Choosing an AI coding tool used to be like picking a religion — Cursor stans and Copilot loyalists were sworn enemies, Windsurf users occupied their own lane, and Claude Code purists thought everyone else was insufficiently "native." But after JetBrains shipped version 2026.1 in July 2026, that question may no longer matter.

The new release introduced an unassuming but far-reaching feature: ACP Registry. ACP stands for Agent Client Protocol — an open JSON-RPC 2.0 protocol in the spirit of LSP (Language Server Protocol), designed to let AI agents from different vendors collaborate inside the same IDE. The official repo, agentclientprotocol/agent-client-protocol, has already racked up 3.2k+ stars.

Assembling Agent Teams in One IDE

ACP doesn't solve "which model is strongest" — it solves something more fundamental: standardized communication between editors and coding agents.

Before ACP, every AI coding tool was an island. Cursor's agent only ran inside Cursor. GitHub Copilot only worked in VS Code and Visual Studio. Claude Code was a standalone CLI. Junie was JetBrains' in-house tool. Want Cursor for frontend, Copilot for reviews, and Claude Code for architecture refactoring? Sorry — juggle three windows and copy-paste context between them.

ACP changes that. The protocol defines standard interactions between agents and IDE clients: task dispatch, context passing, file operations, terminal execution, and diff previews. Theoretically, any ACP-compatible agent can plug into any ACP-compatible IDE. Inside IntelliJ IDEA, you could have Cursor's agent handle frontend components, Copilot handle code review, and Claude Code handle backend refactoring — three agents working for you simultaneously, in the same editor.

It sounds like a miniature dev team. You used to hire humans to form teams; now you orchestrate agents to form teams. Frontend agent, backend agent, testing agent, review agent — each from a different vendor, each with its own specialty, all collaborating in the same workspace via ACP.

Moats Dissolve, a Skill Market Emerges

ACP's deeper implication is that AI coding tool moats are dissolving on a weekly basis.

AI coding tools historically had three layers of defense: model capability, IDE integration depth, and user habit. MCP (Model Context Protocol) already broke down the tool-calling layer, letting agents use any MCP-compatible tool. ACP now breaks down the IDE integration layer, decoupling agents from specific editors. When integration barriers vanish, competition shifts from "I'm a complete tool" to "I'm a skill that excels at X."

That's why some observers say ACP is gestating an "AI coding skill marketplace." Just as the VS Code extension marketplace hosts thousands of themes, language supports, and linters, an agent skill marketplace could emerge — some agents specialize in React performance optimization, others in SQL tuning, security auditing, or test writing. You orchestrate on demand, paying per token or per task.

JetBrains embracing an open protocol at this juncture is a fascinating strategic pivot. As the longtime IDE king, JetBrains一度 fell behind Cursor and Copilot in the AI coding race — but Junie topping SWE-rebench combined with embracing the ACP open ecosystem signals JetBrains has found its position: not building an agent, but building the platform for agents. It's the same play Microsoft made with LSP in VS Code — if you can't beat 'em, become the ecosystem.

For developers, this is the best of times. You no longer need to go all-in on one tool; you can assemble a team of agents like a sports roster, picking the right player for each task. The bad news: decision fatigue is real, and you'll need to learn each agent's strengths and weaknesses. For AI coding vendors, the moat shifts from "my IDE experience is better" to "my agent is SOTA at task X" — a more brutal but fairer arena.

The IDE war has no winners, only survivors. When protocol standardization completes, the endgame for AI coding may not be one tool to rule them all, but a vibrant marketplace of agent skills.

See you tomorrow.