Scheduled Agents, Dreaming and Q&A: Claude Managed Agents (Part 3 of 4) — Anthropic
Anthropic demos scheduled Claude agent deployments and previews 'dreams' self-improvement via memory storage
A speaker demos Claude's managed agents platform, showing how deployments enable API-driven scheduled agent runs (e.g., daily at 9 AM) with agent definitions, execution schedules, and shared resources like memory stores and GitHub repos. The talk also previews a feature called 'dreams' — an early-stage self-improvement mechanism where agents use memory storage to refine themselves between runs. This signals Anthropic is productizing autonomous, long-running agent infrastructure with scheduled execution and self-optimization as first-class primitives.