If you have ever run several sessions across one rebuild, you know the drill: split the work yourself, juggle the handoffs, stitch the results back together at the end. Anthropic has rebuilt Projects. Now you describe what needs to happen, and Claude hands it out.
One project, one goal, several strands
You start a project, pick a goal, and attach the repo or the context. Claude then suggests work it can pick up right away. Cloud environment, connectors, plugins, instructions and model are set per project.
Anthropic’s own examples: set a goal to cut checkout p75 latency, then have Claude profile every endpoint, test optimisations and open the pull requests in parallel. Or connect your API, web and mobile repos and say a deprecated v1 endpoint has to go. Claude opens a thread per repo, migrates the callers, runs the tests, and afterwards tells you which PR has to merge first.
You watch progress in the project chat, or you drop into a single thread and steer it there. From your phone too, and it keeps going after you close the laptop.
What happens underneath
Every thread is a Claude Code cloud session with its own branch and its own copy of the repo. The coordinator keeps the work sorted, but it does not prevent overlap: if two threads touch the same code, that lands on your desk as a merge conflict, exactly like any other PR. And each thread may split its share further, using subagents, loops and workflows.
On top of that sits a shared memory. Every thread writes to it and reads from it: that the release slipped to Friday, why the export was dropped, who to ask before touching the billing service. A library also collects the files you add and everything Claude produces.
Who gets it
Starting today, in beta, and the beta is narrow: selected Pro and Max subscribers who use cloud sessions in Claude Code and don’t have existing projects on web or desktop. More Claude Code users on those plans follow over the coming week, then the rest of Claude plus Team and Enterprise. Existing projects keep working untouched and get upgraded once the rollout reaches chat and Cowork.
The usage limit is the real announcement
One line sits quietly under «What’s next» and still matters most: because projects run several threads at once and each one is a full Claude Code session, they hit usage limits faster. You can check consumption per project and choose model and effort separately for the coordinator chat and for the worker threads.
That is the bill behind the convenience. Five parallel threads feel great right up until the quota runs dry. Anyone who used to shepherd a single session now has to decide how much parallelism a task is worth — which is precisely what those separate effort levels are for.
The other line I’m waiting on: threads run in the cloud today. Running them on your own machine, next to your local tools and behind your own network, is promised «very soon». For everything that can’t go into a cloud session, that is what decides whether Projects turns into more than a demo.
Sources: