JetBrains 2026: 90% Devs Use AI Agents Daily

JetBrains published early results from its Developer Ecosystem Survey 2026: 90% of professional developers now use AI coding agents at least weekly, and 68% use them daily.

The bigger shock is the leaderboard: Claude Code jumped to 39% adoption and dethroned GitHub Copilot, which fell to 21%. Here is the context, what it actually means, and what I would do about it this week.

1. Context: What This Survey Actually Is

This is not a Twitter poll. The Developer Ecosystem Survey 2026 is the tenth edition of JetBrains’ large-scale, globally representative study, run by its Strategic Research and Market Intelligence team. The early AI-agent findings, presented by Mikhail Bogdanov on the JetBrains blog on August 18, 2026, cover more than 15,000 professional developers (heise online reports 15,509 participants) surveyed between May and July 2026, localized into eight languages.

Two things make the numbers credible: the survey is statistically reweighted by region, employment status, programming language, and familiarity with JetBrains products, and it continues a series — so we can compare against the January 2026 AI Pulse wave and the 2025 survey instead of staring at a single snapshot.

Why I trust the trend, not just the numbers

It is still one vendor’s survey, and “adoption” means self-reported use at work. But the direction is consistent across independent coverage (heise online, i-programmer, GIGAZINE), and the panel design lets us see movement over time. Treat exact percentages as estimates — treat the direction as fact.

2. The Scoreboard: Agents Went Mainstream

As of May–July 2026, 90% of professional developers use AI coding agents at work at least weekly — local agents or cloud agents — and 68% use them daily. Agents are no longer an experiment for early adopters. They are routine infrastructure, like version control.

📊

90% weekly / 68% daily

All professional developers

Agent usage at work. Daily users are now the majority — the autocomplete era quietly ended.

👑

Claude Code 39%

Up from 18% in Jan 2026

From ~3% in early 2025 to the most-used AI coding tool at work. 47% in the US — nearly one in two developers.

📉

Copilot 21%

Down from ~29–31%

The former leader held 29–31% from mid-2025 to early 2026, then dropped. Same 79% awareness as Claude — but half the adoption.

🚀

Codex 16%

Up from 3% in January

More than quintupled in six months. Awareness jumped from 27% to 65% — OpenAI distribution still matters.

🔒

31% single-tool users

Claude Code stickiness

Claude Code is the single most-used tool for 31% of devs — an ~80% conversion rate from trial to main tool.

🌍

Cursor 12%, OpenCode 7%

The open field

Cursor slipped from 18% (and 28%→16% in China). OpenCode hit 7% adoption with 42% mindshare and no big brand behind it.

My favorite stat is the awareness tie: Copilot and Claude Code both sit at 79% mindshare, yet adoption is 21% versus 39%. Developers have heard of both — they chose one. That gap between knowing and using is where the real signal lives.

3. Four Implications I Take Seriously

Numbers without interpretation are trivia. Here is what I actually take away from this shift — and what I would bet on for the next twelve months.

🤖

The autocomplete era is over

68% daily usage means agents write code inside real workflows, not demos. If your team still treats AI as “fancy autocomplete,” your mental model is two generations behind. Plan reviews, branching, and testing around agent-generated diffs — that is the default input now.

🧲

Retention beats awareness

79% vs 79% awareness, 39% vs 21% adoption, ~80% trial-to-main-tool conversion for Claude Code. When you evaluate tools for your team, stop counting feature lists and start measuring 30-day retention: who is still using it after the novelty wears off?

🏟️

The field is still wide open

Codex quintupled in six months, an open-source agent without corporate backing reached 7%, and Cursor lost a third of its share. Do not marry one tool in 2026. Build your workflow (rules files, evals, review gates) so the agent underneath is swappable.

🔌

The moat is IDE integration

39% of Copilot users run it inside JetBrains IDEs, Claude/Codex/Copilot/OpenCode plug into JetBrains AI chat, and Cursor joined via the Agent Client Protocol. The winner will not be the best model — it will be the agent with the deepest context in your repo. Choose by integration depth, not benchmark scores.

Golden rule

Adoption numbers tell you where the ecosystem is going, not which tool fits your codebase. Copy the direction (agent-first workflows), not the leaderboard (someone else’s favorite agent).

4. What to Do This Week

Five concrete moves, each under two hours. I have done the first three with my own projects, and they changed how much I trust agent output:

✅ Your 5-step agent audit

1. Run a one-week agent trial: Pick one real branch and do it agent-first (Claude Code, Copilot agent mode, Codex — your call). Track merged PRs, review cycles, and rework rate. One week of your own data beats ten benchmark posts.
2. Write the rules file first: Before blaming the tool, add an AGENTS.md or rules file with your stack, conventions, and test commands. Most “agents are dumb” complaints I see are really “my repo has zero machine-readable context” problems.
3. Move one boring task to agent-first: Test generation, migrations, changelogs, boilerplate CRUD — pick one repetitive flow and make the agent the default author with you as reviewer. That is exactly the 68%-daily-use pattern the survey describes.
4. Stop copy-pasting from chat: Check agent support inside your IDE (JetBrains AI chat with ACP, VS Code agent mode). 39% of Copilot users already work inside JetBrains IDEs — in-editor agents with repo context beat browser tabs every time.
5. Set guardrails before scaling: Branch-only work, mandatory human review, no secrets in prompts, and a defined blast radius. JetBrains is already building governance layers (Central) because fleets of agents need oversight. Do the lightweight version now.

Sources

Conclusion

The headline is simple: AI coding agents crossed the chasm in 2026. 90% weekly and 68% daily use, a new leader in Claude Code, and a former monopolist cut nearly in half — that is what a platform shift looks like while it is happening.

But the leaderboard is the least durable part of this story. Tools will keep swapping places — Codex quintupled in six months, open-source agents are climbing without marketing budgets. What compounds is workflow discipline: rules files, review gates, and agent-first habits. Build that, and you win no matter which agent tops next year’s survey.

The story in 30 seconds

The fact

• 90% weekly, 68% daily agent use
• 15,000+ devs, May–Jul 2026
• 10th edition, reweighted panel

The upset

• Claude Code 39% (from 18%)
• Copilot 21% (from ~29%)
• Codex 16% (from 3%)

Your move

• 1-week trial on a real branch
• AGENTS.md before new tools
• Review gates + guardrails

JetBrains 2026: 90% of Developers Now Use AI Coding Agents