Anthropic appears to be extending Claude for Desktop with a more structured project layer for Cowork, pushing the product closer to a persistent local workspace rather than a one-off agent session. The main addition spotted here is project creation for Cowork, with users seemingly able to attach either an existing local folder or create a new one during setup. This is significant because Cowork already revolves around access to user-approved folders, and Anthropic has been positioning it as a desktop agent for knowledge work beyond coding. A project wrapper would provide a tighter boundary for files, instructions, and ongoing work.
One notable detail is the appearance of scheduled tasks inside the project scope. This feature would allow users to keep recurring work tied to a single folder, instruction set, and objective, which is a logical next step for teams or individuals running repeatable research, reporting, or operations workflows. It's important to note that scheduled tasks in Cowork were already introduced in Anthropic’s Claude app release notes on February 25, 2026. Therefore, the new angle here is less about the existence of tasks and more about their apparent integration into project-based organization inside Desktop.
This development fits Anthropic’s broader product direction. Cowork is still framed as a research preview, but the company has been steadily adding the scaffolding required for longer-running agent workflows across Desktop, Chrome, and Claude Code. A project system for Cowork would align with that push by making local context reusable and easier to isolate.
Two other changes point in the same direction:
- The revived pixelated avatar flow looks cosmetic, but it suggests Anthropic is still polishing identity and profile surfaces inside Claude.
- More consequential is the apparent work on a new Claude Code auto-permission mode. Anthropic already supports a bypass-permissions setting, with explicit warnings to use it only in sandboxed environments. An intermediate mode that lets Claude decide when approval is necessary would sit between full bypass and constant prompts, which is exactly the tradeoff Anthropic has been exploring in its work on safer autonomy and sandboxing.
No public rollout date is visible yet, but the direction is clear: Anthropic is making Claude Desktop less like a chat app and more like a managed operating surface for ongoing agent work.
