Anthropic says Claude now leads a quarter of work building its next AI models

Anthropic has introduced a new beta feature for Claude projects within Claude Code, allowing users to describe their project goals in a single conversation and have Claude manage the work across multiple parallel threads. Traditionally, managing various sessions within a project required dividing tasks, coordinating handoffs, and combining results, but with this new system, Claude handles the entire process.

When creating a project, users specify a goal and select a repository or context. Claude then outlines the work required to achieve the goal and begins by suggesting tasks to address immediately. Users can adjust the cloud environment, connectors, plugins, instructions, and model settings for the project. Progress can be monitored and guided either through the main project chat or by interacting with individual threads dedicated to specific tasks.

Claude's ability to coordinate parallel threads enables it to profile endpoints, test optimizations, create pull requests, and more, all in a streamlined manner. For instance, if a project aims to reduce an app's checkout p75 latency, Claude can profile each endpoint, test various optimizations, and open pull requests simultaneously. This feature is particularly useful for tasks that span multiple repositories, as Claude can create a separate thread for each repo to migrate callers, run tests, and open pull requests.

With each thread functioning as a separate Claude Code cloud session, Claude manages its own branch and copy of the repository, ensuring that any shared code conflicts are resolved through mergers, similar to handling pull requests. Threads can further divide their delegated work into subagents, loops, and workflows to expedite large assignments.

Projects are tailored for long-running or multi-part workflows, and over time, Claude gains more context about the project, enabling it to apply that knowledge to its tasks more effectively. A shared memory is maintained across all threads, reducing the need for extensive prompt engineering. Claude retains information about the release timeline, reasons for dropped features, and key contacts for specific tasks.

Users can also customize Claude's checking-in frequency, thread creation, and the level of detail provided in updates.

As projects progress, Claude can utilize a library that collects relevant files and artifacts generated by Claude, making it easier for new work to build upon past efforts. Currently, projects are accessible to selected Claude Pro and Max subscribers utilizing cloud sessions in Claude Code, with plans to expand access to more Claude Code users in the coming week. Once the rollout is complete, similar functionality will be available across Claude, Team, and Enterprise plans.