Eight months. That's how long it took Claude Code to go from public launch to leading the category on every metric that matters — satisfaction, loyalty, and daily active use among the developers who actually ship production software.
GitHub Copilot had years. A Microsoft distribution deal. Native IDE integration. 76% brand awareness among developers. And it's still losing on the metrics that predict long-term retention: CSAT and NPS.
The Numbers Are Not Close
The JetBrains Developer Ecosystem Survey for 2026 puts Claude Code at 91% CSAT and an NPS of 54. Those are the highest loyalty scores in the category. Not by a little.
GitHub Copilot is still the most widely known tool — 76% of developers have heard of it, 29% use it at work. But knowing something and loving it are different things. Copilot's satisfaction scores have been sliding as power users migrate to tools built around agentic workflows.
By the numbers (JetBrains Research, 2026)
Claude Code: 91% CSAT, NPS 54 — highest in category. GitHub Copilot: 76% brand awareness, 29% workplace adoption, satisfaction declining. Cursor: briefly led the agentic wave, now third in satisfaction rankings.
What Actually Changed: Copilot vs. Agent
The name 'Copilot' was always a metaphor — and it turned out to be the wrong one. A copilot sits next to you and helps. An agent goes and does the thing.
For years, AI coding tools were glorified autocomplete. Useful, sure. But you were still the one holding every context switch, every terminal command, every file edit. The AI was a fast typist riding shotgun.
Claude Code shipped with a different assumption: you describe what you want, it figures out how to do it. Terminal-first. Agentic by default. No IDE plugin required, no hand-holding needed. It treats you like an engineer, not a user.
“The shift wasn't about better autocomplete. It was about who's driving.”
55% of developers now regularly use AI agents — and among staff-level and above engineers, that number is 63.5%. These are the people whose opinions propagate through teams. When they switch, teams switch.
Cursor's Brief Window
Cursor deserves credit. It saw the agentic shift coming and built for it inside a familiar IDE wrapper. For a lot of developers, Cursor was the first taste of what autonomous coding assistance could actually feel like.
But Cursor was still fundamentally IDE-shaped. Claude Code didn't try to be a better IDE. It ran in the terminal, talked to your filesystem, spawned subprocesses, read your git history. It acted more like a senior engineer pair-programming over SSH than a plugin sitting inside VS Code.
The Copilot Token Billing Moment
On June 1, 2026, GitHub flipped Copilot to usage-based token billing. The reaction was immediate and ugly. Power users — the exact developers driving Claude Code adoption — did the math and didn't like it.
When you're running complex, multi-step agentic tasks, token costs are not predictable. Flat-rate pricing is a feature, not a pricing strategy. Teams paying $100–200 per engineer per month for Claude Code Max knew exactly what they were getting each month. Token billing introduced the kind of uncertainty that finance teams and engineering leads hate equally.
“'What a joke' was the top comment on Hacker News the day Copilot's new billing dropped. That's not a recoverable moment.”
The 95% Baseline
95% of developers now use AI coding tools at least weekly. This isn't a category that's still trying to prove value. It's a category where the question is purely which tool wins.
- 95% of devs use AI coding tools at least weekly
- 55% regularly use AI agents (not just autocomplete)
- 63.5% of staff+ engineers are on agentic workflows
- $100–200/month per engineer is the new normal for 'max' tier plans
GitHub Copilot optimized for the developer who was skeptical of AI. Claude Code optimized for the developer who was ready to hand off the boring parts entirely. In 2025 that was a risky bet. In 2026 it's the obvious call.