AGI Pulse
ProductAnthropic · agipulse

Claude Managed Agents gets dynamic workflows in public beta: a lead agent writes the plan and runs up to 1,000 agents per run; Claude Code Projects opens to every waitlisted Pro and Max user

0 reads

Published
2026-10-10
Type
Product
Organization
Anthropic
Source
agipulse

Why it matters

【Key points】

On Oct 9 Anthropic announced, via @ClaudeDevs and the Claude Platform release notes, that dynamic workflows for Claude Managed Agents are in public beta. When a job has many parts, such as reviewing hundreds of documents, the lead agent writes a program that fans the work out to many agents in phases in the background and merges the results; Anthropic says a run can orchestrate up to 1,000 agents.

Later the same day @ClaudeDevs said every Pro and Max user on the Claude Code Projects waitlist has been let in. A project is an ongoing conversation where Claude coordinates the work and runs each task in its own parallel thread. Projects remain in beta.

【Verifiable】

The release notes give the switch: set the agent's multiagent field to multiagent_20261001 with workflows enabled, under the managed-agents-2026-04-01 beta header, and follow runs through workflow_run events. The docs note that subagent delegation is one level deep with at most 25 child threads. Anthropic's internal test planted 70 bugs in a 116k-line codebase: a single agent found 14, 15 and 27 across three runs; a workflow found 66 every time. Anthropic warns workflows can use many tokens. The two launch posts drew roughly 450k and 500k views, and about 870 related X posts appeared within six hours.

【Our read】

The headline number is 1,000 agents, but the real shift is that orchestration moves into the model: developers stop hand-wiring multi-agent pipelines and instead tell the lead agent when to start a run. The jump from the teens to 66 bugs suggests that for find-everything tasks, breadth beats one agent thinking longer. The cost is just as direct: a run can burn many times the tokens of a single agent, so it pays off only when the work splits and verifies cleanly. 66 of 70 is Anthropic's own benchmark; until others reproduce it, treat it as a ceiling rather than a typical gain.

Read original →

Related developments

Aggregated by AGI Pulse. Titles and links only; no full text is republished. Editorial policy