By The assumption that GitHub Copilot is the default AI coding tool is no longer accurate. JetBrains surveyed more than 15,000 professional developers between May and July 2026 and found Claude Code at 39% adoption — nearly double GitHub Copilot’s 21%. Six months earlier, in January 2026, those numbers were flipped: Copilot led at 29%, Claude Code sat at 18%. That reversal happened in roughly half a year.
The Numbers Behind the Shift
This is the tenth year JetBrains has run its Developer Ecosystem Survey, which gives the data historical credibility. The 15,000+ respondents span experience levels and regions, weighted by geography, employment, and primary programming language. It is not a random Twitter poll.
The headline figure is the adoption reversal, but the depth of it is what stands out. Claude Code is now the primary tool for 31% of developers — meaning most who try it at work end up making it their main tool. Copilot still has 79% awareness (nearly everyone knows it exists), but awareness has stopped converting to usage at the same rate. In the United States, Claude Code adoption hits 47%. Among senior engineers, 46% name it their most-loved coding tool versus 9% for Copilot — a five-to-one gap at the segment that typically drives tool adoption decisions.
Cursor, which entered 2026 at 18% alongside Claude Code, dropped to 12% over the same period. It still holds its ground among developers who prefer an IDE-first workflow, but the terminal-first crowd appears to be moving decisively toward Claude Code.
Why Claude Code, and Why Now
The explanation is partly category. Copilot was built for inline completions and suggestions — a co-pilot model. Claude Code is built for task delegation: you hand it a goal, it executes a sequence of file reads, edits, and shell commands to get there. That is a different product philosophy, and the survey data suggests many developers prefer the latter for serious work.
Small companies and startups show this most clearly: Claude Code adoption in that segment reaches 75%. Large enterprises still lean toward Copilot, largely because existing Microsoft licensing agreements and compliance requirements create structural inertia. Survey numbers do not capture contracts.
The Trust Problem Nobody Wants to Talk About
The more interesting tension in the data is not which tool is ahead — it is what the adoption numbers obscure. Developers report that 47% of their new code is now fully AI-generated. For JavaScript, TypeScript, and Go developers, that number climbs to 54–55%. The 31% of developers JetBrains identifies as “agentic coders” report that 84% of their code is agent-written.
Yet only 4% of developers say they completely trust AI-generated code to be functionally correct. Just 48% say they always verify AI output before committing. Code review time has reportedly increased 441% as teams try to cope with the volume of AI-generated changes landing in pull requests. Developers are producing more code than ever, trusting it less than ever, and struggling to review it fast enough.
When the survey asks what skills will matter most in the AI era, the top answer — cited by 47% of respondents — is “reviewing and validating AI-generated code for quality and security.” Not prompting. Not architecture. Not shipping faster. Reviewing.
GitHub Copilot Is Not Standing Still
Microsoft’s response has been visible. In May 2026, the company quietly cancelled internal Claude Code licenses and directed engineers to GitHub Copilot CLI — a move that got noticed precisely because it suggested Microsoft’s own teams found Claude Code valuable enough to be worth banning. At Build 2026, GitHub announced MAI-Code-1-Flash (its first model built specifically for Copilot), a revamped CLI, and a developer SDK. The framing shifted: Copilot is now positioned as an “agentic execution platform,” not a code assistant.
Whether that repositioning lands depends on whether enterprise contracts hold and whether Copilot’s agentic capabilities catch up. The survey is a snapshot of May through July 2026, before most of those announcements reached developers at scale.
What to Do With This
If you are still using GitHub Copilot by default and have not tried Claude Code, the JetBrains data is a reasonable prompt to experiment. That is not a dismissal of Copilot — enterprise constraints are real, and a VS Code integration that works is worth something. But if 46% of senior engineers have made the switch and are calling it their most-loved tool, there is probably something worth an afternoon of your time to evaluate.
The raw adoption comparison between Claude Code and Copilot tells you where the market went. The trust gap — 47% AI-generated code, 4% complete trust — tells you where the real problem is. One is a competitive story. The other is an industry-wide challenge that no survey result fixes.
