You open IntelliJ every morning, keybindings tuned like a vintage guitar, plugins stacked three layers deep — and every week some new AI editor demo shows up making you feel like you're coding with a stone axe. Cursor ships an agent. Codex ships a sandbox. Windsurf ships... something. The FOMO is real.
The conventional wisdom says switch. AI-native editors were born with AI in their DNA; legacy IDEs are playing catch-up. But the data from this month tells a different story — and the structural argument goes deeper than anyone on your timeline has bothered to explain.
The numbers say one thing, the architecture says another
On April 8, JetBrains published their AI Pulse survey — 10,000+ developers, real-world work adoption. GitHub Copilot leads at 29%. Cursor and Claude Code each sit at 18%. Junie — JetBrains' own coding agent — trails at 11%.
The same survey puts Claude Code's CSAT at 91% and its NPS (Net Promoter Score — how likely users are to recommend it) at 54. Highest on the market. Claude Code grew from 3% adoption in July 2025 to 18% by April 2026 — nine months of near-vertical growth. That's the kind of loyalty you can't absorb with a protocol.
Surface reading: JetBrains is losing the AI race. Actual reading: JetBrains isn't running it.
The move nobody framed correctly
Back on March 4, Cursor announced it now works inside JetBrains IDEs through ACP — Agent Client Protocol, essentially a USB port that lets AI agents plug into any editor. Aleksey Stukalov, Head of IDEs Division at JetBrains, said in the same announcement: "Developers stay in control of their environment, while Cursor brings the powerful AI assistance." Read that again. Cursor — supposedly JetBrains' biggest competitor — now runs as a plugin inside IntelliJ.
Five days later, on March 9, JetBrains launched Air — an agentic IDE built on Fleet's remains — and Junie CLI, an LLM-agnostic coding agent for $10–60/month.
Those March announcements were setup. The payoff landed this month.
On April 7, JetBrains connected Junie CLI to the IDE's semantic index — the AST-level understanding their IDEs spent 20+ years building. AST stands for Abstract Syntax Tree: a detailed map showing how every function, class, and variable connects to everything else. Junie inherited that entire map instantly.
Cursor and Windsurf? They infer your project structure from raw file contents. Every time. It's the difference between giving someone a GPS versus telling them "the coffee shop is somewhere northeast, probably."
IntelliJ 2026.1, released the same week, shipped with a built-in ACP Agent Registry — a one-click marketplace for browsing and installing AI agents the way you install plugins. Codex, Cursor, any ACP-compatible agent. All running in Git worktrees — isolated branches where agents work without touching your main code.
The price of being Switzerland
JetBrains' strategy has a name: platform-as-protocol. Let every agent in, own none of them. Swap AI models like batteries. The upside is obvious. So is the downside.
JetBrains owns no proprietary model. No accumulated context moat — the persistent memory of your codebase that Cursor builds with its @Codebase feature. No cloud sandbox like GitHub Copilot's cloud agent. And Air, as The Register noted on March 10, runs on Fleet's codebase — the editor that never found an audience. JetBrains themselves admitted in the Air launch blog: "Complex codebases aren't yet ready for pure agentic coding." Not exactly a rallying cry.
Claude Code's 91% satisfaction and 54 NPS — again from that same AI Pulse survey — say something uncomfortable for the socket thesis: developers don't just want interchangeable parts. They want the agent that gets them. A universal plug doesn't automatically beat a purpose-built tool that users love.
The desk won, not the sitter
Here's my take. If you're among JetBrains' 16 million developers, your FOMO bill just dropped to zero. As of today — April 12, 2026 — you can run Cursor's agent, install Junie CLI, browse an agent registry, and keep every keybinding you've had since 2014. All without leaving your IDE.
The switching cost argument that AI-native editors depended on just collapsed. Not because JetBrains built a better AI — they explicitly didn't. Because they built a better socket.
The editor war isn't about who has the smartest agent. It's about who has the desk the agents sit at. JetBrains bet that 16 million developers were already at the right one. Whether the agents sitting across from them are good enough — that's a question JetBrains deliberately chose to let someone else answer.
