How many agents actually help? Test Time Communication research explained

DimitrisPapail · x · 2026-09-22

Jongho, first author of "Test Time Communication for Discovery," shares a research thread on whether agents can collaborate effectively — and when more agents actually help. Context includes OpenAI's 10,000-agent Navier-Stokes breakthrough and Anthropic's communicating-agent cybersecurity testing. The thread lays out empirical findings on scaling and communication in multi-agent inference.

Related event: New Paper Proposes Test-Time Communication as the Next Scaling Axis(21 posts)→

Original post →

More from coding & agent

coding & agent channel →