During the panel “Venture and growth: The AI opportunity in early focus”, I found myself asking about model routing. When a company chooses which model handles a task, how much is driven by economics, and how much by the support available from a provider?
That choice seems more revealing than the name of the model in a product announcement. It brings the discussion down to the work: what the task needs, what a satisfactory answer costs, and how much effort it takes to keep the system reliable.
My question is whether some of the advantage sits in making those decisions repeatedly and well. A company can have access to the same models as its competitors and still make very different choices about when to use them.
I would like to see this at the level of one real workflow: which calls go where, which need checking, and what changes when usage increases. That would tell me more than hearing that a company uses several models.
The question I left with: who inside the business owns those trade-offs, and how do they know when the routing decision needs to change?