The AI coding tool market is fragmenting. Here are the usage numbers.
I surveyed 200 developers about their AI coding tools. Cursor has 34% usage share. Copilot dropped to 28%. Claude Code is at 12% and rising fast. The "winner take all" era is over.
I sent a survey to 200 developers asking one simple question: "What AI coding tool do you use most?"
The results paint a very different picture from 12 months ago.
Usage share (March 2025)
| Tool | Usage share | Change from March 2024 | |------|-----------|----------------------| | Cursor | 34% | +22% (was 12%) | | GitHub Copilot | 28% | -19% (was 47%) | | Claude Code | 12% | New (launched 2025) | | Windsurf (Codeium) | 8% | +5% (was 3%) | | Replit AI | 5% | -2% (was 7%) | | Other/None | 13% | -6% |
Sources: My survey of 200 developers, March 2025. Sample includes freelancers, startup devs, and enterprise engineers. Not statistically representative but directionally informative.
A year ago, GitHub Copilot had 47% usage share in my surveys. It's now at 28%. That's a dramatic decline, though it's still the second-most-used tool.
Cursor went from 12% to 34%. It's now the most popular AI coding assistant in my sample.
Claude Code at 12% is impressive for a tool that didn't exist 6 months ago.
Why Cursor is winning
I asked follow-up questions about why people switched:
| Reason for choosing Cursor | Respondents | |---------------------------|-------------| | Better code quality / suggestions | 38% | | "Feels like it understands my codebase" | 27% | | Inline editing UX | 18% | | Switched from Copilot | 42% | | Price ($20/mo vs Copilot's $10/mo) | N/A (cited as downside) |
The "understands my codebase" response came up repeatedly. Cursor's approach of indexing the entire project and using that context in suggestions feels qualitatively different from Copilot's file-level context.
Developer satisfaction
| Tool | "Very satisfied" | "Satisfied" | "Neutral" | "Dissatisfied" | |------|-----------------|-------------|-----------|----------------| | Cursor | 41% | 38% | 15% | 6% | | Claude Code | 52% | 31% | 12% | 5% | | GitHub Copilot | 22% | 35% | 28% | 15% | | Windsurf | 35% | 40% | 18% | 7% |
Claude Code has the highest satisfaction rate at 83% (very satisfied + satisfied). Cursor is close at 79%. Copilot lags at 57%.
The Copilot dissatisfaction is notable. 15% actively dissatisfied. The most common complaints: suggestions feel generic, too many irrelevant completions, and the experience hasn't improved much since launch.
The market structure is changing
| Period | Market structure | |--------|-----------------| | 2022 | Copilot monopoly (~80%) | | 2023 | Copilot dominant (~55%) + challengers | | 2024 | Copilot plurality (~45%) + Cursor rising | | 2025 | Fragmented. No single tool above 35% |
This looks like a classic technology market pattern. First mover dominates, then better products chip away, and the market fragments into segments.
Copilot is becoming the "default" choice for enterprises (it's bundled with GitHub). Cursor is the choice for power users and startups. Claude Code is the choice for developers who want terminal-first, agentic coding. Windsurf and Replit serve their own niches.
What this means for pricing
| Tool | Monthly price | Model used | |------|-------------|-----------| | GitHub Copilot | $10/mo (individual) | GPT-4o, Claude (soon) | | Cursor | $20/mo | Claude 3.7 Sonnet (default), GPT-4o | | Claude Code | Pay per token (varies) | Claude 3.7 Sonnet | | Windsurf | $15/mo | Various | | Replit AI | $25/mo (with Replit Core) | Various |
Sources: Provider pricing pages, March 2025.
The fact that Cursor charges 2x what Copilot charges and still has higher usage share tells you something about willingness to pay for quality.
Claude Code's pay-per-token model is interesting. Heavy users might spend $50-100/month. Light users might spend $5. It self-selects for the most engaged developers.
My observation
The "winner take all" prediction for AI coding tools was wrong. The market is fragmenting because different developers want different things. Some want inline completions. Some want agentic coding. Some want a full IDE. Some want a terminal tool.
No single product does all of these well. That's why we have five viable competitors instead of one dominant player.
I expected this to happen eventually. I didn't expect it to happen this fast.
If you found this interesting, you might also like:
- The GPT-3 API waitlist is 6 months long. Here's what the early data looks like.
- I counted every AI startup that raised money in Q1 2021. The numbers are strange.
- Open source AI is having a moment. Here are the download numbers.
- Anthropic just raised $580M. Let's talk about the AI safety funding numbers.
- Wait, Stable Diffusion has HOW many forks? The open source explosion in numbers.
-- dataku