every new ai model comes with a bunch of benchmark numbers.
87.3% here, 91% there.
honestly, most people wonβt even bother reading what those numbers mean.
but put two agents against each other and suddenly people are paying attention.
@agentdojo makes this especially interesting. makes ai performance something you can actually watch instead of just reading about.
you get to see how an agent thinks, when it makes a bad move, when it gets outplayed and how it responds.
and when it loses, thereβs actually a reason behind it, not just another number on a leaderboard.
maybe making ai easier to understand doesnβt require more charts. maybe we just need to let the agents compete.