It's true that you can redirect API calls fast with minimal infrastructure, but the idea that firms' adjustment costs for custom prompts somehow contributes to this ease is missing the bigger picture.
If we hadn't already poured hours into training our staff on one specific AI system, would we re-enter that commitment with the same provider if a cheaper, slightly better option appeared?
Probably not, because the actual sunk cost isn't the API, it's the custom coding, the embedded workflows, and all the employee retraining.
For example, if we’ve spent months integrating a specific AI for our scheduling system, shifting to a new one means not just changing the API key, but rewriting all the custom scripts and retraining every manager.