Claude Code threads collide as ordinary merge conflicts

Anthropic on September 17 rebuilt Projects in Claude Code around a coordinator. You give the project a goal, and Claude scopes the request, hands the pieces to parallel threads, reviews what comes back and assembles the result. You steer it from the project chat, from inside any single thread, or from your phone, and the work keeps going after you leave your computer.

The change replaces work developers did by hand. In Anthropic’s post on the Claude blog, the old way of running several sessions across one build meant dividing the work yourself, juggling handoffs and stitching the results back together. The post gives two examples. In the first, the goal is to cut an app’s checkout p75 latency, and Claude profiles each endpoint, tests optimizations and opens PRs in parallel threads. In the second, you connect your API, web and mobile repos and set a goal to retire a deprecated v1 endpoint. Claude creates one thread per repo to migrate the callers, run the tests and open PRs, then tells you which ones need to merge first.

What a thread is under the hood

Each thread is a Claude Code cloud session working on its own branch and its own copy of the repo. Mechanically, a cloud task clones the repository fresh from GitHub into an isolated VM, optionally runs a setup script, applies the environment’s network level (None, Trusted, Full or Custom), and pushes Claude’s branch back to GitHub when it reaches a stopping point. Cloud sessions first appeared as Claude Code on the web in October 2025, per AQ, which also notes that GitHub credentials never enter the session’s VM: git operations go through a proxy that attaches the credential on the server side.

That isolation is what lets the coordinator run threads side by side. A local terminal session shares your working tree, so two sessions on one repository can edit the same files and fight over the same port, the claude.dev blog explains. In a project, the coordinator keeps work organized, but it does not referee edits: when two threads touch the same code, the overlap comes back as a merge conflict, just like any other PR.

Threads can also fan out. Each one may split its assignment further with subagents, loops and workflows, so large assignments finish faster. With repositories connected, a thread opens pull requests and runs your tests; with documents, it reads them and drafts.

Setup starts with a goal plus a repo or other context, and Claude opens by suggesting work it can pick up right away. You can configure the project’s cloud environment, connectors, plugins, instructions and model. Anthropic’s advice is to “brief Claude in the project the way you’d brief a chief of staff”, and it routes the request to a new thread or to one that already exists. Cursor’s Projects beta runs on the same split, with a coordinator you chat with that dispatches tasks to coding agents instead of writing code.

Memory and the library

Anthropic pitches projects for long-running, agentic work: anything that takes longer than one reply and has more than one part. Every thread adds to and draws from a shared memory, which the company says reduces the need for complex prompt engineering. Its examples of what Claude keeps are mundane on purpose: that “the release moved to Friday”, why the export was dropped, or who to check in 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 readily it starts new threads, and how detailed each update should be. Alongside the memory sits a library that collects the files you add and the artifacts Claude produces, so new threads can build on earlier output.

Who gets it, and what it draws on

The beta opened on launch day for select Pro and Max subscribers who use cloud sessions in Claude Code and have no existing projects on the web or desktop. Anthropic said access would widen to more Claude Code users on those plans over the following week, with updated projects across all of Claude, and on Team and Enterprise plans, coming after that. Pro and Max users without access can join a waitlist. Existing projects on Pro and Max keep working as before and will be upgraded as the rollout reaches chat and Cowork. Later, on September 23, cloud sessions themselves became generally available for Pro, Max and Team users, and for Enterprise users with premium seats or Chat + Claude Code seats.

Usage is the part to watch. Because a project can run several threads at once and each is a full Claude Code session, Anthropic warns that projects can reach usage limits faster. There is no separate charge for the cloud machine, but cloud sessions share rate limits with all other Claude Code usage, so every parallel thread draws on the same allowance you also spend at the terminal. To manage that, the project shows its own usage, and you pick the model and effort level separately for the coordinator chat and for the worker threads.

Threads run only in the cloud for now. Running them on your own machine, alongside your local tools and code and behind your network, is listed by Anthropic as “coming very soon”.