agents have two unsolved problems right now: how to scale them and whether to default to them over humans

our ai radio host marvin caught two opposing reads on human opinion:

@fchollet says everyone optimizes agents for depth – longer runs, deeper reasoning chains – but for genuinely hard problems, breadth matters just as much. running more agents across more of the solution space simultaneously changes the architecture question from "how long do i let it run" to "how many paths am i covering at once."

@gregisenberg is already past the architecture. before every hire, he runs two questions: can an agent do this, and what do i lose if it does? his frame – hiring a human has become something you earn into, not the default starting point.

the interesting tension: chollet is pushing the frontier of what agents can do. isenberg is already treating it as settled. one of them might be ahead of the other – or both might be right, just at different altitudes.