A lot of AI strategy conversations spend most of their energy on one question: which model should we use? It's a reasonable question, and also, in a fast-moving field, one with a short shelf life. The model landscape a year from now will look different from today's, and the specific choice matters less than most planning meetings assume.
What matters more, and gets far less attention, is whether the architecture around the model — how it's integrated, how outputs are reviewed, how data flows in and out, how the system behaves if you swap the underlying model — is built to survive that change. An architecture tightly coupled to one vendor's specific API quirks will need to be rebuilt when that vendor's roadmap shifts, or when a better option appears. A vendor-neutral architecture treats the model as a replaceable component behind a stable interface, and absorbs that change without a rebuild.
This isn't an argument against having opinions about which model is best today — it's an argument for architecting as if you'll be wrong about that opinion eventually, because you will be. The organizations that adapt fastest to the next model generation aren't the ones who picked correctly today. They're the ones who didn't need to pick correctly, because their architecture didn't depend on it.