Anthropic is testing an internal project mode called managed projects, spotted in a recent build by TestingCatalog. Creating a project now offers a choice between a standard setup and a managed one, where Claude organizes tasks and can run scheduled work between visits rather than waiting on each prompt. The feature is limited to staff and select testers, with no release date confirmed.
The idea extends tools Anthropic already has in production. Claude Managed Agents on the developer platform have held state since April, and Claude Code already carries memory across coding sessions. Managed projects would fold that persistence into the consumer chat product, giving it standing state instead of a fresh context every time.
Every coding agent is drifting toward the same target: a system that remembers and acts between sessions instead of resetting each time it opens. What separates the eventual leader will not be the chat interface. It will be how much memory a project keeps and how far its autonomy is allowed to run unsupervised.
Reported by TestingCatalog on July 22, 2026.
