AI agents, benchmarks and the productivity debate
Coinbase's CEO described an agent-based workflow while others questioned benchmark validity and whether AI progress is accelerating or a one-time speedup.
What we know
Brian Armstrong runs lead agents with a chief-of-staff agent. Toly and an account on X discuss benchmaxxing via MaxxBench, where models near 74% on DeepSWE end up 23 points apart. Robin Hanson reads METR's data as a one-time speedup, while Aaron Levie reports 10-100x productivity gains.
Sources
@brian_armstrong
“one lead agent for each area of focus, running all the time, and a chief of staff agent”
View post →By the Global Eye 24 desk, with AI assistance. Updated as the story develops. Spot an error? Write to corrections@globaleye24.com.