How Anthropic’s Dynamic Workflows Coordinate Claude Agents
Anthropic showed dynamic workflows that let Claude plan and orchestrate its own multi-agent topology.
At AI Engineer New York, Anthropic engineers described dynamic workflows, or flows, a multi-agent mode for Claude Managed Agents announced the previous Friday. A lead agent writes an orchestration program that launches sub-agents, collects results, and manages information flow, instead of requiring a developer-configured coordinator and roster. Users can still supply specialist sub-agents, set a session budget the model cannot see while running, and inspect every handoff in Claude Console. A research-preview feature called dreaming can reorganize an agent's memory store from past session transcripts.
- Dynamic workflows let a lead Claude agent design its own multi-agent plan.
- Developers can set a session budget the model does not see while running.
- Dreaming, a research preview, reorganizes memory from past transcripts.
- Claude Console graphs show payloads passed between lead and sub-agents.
From the original article
5 min read at latent.space
At AI Engineer New York, Anthropic drew on lessons from Unix and Linux to explain how agents can tackle bigger projects without overwhelming developers.
By Richard MacManus · latent.space
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.