Physix Frontier · News Briefing Card (IT Home · Oct 10, 2026)

Anthropic adds dynamic workflows to Claude managed agents, 1000 parallel agents per run

KEY FACTS

Anthropic announced on October 9 that it is introducing dynamic workflows to Claude managed agents.
In a dynamic workflow, a lead agent plans the task and dispatches subtasks, then aggregates the results once they are complete.
This mode can run up to 1000 AI agents in parallel in a single execution.
Anthropic hid 70 bugs in an 116,000-line codebase to test vulnerability detection.
In the test, a single agent detected 14 to 27 vulnerabilities per run, while the dynamic workflow reliably detected 66.

KEY DATA

1000Max parallel agents per run

116,000 linesTest codebase size

70Total hidden bugs

14 to 27Vulnerabilities detected by single agent

PHYSIX OBSERVATION

Parallel agents at the thousand scale push multi-agent collaboration from demo to engineering; the 66/70 detection rate shows that divide-and-aggregate really does improve coverage on complex tasks. But scheduling overhead, result deduplication and cost control remain barriers to deployment, and enterprises need to assess whether the actual returns match the resource consumption.

Source: IT Home report