Claude Managed Agents gain persistent memory stores

Anthropic is aggressively tackling the "amnesia" problem in AI agents by moving state management from the developer's application layer directly into the model's managed runtime.

• –Workspace-scoped memory stores provide a reliable, versioned "past" for agents, enabling long-term learning and context retention that survives session restarts.
• –Filesystem-based mounting allows agents to use standard file tools to read and write state, significantly simplifying the development of complex, multi-step workflows.
• –Integration with the `claude-api` skill in Claude Code provides a unified CLI interface for managing agent memories and accessing real-time API documentation.
• –The shift to managed agent runtimes mimics the "Lambda-fication" of compute, prioritizing developer convenience and secure sandboxing over raw control, though it risks increasing vendor lock-in.