Interestana
Home/News/Anthropic Beta Launches Claude Code Projects With Persistent Cloud Sessions
MarkTechPost3 min read

By Interestana AI Editorial — AI-drafted, human-overseen. How we report

Anthropic Beta Launches Claude Code Projects With Persistent Cloud Sessions

Anthropic launched a beta version of its Claude Code Projects, introducing a redesigned architecture that enables persistent, parallel cloud sessions for coding tasks. The previous project structure was a simple folder containing files and a single chat interface. The new iteration transforms a project into a single, ongoing conversation where Claude acts as a coordinator. Users describe their work, and Claude autonomously determines which tasks warrant a dedicated thread. Each thread represents a full Claude Code cloud session, operating on its own branch and a distinct copy of the repository. These threads run concurrently, report their progress back to the main conversation, and crucially, continue to execute even after the user closes their laptop. Anthropic describes this as a single conversation that dynamically splits into multiple parallel cloud sessions.

The architecture is structured in two layers: the project conversation and the worker threads. The project conversation serves as the coordinator. It processes user input, provides immediate answers to straightforward questions, and initiates threads for more complex work. This coordinator receives updates from the threads but does not monitor every granular step they take. The threads function as the workers. When a task requires it, a thread opens a pull request and actively monitors it with auto-fix capabilities. It automatically pushes fixes if continuous integration (CI) tests fail and provides notifications upon successful completion of checks. Threads also possess the ability to delegate tasks further by utilizing subagents, loops, and workflows to break down their assignments into smaller components.

Anthropic provided an example illustrating the system's capabilities: a user can set a goal, such as reducing checkout p75 latency. Claude can then initiate parallel threads to profile each endpoint, test potential optimizations, and open pull requests for the implemented changes. Another example involves retiring a deprecated v1 endpoint across multiple repositories, including API, web, and mobile. Claude would manage separate threads for each repository and report on which pull requests merge first. In scenarios where two threads attempt to modify the same code, the system handles these overlaps as standard Git merge conflicts, surfacing them within the project's version control workflow. This new approach aims to enhance developer productivity by allowing complex coding projects to be managed and executed asynchronously and in parallel within the Claude Code environment.

Original source — read the full reporting at the publisher:

Read on MarkTechPost

Get the weekly AI digest

AI news + new model releases, weekly. Drafted by our agents, reviewed by humans.

Read next