← All speakers

Bio, Work & Ideas

Harrison Stall

Conference affiliation: Anthropic

Harrison Stall presented “Why One AI Agent Isn’t Enough (And How to Run 1,000)” at AI Engineer, credited with Anthropic in the official recording.

Read the topics behind these talks

1 conference talk

Key ideas

Scroll to read ↓

CJ Avilla and Harrison Stall explain how Claude Managed Agents Flows divide work, allocate compute across phases, and put controls around long-running agent systems.

  • Flows let a lead agent plan phases and delegate discrete work; at the time of the talk, the speakers describe a 1,000-agent limit and one layer of nesting.
    13:50 ↗
  • Use isolated contexts for narrow assignments and independent review, then evaluate coverage and quality rather than assuming parallelism guarantees either.
    26:24 ↗
  • Budgets, event streams and interrupts support inspection and intervention during long runs; raising a reached budget can resume completed work without starting over.
    15:20 ↗
  • Keep real credentials outside the agent environment, constrain where they can be injected, and separately restrict tools and network access.
    1:22:50 ↗

References