In December 2025, JetBrains quietly killed Fleet, its boldest attempt in a quarter-century to reinvent the developer IDE. Three months later, they launched Air , and what felt like an embarrassing retreat now looks like the most strategically timed pivot in developer tooling history.
What Actually Happened
JetBrains launched Air in public preview on March 10, 2026, built on the codebase of the otherwise abandoned Fleet IDE. Fleet had been introduced years earlier as a potential successor to IntelliJ IDEA , JetBrains' flagship Java and Kotlin IDE that also forms the foundation for Android Studio and dozens of specialized tools. Fleet never left preview and was officially discontinued in December 2025. Within three months, its codebase was repurposed for something entirely different: an agentic development environment that doesn't just assist developers, but delegates entire tasks to AI agents running concurrently in isolated environments.
Air's central concept is the "task" , a unit of work described by the developer in natural language and executed by an AI agent in a Git worktree, local workspace, Docker container, or (in upcoming releases) a cloud container. The key architectural choice is multi-agent parallelism: Air supports OpenAI Codex, Anthropic Claude Agent, Google Gemini CLI, and JetBrains' own Junie agent, all connectable through the Agent Client Protocol (ACP) , a vendor-neutral protocol co-developed by JetBrains and Zed that allows any compliant agent to plug in. Air launched on macOS in March 2026, with Windows and Linux support promised for subsequent releases. Two weeks later, on March 24, JetBrains announced Central , a platform designed to govern, orchestrate, and monitor those same agents working across entire enterprise development teams.
Why This Matters More Than People Think
The conventional reaction to Fleet's death was to write JetBrains off as a company that missed the AI era. But that framing misses something important: JetBrains has shipped developer infrastructure used by tens of millions of developers for 26 years. IntelliJ IDEA's code intelligence engine powers not just PyCharm, WebStorm, and RubyMine, but also Google's Android Studio. JetBrains knows how developers actually work, in production, at scale , and they know where the friction is. Air is not a chatbot bolted onto a text editor. It's an attempt to redesign the developer workspace around agent orchestration as the primary mode of interaction.
The timing is also critical. A January 2026 JetBrains survey of 11,000 developers found that 90% already use AI at work. But GitHub Copilot still leads the market , 76% brand awareness, 29% active usage , while JetBrains AI Assistant sits at just 9% and Junie at 5%. Air and Central represent JetBrains' answer to a simple problem: they are the deepest platform player in the IDE space, with 26 years of tooling knowledge, and they are being outpaced in the AI race by products a fraction of their age. That gap is the opportunity Air is designed to close.
The Competitive Landscape
The developer tools AI market in early 2026 looks like this: GitHub Copilot is the safe enterprise default; Cursor , recently valued at $50 billion after a $2 billion funding round in April 2026 , is the performance-obsessed upstart attracting individual developers and startups; Google's Antigravity platform (built on Gemini 3 Pro) hit 76.2% on SWE-bench Verified and is positioning as a Cursor alternative; and Cursor 3.0 shipped in April 2026 with an "Agents Window" for multi-session orchestration. Into this market, JetBrains is launching not one product but two: Air (the agent-native IDE) and Central (the team-level orchestration layer).
The crucial differentiator JetBrains is betting on is depth. Cursor started as a fork of VS Code and was always designed to augment a code editor. Air was designed from the ground up as an agent execution environment, with tasks as first-class primitives. JetBrains Central introduces something none of the Cursor-tier competitors have articulated clearly: governance infrastructure for enterprises that need to audit, monitor, and control what their AI agents are doing. JetBrains retired "Code With Me," its pair programming feature, to redirect resources toward Central , a signal that they are treating agentic orchestration as a replacement for synchronous human collaboration, not merely an enhancement to it.
Hidden Insight: The Real Value of 26 Years of IDE Data
Every JetBrains IDE , IntelliJ IDEA, PyCharm, WebStorm, GoLand, Rider, CLion , has been collecting structured information about how developers navigate, refactor, debug, and review code for over two decades. This is not chat logs or GitHub stars. It's fine-grained behavioral data: which refactors get reverted, which code patterns trigger bugs, which test failures correlate with specific architectural decisions. This data is the raw material for training coding agents that understand not just what code is correct, but what code is maintainable and operationally safe.
When JetBrains introduced Junie , its own agent that costs $10/month for individuals and $60/month for enterprise , the goal was less about competing on model quality and more about building an agent that leverages this proprietary behavioral corpus. Junie's ACP-native integration with Air means it can operate with IDE-level context: not just the file you're editing, but the call graph, the test coverage map, the dependency tree, and the full project structure. That's a fundamentally different surface area than a floating chatbot operating on clipboard context.
The uncomfortable implication is that JetBrains' competitive moat isn't really about Air or Central as products , it's about being the company that has watched millions of developers work for 26 years and built the tooling that encodes that institutional knowledge. As agentic development matures, the agents that win won't just be the ones with the largest context windows. They'll be the ones with the best priors about what good software development actually looks like , and JetBrains has more of that data than almost anyone else in this race.
What to Watch Next
The first leading indicator to track is JetBrains AI market share in the next quarterly developer survey. If Air drives JetBrains AI Assistant and Junie usage from the current 9%/5% levels toward the 20%+ range by Q4 2026, it signals the agent-first approach is pulling developers away from Copilot. Watch for JetBrains to release its next survey data around July 2026. A second critical metric: whether JetBrains Central exits its Early Access Program before GitHub Copilot Workspace , Microsoft's own attempt at team-level AI orchestration , moves from beta to enterprise GA. The race to be the enterprise standard for agentic development team management is genuinely open, and the first to GA in enterprise will set the reference architecture.
Longer-term, watch whether ACP , the Agent Client Protocol co-sponsored by JetBrains and Zed , gains adoption the way LSP (Language Server Protocol) became the universal bridge between code intelligence and editors. If Microsoft, Google, or Anthropic announce ACP support in the next 90 days, that's the signal JetBrains is on the path from IDE vendor to infrastructure layer , a positioning shift that would entirely reframe how investors and developers think about what this company is worth.
JetBrains didn't fail at building an IDE , it failed at building the IDE for the old era, and succeeded at recognizing that the new era needs something else entirely.
Key Takeaways
- Air launched March 2026 , built on Fleet's abandoned codebase, Air executes developer-described tasks via AI agents in Git worktrees, Docker containers, or local workspaces
- Central launched March 24, 2026 , a governance and orchestration layer for enterprise AI development teams, retiring Code With Me and introducing team-level agent management
- Only 11% of developers use JetBrains AI tools vs. 29% using GitHub Copilot, per a January 2026 JetBrains survey of 11,000 developers worldwide
- Junie costs $10 $60/month and connects to Air via ACP, a vendor-neutral protocol supporting OpenAI Codex, Anthropic Claude Agent, and Google Gemini CLI in a unified interface
- 26 years of behavioral IDE data from millions of IntelliJ IDEA, PyCharm, and WebStorm users gives JetBrains a domain-grounded training advantage that pure AI companies cannot replicate
Questions Worth Asking
- If agents write the majority of code, does it matter which IDE the human uses , or does the winner become the governance layer that controls the agents?
- GitHub Copilot has 29% market share and Microsoft's distribution. Can any IDE player realistically displace it in enterprise without a non-Microsoft channel?
- If ACP becomes the agent protocol standard the way LSP became the code intelligence standard, how does that change JetBrains' valuation , and should developers treat it as infrastructure, not tooling?