39% of Developers Use Claude Code. Cursor Went Backwards.

TL;DR

JetBrains published the AI coding agent numbers from its Developer Ecosystem Survey 2026, the tenth edition, fielded May through July across more than 15,000 professional developers. Claude Code sits at 39% workplace adoption worldwide and 47% in the United States, up from 18% in January. GitHub Copilot, the tool that created this category, is at 21%, down from 29% a year earlier. Cursor fell from 18% to 12% even as its name recognition rose. And 90% of professional developers now use a coding agent at work at least weekly, 68% of them daily. The report got almost no attention when it landed, which is strange, because it is the clearest picture anyone has published of a market that reordered itself in seven months.

The standings

These are shares of professional developers who report using each tool for work, not seats sold and not revenue. One developer can appear in several bars, which is why the column adds to well over 100%.

The number under the number is stickiness. Of the 39% using Claude Code at work, 31% of all developers name it their single most used tool, which JetBrains puts at close to an 80% conversion from regular use to primary use. Most tools get tried. This one gets kept.

Seven months is not a long time

In January, on JetBrains' previous wave, Claude Code and Cursor were tied at 18% and Copilot led at 29%. Here is what happened next.

Codex is the other mover: 3% to 16%, roughly 5x, with awareness climbing from 27% to 65%. That awareness jump is the more telling half. In January, despite OpenAI's brand, fewer than three in ten developers had heard of its coding agent. The desktop app and the ChatGPT placement fixed that faster than the product could convert it.

Cursor got more famous and less used

Cursor's mindshare went up, from 69% to 75%, while its workplace adoption fell by a third. Those two lines are supposed to move together, and when they split it usually means the top of the funnel is fine and something downstream is leaking. The steepest drop was in China, from 28% in January to 16%.

Awareness is the gym membership. Adoption is showing up in February.

The one to watch at the bottom is OpenCode. No vendor, no ad budget, no bundling deal, and 42% of professional developers worldwide have heard of it, with 7% using it at work. That is more mindshare than Google's Antigravity, which has Google.

Copilot's problem is not awareness

Copilot still has 79% recognition globally, 86% to 90% across Europe, the UK and the US. It is pre-installed in the editor half the industry already pays for. It lost anyway, and it lost while being the best-distributed product in the category, which is the part worth sitting with if you are betting on bundling as a moat.

JetBrains adds a detail that explains some of the erosion: 39% of Copilot users run it inside JetBrains IDEs, among other surfaces. Copilot is not just competing with Claude Code, it is competing inside editors that now ship Claude Agent, Codex and OpenCode in the same chat panel, plus anything else you can bolt on through the Agent Client Protocol. When the incumbent becomes one entry in a dropdown, distribution stops being a moat and starts being a menu.

Why the ones that stick, stick

The survey says what developers use. A separate and genuinely independent dataset says why it spreads. In a July arXiv paper, Microsoft researchers Emerson Murphy-Hill, Jenna Butler and Alexandra Savelieva studied tens of thousands of Microsoft engineers through the company's early-2026 rollout of Claude Code and Copilot CLI. Their finding: first use spread primarily through social networks, retention tracked engineers' coding activity rather than their demographics, and adopters merged roughly 24% more pull requests than they otherwise would have, a lift that held across the four-month window.

Nobody adopted a CLI agent because of a landing page. They adopted it because the person next to them was visibly using one. That is the mechanism behind a 21-point swing in seven months, and it is also why mindshare alone did nothing for Cursor.

Read the fine print

• JetBrains is not a neutral referee. It sells the IDEs Copilot runs in, ships its own agents in Junie and JetBrains AI, and closes this very report with a pitch for Air and JetBrains Central. That does not make the numbers wrong. It does mean you should not treat the framing as disinterested.
• The comparison mixes two instruments. The May-July figures come from the Developer Ecosystem Survey; the January baseline comes from the AI Pulse survey, a different wave with a different sample. Directionally consistent, but not the same ruler twice.
• The sample is weighted on JetBrains familiarity. JetBrains says the Developer Ecosystem Survey is statistically reweighted by region, employment status, programming language and familiarity with JetBrains products. To its credit, the January AI Pulse wave was debranded, promoted through Instagram ads and Zhihu in China, made no mention of AI in the promo, and drew only about 16% of responses from the JetBrains research panel.
• Adoption is self-reported use, not spend. A developer who opened Claude Code twice last week counts the same as one who lives in it.

An independent check exists and it points the same way. The Pragmatic Engineer's survey of 906 respondents, fielded January 27 to February 17 and published in March, found Claude Code the most-loved tool at 46%, against 19% for Cursor and 9% for Copilot, with 95% of respondents using AI tools at least weekly. Different population, different question, same ordering.

Key Takeaways

• Claude Code is the default now, not the insurgent. 39% worldwide, 47% in the US, and 31% of all developers call it their primary tool, close to an 80% conversion from user to main user.
• Bundling did not save Copilot. 79% awareness, deep enterprise distribution, and adoption still fell from 29% to 21% in a year. Being everywhere is not the same as being chosen.
• Cursor is the warning case. Mindshare up from 69% to 75%, adoption down from 18% to 12%, and a China drop from 28% to 16%. When those two lines diverge, the problem is retention, not marketing.
• Agents are table stakes. 90% weekly, 68% daily. If your team's tooling debate is still whether to allow agents, the survey says you are arguing about last year.
• Peer visibility drives adoption. Microsoft's own rollout study found first use spread through social networks and adopters merged about 24% more PRs. Rollout strategy is a social problem before it is a licensing one.
• Watch OpenCode. 42% mindshare and 7% adoption with no vendor behind it, ahead of Google's Antigravity on awareness. Open-source agents are now a real slice of the board.

Sources: JetBrains Research, AI Coding Agents: Adoption Trends, JetBrains Research, Which AI Coding Tools Do Developers Actually Use at Work?, Murphy-Hill, Butler and Savelieva, Adoption and Impact of Command-Line AI Coding Agents (arXiv, 1 July 2026), The Pragmatic Engineer, AI Tooling for Software Engineers in 2026