TL;DR
JetBrains just published the second wave of their AI Pulse survey, covering 10,000+ professional developers across eight languages. The headline: 90% of developers now use AI tools at work, but the market is reshuffling fast. GitHub Copilot still leads at 29% adoption but its growth has flatlined. Claude Code jumped from 3% to 18% in eight months with the best satisfaction scores in the industry. And Google Antigravity already hit 6% adoption two months after launch.
The Survey That Actually Matters
Most “developer survey” articles are thinly-veiled marketing. JetBrains’ AI Pulse is different. They surveyed over 10,000 professional developers worldwide in January 2026, localized the survey into eight languages, promoted it without mentioning AI in the ads (to avoid self-selection bias), and applied raking weighting to align demographics with the broader developer population.
That methodology matters. When a company like Stack Overflow runs a survey, the sample skews toward developers who hang out on Stack Overflow. JetBrains went further to get a representative cross-section.
The previous wave ran in September 2025, so we have trend data. Four months of movement in a market that shifts quarterly.
90% of Developers Use AI Tools at Work
Nine out of ten professional developers now regularly use at least one AI tool for coding and development tasks. Regular, ongoing usage, not “tried it once and forgot about it.”
74% have adopted specialized AI dev tools like coding assistants, AI editors, and agents. The remaining 16% use general-purpose chatbots (ChatGPT, Gemini, Claude) for coding tasks without a dedicated integration.
About one in six developers who use AI for coding are still copy-pasting between a chat window and their editor. That’s a large addressable market for tools with tighter IDE integration, and it helps explain why every major player is racing to build deeper editor hooks.
The Adoption Scoreboard
The tool-by-tool breakdown from January 2026:
| Tool | Awareness | Work Adoption |
|---|---|---|
| GitHub Copilot | 76% | 29% |
| ChatGPT (chatbot) | — | 28% |
| Cursor | 69% | 18% |
| Claude Code | 57% | 18% |
| JetBrains AI Assistant | — | 9% |
| Gemini (chatbot) | — | 8% |
| Claude (chatbot) | — | 7% |
| Google Antigravity | — | 6% |
| Junie | — | 5% |
| OpenAI Codex | 27% | 3% |
A few things jump out.
ChatGPT at 28% is almost tied with Copilot. A general-purpose chatbot is nearly as widely used for coding as the most established coding-specific tool. That should make GitHub’s product team uncomfortable.
Cursor and Claude Code are tied at 18%. Six months ago, Cursor had a comfortable lead. Claude Code closed that gap completely. (For a hands-on comparison of both tools, see Cursor vs Claude Code vs Windsurf.)
Google Antigravity hit 6% in two months. It launched in November 2025. Getting to 6% adoption by January 2026 with a free preview is aggressive growth, faster than Cursor or Claude Code were growing at the same stage.
Copilot’s Ceiling Problem
GitHub Copilot is still the king by raw numbers: 4.7 million paid subscribers, deployed at roughly 90% of Fortune 100 companies, 76% developer awareness. On paper, everything looks dominant.
But the JetBrains data reveals something the subscriber counts don’t: Copilot’s growth has stalled.
Between September 2025 and January 2026, both awareness and adoption barely moved. Compare that to Claude Code, which went from 31% awareness and ~3% adoption in mid-2025 to 57% awareness and 18% adoption by January 2026. Copilot went from roughly 75% to 76% awareness in the same period.
Why the plateau? A few things stacked up in late 2025 and early 2026:
- Developers on Reddit and Hacker News documented measurable declines in Copilot suggestion quality since late 2025. Accuracy, latency, and context awareness all took hits.
- In March 2026, Copilot got caught injecting promotional “tips” into pull requests. The phrase appeared in over 1.5 million PRs before GitHub killed it. This came on top of the default data collection policy change that angered developers in late March.
- Microsoft 365 Copilot has only 3% penetration (~15 million seats out of 450 million). When employees have access to both Copilot and ChatGPT, 76% choose ChatGPT. The “Copilot” brand is getting stretched thin.
None of this means Copilot is dying. A 29% adoption floor with 4.7 million paying users is a massive business. But the growth engine has stalled, and that creates openings.
Claude Code’s Eight-Month Blitz
Look at the growth curve:
- April–June 2025: ~3% adoption, 31% awareness
- September 2025: ~12% adoption, 49% awareness
- January 2026: 18% adoption, 57% awareness
- US/Canada: 24% adoption (January 2026)
That’s a 6x increase in adoption over eight months. And in the US and Canada, Claude Code has already passed Cursor.
The satisfaction numbers are equally striking. Claude Code posted a 91% CSAT score and a net promoter score of 54, the highest of any AI coding tool in the survey. For context, an NPS above 50 is considered “excellent” in most SaaS categories. Most developer tools hover in the 20-40 range.
What’s driving it? A few things I’ve noticed from using Claude Code daily:
The terminal-native workflow eliminates context switching. You stay in your editor, run claude in the terminal, and the agent reads your codebase, makes changes, runs tests, and commits. There’s no second window, no separate IDE, no browser tab.
The model quality helps too. Claude Opus 4.6 consistently handles large codebases (50+ file changes, cross-cutting refactors, test generation across multiple frameworks) in ways that GPT-5.4 and Gemini 3.1 still struggle with.
And pricing played a role. At $120/month for the Max plan with extended thinking and high rate limits, Claude Code costs more than Cursor’s $20/month. But developers who switched report spending less overall because they aren’t burning through Cursor’s “fast request” limits and falling back to slow models mid-task.
Cursor’s Growth Plateau
Cursor crossed $2 billion in annualized revenue by February 2026. That’s a genuine business achievement for a startup that barely existed two years ago.
But the JetBrains data shows adoption growth has slowed. Cursor reached 18% and stayed there while Claude Code caught up from behind.
The pattern in the data suggests churn at the individual developer level. Cursor’s enterprise business (45-60% of revenue by late 2025) is holding steady, but individual developers, the early adopters who evangelized it, are the ones leaving. Reddit threads and Hacker News comments tell a consistent story: developers who tried Claude Code for a week didn’t come back.
Cursor also made a controversal bet by building Composer 2 on top of Moonshot’s Kimi K2.5, a Chinese foundation model. The technical choice makes sense (cheaper inference, custom fine-tuning), but it introduced geopolitical friction that some enterprise buyers flagged.
Cursor is an IDE. Claude Code is a CLI tool that works with any editor. As AI coding shifts from “autocomplete in your editor” to “autonomous agent that runs tasks,” the IDE-centric model may be the wrong abstraction layer. You don’t need a purpose-built editor when the agent operates at the file system level.
Google Antigravity: The One to Watch
Google Antigravity launched in November 2025, and by January 2026 it had already reached 6% work adoption. No other tool in the survey grew that fast from zero.
Antigravity is Google’s agent-first IDE. Instead of inline suggestions, you dispatch autonomous agents that plan, execute, and verify coding tasks. You can send five agents to work on five different bugs simultaneously. Each one produces “Artifacts” (task lists, implementation plans, screenshots, browser recordings) that you review like code review comments.
The free preview price (currently $0) obviously helps adoption. But the product has substance behind it: Antigravity scores 76.2% on SWE-bench Verified, which beats most competitors. And it supports multiple model backends: Gemini, Claude Opus 4.6, and an open-source OpenAI variant.
The question is whether Google can convert free preview users into paying customers. Google has a history of launching great developer tools and then abandoning them (see: Google Code, Allo, Stadia, Cloud Functions for Firebase pricing chaos). Developers remember.
What the Satisfaction Gap Reveals
The satisfaction gap between tools deserves more attention than it’s getting. Claude Code’s 91% CSAT and 54 NPS sit in a completely different tier than the competition.
JetBrains didn’t publish full CSAT/NPS data for every tool, but the Pragmatic Engineer’s separate survey found Claude Code had the highest “most loved” rating at 46%. Nearly half of the developers who use it consider it their favorite tool across every category.
High satisfaction creates a flywheel. Satisfied users recommend the tool to teammates. Teammates try it. The 24% adoption rate in the US and Canada (vs. 18% globally) suggests this word-of-mouth effect is strongest in English-speaking dev communities where Claude Code’s terminal-first UX resonates most.
What Comes Next
Three trends I’m watching based on this data:
The chatbot-to-tool pipeline is real. 28% of developers use ChatGPT for coding, and 7% use Claude’s chatbot. These are future customers for Codex, Claude Code, and Antigravity. The question is which tool captures them as they move from “paste code into chat” to “let the agent handle it.”
Enterprise is the real battleground. Individual developer preferences shift fast, but enterprise contracts lock in for years. Copilot’s 90% Fortune 100 deployment gives it a moat even as individual adoption stalls. Claude Code’s enterprise play through Amazon Bedrock and direct sales is growing but still early.
The fastest-growing tools (Claude Code, Antigravity) treat AI as an autonomous agent. The ones plateauing (Copilot, Cursor) started as inline suggestion engines and are retrofitting agentic features. That architectural difference is starting to show in the adoption numbers.
FAQ
Which AI coding tool has the highest adoption in 2026?
GitHub Copilot leads at 29% work adoption, followed closely by ChatGPT at 28% (used as a general chatbot for coding). Among dedicated AI coding tools, Copilot leads, with Cursor and Claude Code tied at 18%.
Is GitHub Copilot losing users?
Copilot’s growth in both awareness and adoption has plateaued between September 2025 and January 2026. The user base is stable, but competitors like Claude Code grew rapidly in the same window.
What’s the best AI coding tool by developer satisfaction?
Claude Code posted the highest satisfaction scores in the survey: 91% CSAT and an NPS of 54. The Pragmatic Engineer survey separately found Claude Code had the highest “most loved” rating at 46%.
Is Google Antigravity free?
Yes, during its public preview period. Google hasn’t announced pricing for when it exits preview. The free tier helped it reach 6% adoption in just two months.
How many developers use AI coding tools in 2026?
According to the JetBrains survey of 10,000+ developers, 90% regularly use at least one AI tool for coding. 74% use specialized AI dev tools (coding assistants, editors, agents), while 16% use general-purpose chatbots for coding tasks.
Bottom Line
Copilot still leads at 29%, but it’s leading with a stalled growth rate while Claude Code grows 6x in eight months and Google Antigravity sprints from zero to 6% in two months. Developers are spreading their usage across multiple tools, satisfaction matters more than brand recognition, and the next wave of coding tools looks like autonomous agents rather than autocomplete.
If you’re evaluating AI coding tools right now, the numbers point to Claude Code for satisfaction, Copilot for enterprise stability, and Antigravity if you want to bet on where Google goes next. The developers who use multiple AI tools report the highest productivity, so the right answer is probably “more than one.”