JetBrains Air: the IDE bets it still matters
The last download of JetBrains Fleet was served on December 22, 2025. The editor had been announced in November 2021 — a next-generation lightweight tool that could become a full IDE on demand, the company’s long-awaited answer to Visual Studio Code. It never escaped public preview. By December, JetBrains had concluded that the experiment had neither replaced IntelliJ IDEA nor found a differentiated niche, and that maintaining two overlapping IDE families was costing the company more than it was worth.
Nine months after Fleet’s last download, JetBrains unveiled its replacement. On September 22, 2026, the company announced JetBrains Air and called it “one of the most significant steps in our 26-year history.” Fleet’s team had not been disbanded — it had been redirected. Air was built on Fleet’s distributed platform, repurposing the architecture that had powered Fleet’s real-time collaboration and remote-development features for a different problem entirely.
Air arrived as a three-layer system. Air in IDEs lets developers direct agents and verify their output inside JetBrains’ familiar tools. Air Teams coordinates agent-driven delivery workflows across engineering organizations — running agents, accumulating results, managing handoffs. Air Governance, repurposed from JetBrains Central, handles policy, cost tracking, and auditability. A standardized Agent Client Protocol (ACP) connected the whole stack, letting JetBrains’ own Junie agent coexist alongside Claude Code, Codex, and Gemini CLI without any of them caring which IDE they were inside.
The thinking behind Air came from Kirill Skrygan, who wrote the announcement: “AI can produce code. Organizations still have to produce software.” The distinction was not subtle. In 2026, generating code was fast and cheap. Shipping that code — through review, testing, security audit, staging, and production — to a standard a developer was willing to own under their name, remained as slow as it had ever been. JetBrains was betting the tool developers used when making that judgment would still look like an IDE.
JetBrains had spent four years building a better place to write code, at the exact moment the world stopped needing one. Fleet’s public preview launched October 2022; within a year Cursor was demonstrating that the interesting problem in code editing was no longer editing. Fleet was a faster horse; the field had already moved to different transport. Air was the admission that the question had changed — not “where should I write this code?” but “should I ship the code that arrived already written?”
Skrygan named the new bottleneck precisely: as agents became capable of generating thousand-line refactors, the limiting factor was no longer production speed. It was comprehension — knowing whether the change was correct, whether it introduced a regression, whether it was the thing the ticket actually asked for. JetBrains’ 26 years of deterministic code intelligence — static analysis, call-graph traversal, type-aware refactoring — had found a second purpose. None of that infrastructure was built to help agents. All of it turned out to be useful for humans deciding whether to trust them.
ed gave programmers a way to write programs. IntelliJ gave them a way to understand what they had written. Air was a bet that the last job left — the one the agents couldn’t do — was deciding whether to ship it.
Sources
- Introducing JetBrains Air — The JetBrains Blog — September 22, 2026 announcement, Air’s three components, Agent Client Protocol, Kirill Skrygan’s philosophy on accountability vs. code generation.
- JetBrains Fleet — Grokipedia — Fleet’s full history: November 2021 announcement, October 2022 public preview, reasons for failure, December 2025 discontinuation, and connection to Air’s development.
- JetBrains abandons Fleet IDE, pins hopes on forthcoming Air agentic development tool — devclass — December 8, 2025 discontinuation announcement, JetBrains’ stated reasons for killing Fleet, Air as Fleet’s successor built on Fleet’s platform.