The Sequence Opinion - Issue #975: The Architecture of Agentic Swarms
How to design systems that explore together and verify their work

TL;DR
- Agentic swarms involve multiple autonomous agents working towards a shared goal with partial independence, selective communication, and feedback.
- OpenAI's September 2026 Navier-Stokes research utilized around 10,000 concurrent agents, achieving a result after 88 hours, followed by 17 hours of formalization and verification.
- The pattern is traced through frameworks and systems from OpenAI, Google, and Anthropic.
- Key design considerations for swarms include task division, communication protocols, and verification mechanisms.
- Problems with multiple promising avenues and checkable outputs are well-suited for agentic swarms.
- OpenAI reported a finite-time singularity under smooth external forcing for the Navier-Stokes problem, which Clay is currently evaluating.