Teams running 35+ models in production is wild but it tracks.
They live across a team's whole portfolio, not just inside one app. Coding agents want the best reasoning, support bots want fast & cheap, research tools need long context.
Different jobs pick different winners.
@delba_oliveira It really depends on the scope of the task.
Overall I'm trying to avoid doing the thinking for the agent so I'll have things like nitpicky PM, head of function, security engineer, etc.
Each persona comes with examples of what matters to them, things to think about and so on