How do you avoid being locked into one AI vendor as the market keeps shifting?

We're building our product roadmap around AI features and we're effectively betting on OpenAI's API staying competitive and available. Every few months there's a new model that outperforms the one we're using, new pricing, and terms of service changes. How do companies architect their AI dependencies to avoid being locked into a vendor whose costs, capabilities, or policies might change in ways that hurt us?