Anthropic appears to be preparing a new kind of project for Claude, one where the assistant does more than hold files and answer questions. A recent build surfaces a choice, at the moment a project is created, between a standard project and a “managed” one, described with the line that Claude takes on tasks and keeps the project organized. That framing points to a persistent Claude environment that carries the surrounding context, works with some autonomy, and could run scheduled work rather than waiting for each prompt. Beside it, the projects area of the side navigation is being reworked to list only pinned projects, based on tester feedback. Both are marked internal for now, limited to staff and a small circle of trusted testers, with no release date attached.
The pieces line up with work Anthropic already ships elsewhere. Its developer platform has offered Claude Managed Agents since April, cloud-hosted agents that hold state across sessions and refine their own memory through a scheduled process the company calls dreaming. Cowork projects already run scheduled tasks, and Claude Code carries context between sessions through project memory and instructions. Managed projects read like those threads gathered into one consumer surface, where sessions in a project would share memory and instructions and each one feeds the project’s wider knowledge. The build also notes a project can be personal or shared, with shared projects held for Team and Enterprise plans.
For Anthropic, this fits a steady move from chat toward standing, self-organizing workspaces. The company has tried knowledge bases before and has been running Conway, its always-on internal agent, which is set to close on July 24, timing that could mark a handoff rather than an ending. If managed projects reach users, they would offer a project that quietly maintains itself between visits, a shape no rival lab currently sells. A firm date is still missing, but internal testing is plainly live.
