Which AI do you actually use, and does it matter?

This gets asked as a technical question but it is usually a competence check, which is fair enough.

The straight answer

It depends on the job, and I will tell you which and why before we build. Different models are better at different things — some at careful reasoning over long documents, some at speed and cost at volume, some at running reliably inside a defined format.

Anyone who tells you one model is best for everything is either simplifying for your benefit or has only ever used one.

What matters more than the model

What you send it. Most of the quality difference between a good system and a bad one is what context the model receives, not which model receives it. Poor retrieval with an excellent model beats nothing, but good retrieval with a mid-range model beats it comfortably.

Whether you can switch. This is the one to hold me to. Systems should be built so the model is a component, not a foundation. Providers change prices, deprecate versions, and occasionally have bad weeks. If swapping means a rebuild, you have been architected into a corner.

The terms it runs under. Business API tiers and consumer apps have different commitments about your data. That difference matters more to you than a few points of benchmark performance.

The version worth asking

Rather than "which model", ask: if this provider doubled their price next year, what would we have to do? The answer tells you how the thing is built.