JetBrains Shuts Down Fleet, Bets Everything on Agentic AI with Air

After four years stuck in "public preview" purgatory, JetBrains has finally made the call on Fleet. Starting December 22, 2025, their VS Code competitor is done. No more downloads, no more updates. If you've got it installed, you can keep using it until the server-dependent features inevitably break—but that's the end of the road.

The JetBrains Fleet shutdown isn't just a product retirement though. JetBrains is replacing it with something fundamentally different: Air, an "Agentic Development Environment" where you don't write code at all. You delegate tasks to AI agents and review what they produce.

I used Fleet occasionally over the past couple of years—mostly for markdown editing and quick file browsing. It always felt like a product searching for a purpose. Turns out I wasn't wrong.

Why Fleet Never Found Its Market

Fleet launched in November 2021 with a compelling pitch: a modern, lightweight IDE built from scratch, free from the quarter-century of architectural decisions baked into IntelliJ. JetBrains felt the pressure from VS Code's meteoric rise and wanted their own answer.

The problem was that Fleet couldn't decide what it wanted to be. First it was a "lightweight multi-language IDE." Then a "smart editor with coding assistance." Then, when the AI wave hit, they explored making it an "AI-first editor." Each pivot left users confused about whether Fleet was worth investing time in.

JetBrains themselves put it bluntly in their official announcement: "We could neither replace IntelliJ IDEA with Fleet nor narrow it into a clear, differentiated niche. Two similar products created friction and raised questions about ownership and long-term investment."

Here's the thing that always puzzled me: if you're already using WebStorm, PyCharm, or GoLand, why would you switch? Fleet didn't have the deep language intelligence of the IntelliJ-based IDEs. It didn't have VS Code's extension ecosystem. And despite promising to be lightweight, one developer complained that "doing nothing takes up 1GB of RAM."

The warning signs were there. Earlier in 2025, JetBrains deprecated Kotlin Multiplatform support in Fleet, shifting focus back to IntelliJ. Community members connected the dots immediately—and they were right.

What Air Actually Is

Rather than simply killing Fleet, JetBrains is transforming its underlying platform into something radically different. Air isn't an IDE in any traditional sense. It's designed around a single premise: you stop writing code and start supervising AI agents that do it for you.

The workflow looks like this: you give an agent a task—"add unit tests for the authentication module" or "refactor this class to use dependency injection"—and the agent works on it asynchronously. Meanwhile, you can spin up other agents working on different tasks in parallel. When they're done, you review their changes in a dedicated diff panel, provide feedback, and commit what works.

Air offers three isolation modes for agent execution. Git Worktree creates a separate branch while using your local environment. Docker provides complete container isolation—all changes stay sandboxed. Local Workspace is fastest but changes hit your project directly.

Currently, Air works exclusively with Anthropic's Claude, which means you'll need your own Anthropic subscription on top of whatever JetBrains charges. Future plans include support for OpenAI Codex, Google Gemini, and JetBrains' own Junie agent. The product is macOS-only during this preview, with Windows and Linux planned for 2026.

The review-first approach is interesting. You can comment directly in the diff panel, similar to a GitHub code review, and your feedback gets incorporated into future prompts. Local history snapshots are created before each interaction, so you can roll back experimental changes easily.

The Single-Agent Problem: Does Anyone Need This?

Here's where I'm genuinely skeptical. The developer community reaction has been divided, and the core criticism cuts deep.

One commenter on Hacker News put it this way: "Even a single agent is faster at generating code than I am at evaluating its fitness and providing design guidance. Adding more agents working on different things would not speed anything up."

This matches my experience. When I use Claude Code in the terminal, the bottleneck is never "waiting for code generation." It's reviewing what the agent produced, understanding the implications, and deciding whether to accept it. Running three agents in parallel doesn't solve that problem—it multiplies it.

Thorsten Ball, author of Writing an Interpreter in Go and lead engineer at Amp, apparently said something similar in a podcast: he's a single-tasker, despite some of his coworkers preferring fleets of agents. I'm in the same camp.

That said, I can see scenarios where Air makes sense. If you're delegating tasks that take hours—major refactors, comprehensive test suites—and you're comfortable context-switching between review sessions, parallel agents could work. But that's a specific workflow, not a universal improvement.

The more interesting play might be JetBrains' other bet: the Agent Client Protocol (ACP). They're partnering with Zed to create a standard for integrating any AI agent into any IDE. This means your existing IntelliJ, PyCharm, or WebStorm setup could work with Claude Code, Codex, or any ACP-compatible agent. That's the pragmatic path for most developers.

What Fleet Users Should Do Now

The timeline is tight. December 22, 2025 is the cutoff—after that, no more downloads from Toolbox App or any official channel. If you've already got Fleet installed, it'll keep running, but server-dependent features (including AI Assistant) will gradually break.

There's no migration path to Air. They're completely different products targeting completely different workflows. If you were using Fleet as your daily driver, here are your options.

For JetBrains loyalists, IntelliJ-based IDEs remain the safest bet. Many Fleet UX innovations have been backported, and AI features are being actively developed. If you're a Go developer, GoLand isn't going anywhere. Same for WebStorm, PyCharm, and the rest.

VS Code with Cursor or GitHub Copilot is where most of the AI-editor ecosystem lives right now. The extension library is unmatched, and if you're already comfortable with VS Code, the switching cost is minimal.

Zed is worth considering if performance is your priority. It's gaining traction among developers frustrated with Electron-based editors, and the ACP partnership with JetBrains means agent integration is coming.

Claude Code from the terminal is what I use for agentic workflows. It's not as polished as an IDE integration, but for delegating complex multi-step tasks, it works.

A Pattern That Should Worry You

Fleet joins a growing list of recently discontinued JetBrains products: Space, Aqua, Writerside—all shuttered in the past couple of years. This matters because Air faces the same adoption challenge Fleet did. Getting developers to change their development environment is notoriously difficult. JetBrains learned this the hard way with Fleet. There's no guarantee Air will be different.

As DevClass pointed out, "Getting developers to switch to Air may be no easier than it was to tempt them towards Fleet."

JetBrains isn't a company in trouble—they generate over $800 million in annual revenue from their traditional IDEs. But their track record with new product bets should make you cautious about going all-in on Air.

The Real Takeaway

If you're asking whether to invest time learning Air right now, my answer is: probably not, unless you're already doing substantial multi-agent workflows and frustrated with existing tools.

The more pragmatic play is watching the Agent Client Protocol. If ACP matures and gets broad agent support, you could stick with your current JetBrains IDE and bring in whichever AI agents work best for your stack. That gives you the semantic intelligence of IntelliJ-based tools—refactoring, debugging, framework awareness—plus agent delegation for the tasks that benefit from it.

Air is an interesting experiment in a direction development might go. But "might" is doing a lot of work in that sentence. For now, I'll keep using Claude Code in the terminal and my regular IDE for everything else. If hybrid human-agent teams become the default way we build software, I'll reconsider. Until then, I'm not betting my workflow on another JetBrains preview product.

---

December 22 is coming fast. If you want to download Fleet before it disappears, now's the time—though I'm not sure why you would.