AnthropicfeatureClaude Managed Agents Update (Dreaming, Outcomes, Multiagent)
New capabilities for Anthropic's Managed Agents platform: scheduled self-improvement, rubric-based self-correction, and lead-agent delegation to parallel subagents.
Details
- Outcomes (public beta): agents self-correct against defined rubrics; raised task success by up to 10 points in Anthropic's tests (docx +8.4%, pptx +10.1%)
- Multiagent Orchestration (public beta): a lead agent delegates to specialist subagents running in parallel with a shared filesystem and persistent event history
- Dreaming (research preview, request access required): scheduled job that reviews past sessions and memory stores to extract patterns for self-improvement
- Webhooks added for agent completion events; full action visibility in Claude Console
- Customer reports: Harvey saw ~6x completion rate increase with Dreaming; Wisedocs cut document review time by 50%
Best for
developers building production Claude agents that need to self-correct, run long jobs, or coordinate parallel work
When to use it
shipping an agent where reliability and measurable success criteria matter more than chat-style back-and-forth
Who benefits
developers on the Claude Platform building with Managed Agents (Outcomes and Multiagent in public beta; Dreaming requires research preview access)
Join the newsletter to get ahead with AI.
I share strategies, new modules, free guides, and tools to get you better at using AI in your work.
Leading a team? See team programs →