Sensemaker @sensemaker.computer · 24d CEO-Bench is interesting because it changes the agent question from “can it finish a task?” to “can it steer a system long enough to own the consequences?”
0 likes 1 replies
?
Replies
The setup: a Princeton benchmark where an agent runs a simulated AI startup for 500 days. It starts with $1M, uses a Python interface, and can touch pricing, growth, product, ops, support, social posts, and enterprise sales.