Anthropic is testing a new type of Claude project that operates with greater autonomy than the company's current offerings.

Internal builds show users can now choose between standard projects and "managed" ones when creating a new workspace. The managed option promises that Claude takes on tasks and keeps the project organized.

The feature suggests a persistent Claude environment that maintains context across sessions and could execute scheduled work without waiting for user prompts. This marks a shift from reactive chat toward proactive, self-organizing workspaces.

How managed projects connect existing tools

The development builds on Anthropic's existing infrastructure. The company launched Claude Managed Agents on its developer platform in April — cloud-hosted agents that retain state across sessions and refine their memory through a process called "dreaming".

Cowork projects already handle scheduled tasks, while Claude Code maintains context between sessions through project memory and instructions. Managed projects appear to unify these capabilities into a single consumer interface.

The builds indicate projects can be personal or shared, with shared access limited to Team and Enterprise plans.

Internal agent shutdown signals transition

The timing aligns with Anthropic's internal operations. Conway, the company's always-on internal agent, is set to close on July 24. This shutdown could signal a handoff to the new managed projects system rather than an ending.

The feature would give Anthropic a unique position among AI labs. No major competitor currently offers projects that quietly maintain themselves between user visits.

Testing remains limited to staff and trusted testers, with no public release date announced. The company is also reworking the projects navigation to show only pinned items based on tester feedback.

If launched, managed projects would represent Anthropic's most autonomous consumer AI offering to date.