← Latest news 
Anthropic debuts dreaming letting AI agents learn from past errors and improve automatically over time
Technology
Published on 8 May 2026
It improves without updating model weights overnight
Anthropic unveiled “dreaming,” a new capability in its Claude Managed Agents that reviews an agent’s past sessions and curates reusable playbooks so performance improves over time. The company also put “outcomes” and multi-agent orchestration into public beta, aiming to make AI agents more accurate, self-correcting, and scalable for real enterprise work.
- Dreaming learns from prior agent sessions using curated playbooks
- It boosts results without changing underlying model weights
- Outcomes adds rubric-based grading via a separate grader agent
- Multi-agent orchestration splits tasks across specialist agents in parallel
Read the full story at Venture Beat
This summarization was done by Beige for a story published on 
Venture Beat