21:51
2026-09-22
dev.to
ai-agents
We ran 2 vs 4 agents six times. Four agents cost 2.1 and did not improve success
A developer ran a preregistered, capped comparison of 2-agent versus 4-agent configurations on a deterministic public-repair task, completing six live runs across three repeats per group. Both groups …