Anthropic unveiled three major updates for Claude Managed Agents at the Code with Claude conference. The most intriguing one: Dreaming — a feature that literally lets agents learn from their own mistakes.
Dreaming: agents that improve themselves
Dreaming is a scheduled process that reviews previous agent sessions and memory stores, extracts patterns, and curates memories. The key insight: Dreaming surfaces patterns that a single agent can’t see on its own — recurring mistakes, workflows that agents converge on, and preferences shared across a team.
You decide how much control to hand over: Dreaming can update memories automatically, or you can review changes before they land.
Outcomes: define success instead of dictating steps
With Outcomes, you write a rubric describing what success looks like — and the agent works toward it. A separate grader evaluates the output in its own context window, independent of the agent’s reasoning. When something isn’t right, the grader pinpoints what needs to change, and the agent takes another pass.
This is a fundamental shift: instead of prescribing every step, you define the goal and let the agent find the path.
Multi-agent orchestration: divide and conquer
When a job is too big for a single agent, a lead agent can break the work into pieces and delegate each one to a specialist — each with its own model, prompt, and tools. The specialists work in parallel on a shared filesystem.
For example: a lead agent investigates an incident while sub-agents simultaneously comb through deploy history, error logs, metrics, and support tickets. Results flow back into the overall context.
My take
Dreaming is the feature I find most interesting. Agents that improve across sessions without me manually tweaking things every time — that’s a real step toward genuinely useful automation. We’ll see how well it works in practice.
Dreaming is available as a research preview, while Outcomes and multi-agent orchestration are in public beta.
Sources: Claude Blog · SiliconANGLE · SD Times