JetBrains Air

Run Codex, Claude Agents, Gemini CLI, and Junie side by side

Software Engineering

Developer Tools

Artificial Intelligence

Really cool to see JetBrains tackling the multi-agent workflow problem. As a Mac developer, I've been running Claude Code heavily for my projects and the biggest friction is exactly what you described — context switching between agents and terminals. Being able to define tasks precisely and run them in isolation is key. Curious about the review experience: does Air give you a unified diff view across all agents, or do you review each agent's output separately? Also, any plans for Linux support timeline? Love that it's free — will definitely give it a spin.

How does Air handle agent conflicts or differing outputs when running multiple agents like Codex vs. Claude on the same codebase task? Any built-in diff tools or voting mechanisms to streamline reviews?

Looks great so far. Would be cool if you could add a live comparison view between agents in future updates.

Having multiple AI agents in one coherent workflow is the right direction. I've been using Claude Code as my primary dev tool and the biggest friction is switching contexts when you need a different model's strengths. Being able to run them side by side and compare outputs on the same codebase solves a real problem. Curious how it handles conflicts when two agents suggest different approaches to the same task.

What specific “agent friction” did you see with CLI/TUI tools like Claude Code, Gemini CLI, and Codex that convinced you a dedicated workspace was necessary—and what concrete signals (time-to-PR, defect rate, review speed, etc.) did you use to validate that this pain is real in day-to-day engineering?

This is exactly the workflow we've been doing manually.

We're 3 people building Vidtreo — a video recording API (also launching on PH today). Our daily reality is running Claude agents across three repos simultaneously: backend on Cloudflare Workers, browser SDK with WebCodecs transcoding, and a React dashboard. The context switching between terminals and agents is brutal.

The idea of referencing a specific line or method when defining a task for an agent — that alone would save us hours. Right now we paste file paths and line numbers into prompts like cavemen.

Sandboxing agents in worktrees is the other killer feature. We've had agents step on each other's changes more times than I'd like to admit.

Really excited to try this with our stack. Multi-agent development isn't hypothetical anymore — teams are shipping production infrastructure this way today.

Great,

I am a fan of all JetBrains products,
I hope this fixes the problem of using different AI models inside JetBrains

As a JetBrains subscriber for over a decade (grandfathered, still using WebStorm + AI Ultimate), I’m curious: what’s the one key improvement or use case in JetBrains Air that justifies switching from my current workflow, especially given the reported stability issues i'm seeing down below?

Good luck on the launch. Hope it smooths out quickly for longtime users like me

Been using JetBrains for years and their database tool window still feels like having a senior DBA sitting next to me - caught a production query that would've cost us $3k/day in unnecessary joins last week. The way they surface performance hints while you're writing SQL is almost unfair to other IDEs.

I can't recommend this product, because I've tried to make it work and it just doesn't. Command execution didn't start (it got stuck in a scheduled state). I've tried to reach out to @JetBrains on X, but no reply. Eventually I found an issue in their issue tracker, started to wait for the resolution and guess what? The issue is gone now. I used to love JetBrains, I'm paying for WebStorm and JetBrains AI (Ultimate), but this is the last year I'm doing it, because the frustration from their products is immeasurable nowdays.

Everything just flows seamlessly—no bumps, no confusion, just pure elegance

Been juggling Claude Code and Codex on the same codebase and they keep stepping on each other's files. So the isolation part — does each agent get its own worktree, or is it more branch-per-task? Curious how the review works when two agents touch the same file.

been a user of IntelliJ IDEA, Android Studio for ages hence this launch feels personal.

Congrats on the launch team!