> The idea that for basically sub-agents, we can fine-tune them, should reasonably expect to perform as well as Opus for a specific subtask of which my applications have many [...] we can run a general-purpose intelligent model, Sonnet or Opus, orchestrating a fleet of, let's say, 30 to 50 of these sub-agents that have been fine-tuned

I've heard so many people saying this for the last year, and even tried doing it myself too, and never seen a successful application of it, nor succeeded myself either with SOTA models that are smart but slow or local models that are dumb but fast (even with beefy hardware).

What makes you believe this is possible in the first place? Every "swarm of agents" implementation I've seen only been able to produce lowest quality of code, most of the time vastly bloated, but surely you must have seen something working in practice that you could share with the rest of us?

I guess it depends on a task. Opus is already spawning Sonnet/Haiku for simple tasks with a good success rate.