OpenAI Responses API is an attempt at lock-in

OpenAI has a new Responses API, which, in contrast to the de facto standard Chat Completion API that other providers have adopted as well, is stateful.

I know there are probably technical reasons that this API is an improvement. Yeah, you don't have to handle state in conversations yourself. Duh. But I think the change makes more sense from a business perspective: OpenAI would Very Much Like You To Stay With Them ThankYouVeryMuch, and not swap models out under your applications (with the same API, even!) when your evals tell you it's a good idea.

They want to not be a commodity service. I definitely understand their reasoning, but resist the urge! Don't give away the choice to swap model providers!

Data is like gravity. It pulls everything else towards it. Amazon figured this out long ago, which is why it's free to put byte after byte into AWS, but data transfer fees out are super high. S3 is a commodity, too, but with foundation models, YOU still own your data, and foundation models are just a stateless HTTP dependency.

Let's keep it that way.

(This post brought to you by a local foundation model enthusiast.)