Anthropic launched a research-preview feature called ‘dreaming’ for Claude Managed Agents at its Code with Claude developer conference. The system runs scheduled reviews of an agent’s past sessions and memory stores, extracts patterns, and curates memories so agents improve over time — surfacing recurring mistakes, convergent workflows, and team-shared preferences that no single session could identify. Early adopters report significant gains: legal AI firm Harvey saw task completion rates increase roughly 6x; medical document reviewer Wisedocs cut review time by 50%.
Why it matters: Persistent, cross-session self-improvement is a foundational capability for enterprise-grade agentic AI — this is the clearest signal yet that agents are moving from stateless tools to continuously improving systems.