Sarah Hooker: AI agents are the second major driver reshaping compute workloads

sarahookr · x · 2026-09-02

Sarah Hooker argues AI agents are the second major driver of compute demand. Historically, training and inference were predictable, continuous matrix multiplications—like a factory with one oven and one recipe, where only data delivery speed mattered. Unlike training, test-time compute doesn't need tight co-location: parallel agent calls tolerate individual failures, letting developers distribute workloads across hardware of different sizes and locations without penalty—reshaping how inference infrastructure gets built.

Related event: Sarah Hooker: Test-Time Compute and Agents Are Forcing an AI Infrastructure Rebuild(9 posts)→

Original post →

More from AGI Musings

AGI Musings channel →