Anthropic Launches 1,000-Agent Claude Workflows

Anthropic has officially launched the public beta for Claude Managed Agents dynamic workflows on the Claude Platform. Enabled using the versioned configuration parameter multiagent_20261001, the system allows a lead agent to write a custom JavaScript orchestration program that delegates tasks across up to 1,000 managed subagents per run. This ceiling expands upon the previous local research preview, which was limited to approximately 16 concurrent subagents due to resource constraints.

In a benchmark test featuring 70 planted bugs within a 116,000-line codebase, a single agent identified 14, 15, and 27 bugs across three separate attempts. By comparison, the dynamic workflow successfully identified 66 out of 70 bugs in each of its three runs. The parallel structure allows subagents to inspect isolated segments of a repository, cross-examine overlapping findings, and validate defects before submitting results to the primary coordinator.

For software developers and practitioners, this tool opens up new possibilities for large-scale code analysis. Heavy enterprise tasks—such as repository-wide bug hunts, framework or API migrations, multi-perspective security audits, performance investigations, and dependency upgrades—can now be split cleanly across bounded context windows. Users can quickly scaffold this workflow in Claude Code using the command /claude-api managed-agents-onboard bug-hunter.

Despite these gains, multi-agent orchestration adds operational overhead, increased token consumption, and higher latency. Tightly scoped edits or single-file changes remain better suited for single-agent runs. Teams adopting the public beta should establish strict permission allowlists, token spending limits, and verification parameters to avoid runaway API costs or unintended modifications.