Claude Code projects can burn your limits faster

On September 17, Anthropic opened a beta of reworked projects in Claude Code. You give one conversation a goal, and Claude splits it into parallel threads, hands out the work, reviews what comes back and assembles the result. The threads keep running in the cloud after you close the laptop, Anthropic said on X.

Access is narrow for now. The beta covers select Pro and Max subscribers who use cloud sessions in Claude Code and have no existing projects on the web or desktop. More Claude Code users on those plans are due to get it over the following week. After that come updated projects across the rest of Claude and on Team and Enterprise plans. Pro and Max subscribers who don’t have it yet can join a waitlist. Existing projects on Pro and Max keep working as before, and Anthropic says it will upgrade them as the rollout reaches chat and Cowork, now merged into one Claude.

One coordinator, many sessions

A project has two layers. Threads do the work, and a coordinator chat directs them. When you start a project you pick a goal and point it at a repo or other context, and Claude begins by suggesting work it can pick up right away. You set up the project’s cloud environment, connectors, plugins, instructions and model.

From there, Anthropic says you brief Claude “the way you’d brief a chief of staff”: several requests at once, in any order. The coordinator sends each request to a new thread or an existing one and checks in when it needs you. You can follow everything from the main project chat or open a single thread to look at the details and steer it. An Overview panel shows what is waiting on you, and steering works from a phone too.

Under the hood, each thread is a Claude Code cloud session working on its own branch and its own copy of the repo. With repositories connected, a thread runs your tests and opens pull requests. With documents, it reads them and writes drafts. Threads can also split their share of the work further with subagents, loops and workflows when an assignment is large.

The coordinator keeps the work organized, but if two threads work on the same code, the overlap is resolved as a merge conflict, just as it would be in any other PR. Git access in cloud sessions goes through a proxy that limits Claude to authorized repositories and adds the GitHub credential on the server side, so the token never enters the session’s machine, as Claude’s documentation on cloud sessions describes.

Until now, Anthropic says, running several sessions across one build meant you divided the work, juggled the handoffs and stitched the results back together yourself. Its own examples show what moves to the coordinator:

• Set a goal to cut your app’s checkout p75 latency, then ask Claude to profile each endpoint, test optimizations and open PRs in parallel threads.
• Connect your API, web and mobile repos and set a goal to retire a deprecated v1 endpoint. Claude creates a thread per repo to migrate the callers, run the tests and open PRs, then tells you which ones need to merge first.

What the project remembers

Anthropic built projects for long-running and agentic workflows: work that takes longer than one reply and has more than one part. Every thread adds to a shared project memory and reads from it, which Anthropic says reduces the need for complex prompt engineering. Its examples: Claude can remember that the release moved to Friday, why the export was dropped, or who to check with before touching the billing service.

The memory also covers how you like to work. You can ask Claude to change how often it checks in, how often it starts new threads, or how detailed its updates are. Alongside the memory, each project has a library that collects the files you add and the artifacts Claude produces, so new work can build on earlier results.

Where the limits bite

Each thread is a full Claude Code session, so a project running several at once can reach usage limits faster. Cloud sessions carry no separate charge for the cloud machine. They draw on the same limits as the rest of your Claude Code usage, so a busy project spends the same allowance you would otherwise use in the terminal. To manage that, projects show their own usage, and you can choose the model and effort level separately for the coordinator chat and for the worker threads.

The other limit is where threads run. For now they run only in the cloud, which is why they keep going while your computer is offline. It also means they can’t reach files or tools on your machine or anything on your internal network. Running threads on your own machine, next to your local tools and code and behind your network, is “coming very soon”, Anthropic says.