Harrison Stall is a member of technical staff at Anthropic working on agent APIs. He copresented an AI Engineer talk on dividing large agent workflows into phases and inspectable delegated tasks.[1]
Start with Why One AI Agent Isn’t Enough (And How to Run 1,000) - CJ Avilla & Harrison Stall, Anthropic →
Links
X profile Why One AI Agent Isn’t Enough Member of Technical Staff · Anthropic
Contributions
The conference record credits Stall and CJ Avilla with a session on Claude Managed Agents. Its account describes a lead agent dividing work into phases and delegating narrow tasks with separate contexts. Budgets, event streams and interrupts help an operator inspect and intervene in a long run. The central problem is coordinating work and checking its coverage, rather than assuming that a larger number of agents automatically produces a better answer.[2]
Tools, writing & talks
Writing, research & talks
Start here
Why One AI Agent Isn’t Enough (And How to Run 1,000) - CJ Avilla & Harrison Stall, Anthropicyoutube.comA joint talk on organizing, observing and controlling multi-agent workflows.
Start with the phase-and-delegation explanation and map it onto one bounded task. Familiarity with agent APIs helps; treat the product limits described in the recording as dated details to verify before reproducing the setup.
Copresenter with CJ Avilla[2][3]Original source