IT之家 · 智能时代 · 10/10/2026, 13:38:05
Anthropic Adds Dynamic Workflows to Claude Managed Agents: Parallelizing 1,000 Agents Boosts Bug Detection by Nearly 3x
Anthropic introduced dynamic workflows in Claude Managed Agents, enabling a primary agent to plan and dispatch sub-tasks with up to 1,000 parallel agents per execution. In a benchmark involving a 116k-line codebase, this approach consistently detected 66 bugs compared to 14-27 for single-agent runs, demonstrating superior efficacy in complex engineering tasks through massive parallelization.
SOURCE COVERAGEFull Original Report
On October 10, Anthropic announced yesterday (October 9) the introduction of dynamic workflows to Claude Managed Agents, enabling the parallel coordination of up to 1,000 AI agents in a single execution.
Note from IT Home: Dynamic workflows are an execution model where a primary agent plans tasks, dispatches subtasks, and aggregates results once the sub-agents complete their work. Typical use cases include splitting large tasks, such as code reviews, into multiple parts for parallel processing. In Claude Managed Agents, this mode supports running up to 1,000 agents in parallel per execution.
To demonstrate the capabilities of dynamic workflows, Anthropic simulated a code vulnerability detection scenario by hiding 70 bugs within a codebase of 116,000 lines, comparing the number of vulnerabilities detected by a single agent versus those detected using dynamic workflows. The test results showed that while a single agent detected between 14 and 27 vulnerabilities per run, dynamic workflows consistently identified 66.
