Claude Code Now Gives Every Task Its Own Computer
Most people run their AI coding agent the same way: in a terminal, on their own laptop. That works fine until you want two things done at once, or you close the lid and the agent dies mid-task.
Anthropic's answer, laid out in a new field guide on the Claude Code blog, is cloud sessions. Every task gets its own machine. Here's what that actually changes.
The laptop problem
A Claude Code session on your laptop depends on that laptop in three ways. It shares your working folder, so two sessions on the same project can edit the same files and fight over the same port. It runs with your credentials. And it stops the moment your computer sleeps or the Wi-Fi drops.
None of that is a big deal for one task. It becomes a big deal when you want an agent to work like a team member instead of a very fast autocomplete.
What a cloud session is
A cloud session runs Claude Code on a fresh virtual machine. Your repository gets cloned onto a new branch, your setup steps are already done, and the agent works there until it's finished. When it's done, the result sits on a branch you can turn into a pull request.
You can start one from the web, the Claude mobile app, the desktop app, Slack, or your terminal with a single claude --cloud command. Then you can check in on it from wherever you are.
It's included in Pro, Max, Team and Enterprise plans at no extra charge, drawing on the same usage limits as the rest of Claude Code. Existing Pro and Max subscribers can also claim a one-time bonus credit for cloud sessions ($100 on Pro, $250 on Max) until October 7.
Three bugs, 87 seconds
The guide's most convincing part is a real test. The author took a small sample project with three ordinary problems: a test that failed about one run in four, API docs that no longer matched the code, and a logger that built its lines by gluing strings together.
They started three cloud sessions within 16 seconds of each other, one per problem. The sessions ran for 61, 65 and 72 seconds. All three were done 87 seconds after the first one started.
The results were solid. The flaky-test session found a race condition in a cache, fixed it at the root, and ran the test suite 40 times in a row with zero failures. The docs session actually started the server, hit every endpoint, and found five ways the old docs were wrong. The logger session rewrote the logger, added five tests, and then honestly reported that the suite still wasn't clean because of the same race the first session was fixing.
That last detail is the interesting one. The agent noticed a problem outside its task, proposed the fix, and left it alone because that wasn't its job.
Why isolation is the real feature
Two of those three sessions started the same API server to test against it. On one laptop, they'd have collided on the port or stepped on each other's files. In the cloud, each had its own copy of the project, its own processes and its own branch. Neither knew the other existed.
There's a security angle too. Your GitHub token never enters the virtual machine. A proxy holds it, and the session gets a short-lived credential that can only push to its own working branch. If the agent goes off the rails, the blast radius is one branch.
The tradeoff is that isolated agents can't see each other's work. The guide's advice is simple: split parallel tasks along file boundaries, merge branches in a sensible order, and expect one session to flag problems another is already fixing.
The bigger picture
The pattern here goes way beyond coding. An agent that's tied to your laptop is a tool you babysit. An agent with its own machine, its own sandbox, and a clear place to hand back its work is closer to a colleague you delegate to.
That's the same idea behind OpenClaw. Your agent runs in its own always-on environment, keeps working when your laptop is closed, and reports back when it's done, rather than living and dying inside one open window. Anthropic just showed, with real timings, how much faster things move when you stop making agents share a desk.
If you want an agent that runs on its own machine and keeps going while you get on with your day, that's what we built.