Claude Projects Now Manage Themselves, Mostly

Anthropic has overhauled Claude Code projects, transforming static folders into a conversational management layer where a coordinator agent splits goals across multiple parallel cloud sessions. Instead of manually stitching together AI outputs, you simply describe a target like migrating an API or reducing latency, and Claude spawns threads to write code, open PRs, and run tests while you watch from your phone.

• Conversational delegation: A central coordinator scopes your request, assigns tasks to parallel threads, and handles long-running workflows without constant handholding.
• Shared institutional memory: Threads draw from a persistent memory bank that retains project context, past decisions, and your specific communication style over time.
• Early access limits: The beta is currently rolling out slowly to select Pro and Max subscribers using cloud sessions, assuming you do not trip over existing project restrictions.

Why should I care? Could be big
Multi-threaded agent coordination looks promising, but expect heavy token consumption and merge conflicts before it replaces real engineers.

Read the original: Projects redesigned: from folder to conversation