyalty.
Outside the survey, one roundup of vendor data cites a JetBrains AI Pulse survey in which Claude Code workplace adoption reportedly rose from 18% in January 2026 to 39% by the May to July period (Tech Insider summary). The same circle of vendor blogs claims that Copilot had around 4.7 million paid subscribers in early 2026 and that experienced developers run roughly 2.3 AI tools at once (Faros AI blog). Treat all of these as single-source claims from parties with something to sell; we mention them only because they point the same way as Stack Overflow’s independent data.
A developer with Copilot in the editor and Claude Code in the terminal is not being indecisive. They are using two tools for two jobs.
Terminal agents versus IDE assistants
Those two jobs are worth naming because the survey’s two leaders represent them.
IDE assistants
GitHub Copilot started as inline completion inside the editor. It is good at the small, constant work: finishing a line, writing a test that mirrors the one above it, suggesting the obvious next function. It fits an existing workflow without changing it. Visual Studio Magazine’s coverage notes VS Code still leads the editor rankings, which helps explain Copilot’s reach.
Terminal agents
Claude Code runs in the shell, reads the repository, edits several files, runs the tests, and reports back. It changes the workflow: you describe a task rather than typing alongside the tool. That is a bigger ask, so its position at the top of the usage chart says something about how fast developers accepted a new way of working once the output quality justified it.
The two styles are converging, but the split still predicts which one a developer opens for a given task.
What 26.2% on agent workflows means for next year
Automated agent workflows, where the agent runs a multi-step task with little or no supervision, sit at 26.2%. That is roughly 40% of the people who use assistants at all. It is the number to watch in the 2027 survey.
This week’s news points the same way. On November 4, Meta acquired Conductor, a startup whose product runs several coding agents in parallel on one codebase (TechCrunch). You do not buy an orchestration layer unless you expect many agents, not one assistant, to be the normal setup.
A new coding benchmark released this week reportedly showed agents still struggling on realistic tasks. We have only seen aggregator coverage, so hold it loosely, but it fits the trust finding: agents are good enough to use and not good enough to leave alone.
Expect the 26.2% to grow and the review step to stay. If you are building your own agents rather than using someone else’s, our framework comparison covers the trade-offs.
How to read vendor-sponsored surveys
The Stack Overflow survey is the closest thing the industry has to a neutral census. It is self-selected and skews toward people active on the site, but those limits are known and the methodology is published.
Vendor surveys are different. For any adoption chart that arrives in a press release:
- Check who paid for it and whether their product wins.
- Find the base. “66% of users” and “66% of developers” are different claims.
- Look for the question wording. “Have you tried” and “use weekly” produce very different numbers.
- Prefer figures that match an independent source. If only the vendor has the number, say “reportedly” and move on.
We applied those rules above, which is why the JetBrains, Copilot subscriber and 2.3-tools figures carry hedges and the Stack Overflow figures do not.
Our take
The survey confirms what we would have guessed: most developers use AI tools, they use more than one, and they switch when something produces better code. The surprise is the margin. A terminal-first agent that did not exist in its current form two years ago leads a 12,255-person sample, ahead of a tool that ships inside the world’s most popular editor.
What to do with that:
- Pick tools on output quality for your codebase, not on survey rank. Run both leaders on the same real tickets for a week.
- Budget for review. Treat agent output as a pull request from a fast, tireless, occasionally wrong colleague.
- Watch the agent-workflow number. If your team is in the 26.2%, you are early but not alone. If you are not, start with one well-scoped task.
Our tools directory tracks the coding agents in the survey and the ones that arrived after the fieldwork closed. For the weekly digest of this sort of data, join the newsletter.
Questions, answered
What is the most used AI coding agent according to Stack Overflow 2026?
Claude Code, used by about 66% of respondents who use coding assistants or agents, followed by GitHub Copilot at about 59%. Because people picked more than one tool, the percentages add up to well over 100%.
Do developers trust AI coding tools in 2026?
Not fully. Stack Overflow's own summary says sentiment is positive but trust remains conditional: developers use the tools daily while still checking their output before shipping it.
How many developers use AI agents in 2026?
66% of survey respondents use coding assistants or agents, and 26.2% use automated agent workflows that run tasks with less supervision. The survey had 30,903 responses from 169 countries.
