Home » Anthropic Rolls Out Parallel Claude Code Agents to Pro and Max

Anthropic Rolls Out Parallel Claude Code Agents to Pro and Max

By

–

Anthropic Rolls Out Parallel Claude Code Agents to Pro and Max

Until now, a developer juggling a big job in Claude Code had to start and manage each session by hand. Anthropic’s redesigned Projects, released in beta on September 17, hands that bookkeeping to a coordinator.

The user starts with a goal and the relevant repository or context. The coordinator scopes the request, spots the work that can run on its own, sends tasks to new or existing threads, watches the results and pulls them together. Each thread is a full Claude Code cloud session on its own branch and copy of the repository, not a small subtask inside one conversation.

What it looks like in practice

Anthropic offers two examples. In one, the aim is to cut an application’s checkout p75 latency: separate threads investigate different endpoints, test optimizations and open pull requests in parallel. In the other, a deprecated endpoint is being retired across API, web and mobile repositories. Threads migrate callers and test each repository, then the coordinator works out the order in which the changes must merge.

Parallel work does not make integration problems vanish. If two threads edit overlapping code, Anthropic says the clash is handled as a merge conflict, the way conflicting pull requests from human developers are.

Shared memory

Threads can write to and read from project-level memory, so nobody has to rebuild context for each new session. That memory can hold things like a changed release date, the reason a feature was removed, or who must be consulted before a particular service is touched. Claude can also keep a user’s preferences about how often to report progress or create new threads. A separate library holds files the user supplies and artifacts Claude produces.

Persistence cuts both ways. A mistaken decision stored in shared memory can steer several later threads, so important project-level assumptions deserve a human look.

The cost

Every thread is a complete session with its own model usage, so Anthropic warns that Projects can hit usage limits faster. Users can inspect project usage and choose model and effort levels for the coordinator and the workers. Threads run in Anthropic’s cloud for now. Running them on a user’s own machine, next to local tools and code, is promised but not yet available.

Anthropic says access will widen to more Claude Code users on the Pro and Max plans over the following week, with broader Projects support across Claude and for Team and Enterprise plans later. Existing Pro and Max Projects keep working as they do today until the wider rollout reaches them.

The design pays off when a goal splits into truly independent pieces, such as multi-repository migrations or test generation. It helps less when a task is strictly sequential, when agents keep editing the same code or when decisions need constant human judgment.

Sources

Get breaking stories first on our Telegram channel.

More stories by

Most read

AI Tools

Artificial Intelligence

Business

World News