By JetBrains spent 26 years teaching developers how to write code. Today, the company shipped Air — a system designed to hand much of that writing to AI agents while keeping developers in charge of what actually ships. This is not an IDE feature update. It is a new category of tool, and JetBrains is calling it an Agentic Development Environment.
Three Products, One System
Air arrives as three connected products. Air in JetBrains IDEs lets developers direct and orchestrate multiple AI agents — Claude, OpenAI Codex, Google Gemini CLI, and JetBrains’ own Junie — concurrently inside IntelliJ IDEA, PyCharm, WebStorm, and the rest of the JetBrains lineup. Agents run in isolated environments (Local, Git Worktree, or Docker) and tap into JetBrains’ deep code intelligence for type information, dependency graphs, and project structure.
Air Teams moves that agent activity up to the team level, coordinating workflows across developers rather than within a single session. Air Governance — formerly JetBrains Central — gives organizations policy enforcement, cost controls, per-team budgets, and audit trails. If you have ever tried to understand how much your team is actually spending on AI API calls, the governance layer is what makes that answerable.
The ACP Standard Is the Real Story
Most of the Air announcement will be read as “JetBrains enters the agent IDE market.” That is underselling it. The strategically significant move is the Agent Client Protocol (ACP), an open JSON-RPC 2.0 standard that JetBrains co-built with Zed.
If you have worked in editor tooling long enough to remember 2016, the comparison writes itself: LSP (Language Server Protocol) standardized how code editors talk to language servers, and suddenly every editor could support every language without bespoke plugins. ACP is attempting the same thing for the IDE-to-agent connection. Any ACP-compatible agent can plug into Air — or any ACP-supporting editor — without vendor-specific adapters. Twenty-five agents are already compatible, including Claude, Codex, and Gemini CLI.
This is JetBrains playing infrastructure politics, not just product competition. If ACP becomes the dominant standard — and with Google, GitHub, and Zed already on board, that trajectory looks plausible — JetBrains has a seat at the table regardless of which AI model ends up winning the capability race. It is a hedge dressed as a feature.
Junie CLI and the On-Device Option
Alongside Air, JetBrains shipped Junie CLI — a standalone version of its coding agent that runs from the terminal, inside any IDE, in CI/CD pipelines, and on GitHub or GitLab. Junie CLI is model-agnostic, supporting OpenAI, Anthropic, Google, and Grok, with full BYOK and OpenRouter support.
The more interesting announcement for privacy-conscious developers is Junie Local: a coding agent that runs entirely on your Mac, with no tokens consumed, no data leaving the machine, and no cost. On-device AI agents are not yet competitive with cloud models for complex tasks, but the option exists — and for organizations with data-handling constraints, it fills a real gap.
Who Can Start Today
Air is free to use if you bring your own API keys. If you already pay for Claude API access or an OpenAI subscription, the incremental cost to try Air is zero. JetBrains AI Pro subscribers are also covered. Paid tiers start at $30 per month for individuals and $20–60 per month for business seats.
The caveat: Air currently runs on macOS and Linux (Linux support shipped in June). Windows support is in development but has no announced release date. If you are a Windows-primary developer, you are watching from the sidelines for now.
The Bet
The skeptical read on Air is that multi-agent orchestration adds coordination overhead for problems developers mostly solve fine with a single agent in Cursor or Claude Code. That criticism has teeth. But JetBrains’ underlying argument is harder to dismiss: AI agents that understand your actual project — its types, its architecture, its test structure — produce more reliable output than agents reconstructing context from scratch on every task. Twenty-six years of building that project intelligence is not nothing.
Whether Air becomes the platform for agentic development or just one of many tools in a crowded market will depend on whether ACP achieves LSP-level adoption. Read the full announcement and watch the ACP registry growth over the next six months. That is the real metric.
