This is textbook strategy, adding layer after layer of pseudo protocols and "standards" on top of (surprise, surprise) their hard to defend against competition offerings, and ironically attempt to build a moat around something that ideally is open and interoperable. I haven't ruled out that I'm actually an idiot who doesn't get it, but it sure feels like the emperor has no clothes. It's like a recent Cloudflare post about how we're using LLMs to code all wrong; we just need to build (yet more) APIs specifically for MCP servers and then have very simple tools that only use those APIs. Yet more extra effort, time and energy optimizing for someone else's benefit.
It's a new space, everyone is coming up with their own protocols. It's really no different than Stripe offering an API so you can use them as a payment provider. You need to implement this API on your website so ChatGPT can allow users to buy stuff from you.
You need some amount of strictness in the API here, LLMs are not actually sentient. You could say that this is a failure on OpenAI's part in comparison to their marketing, sure.
IMO this view puts the cart before the horse. Suppose ChatGPT builds a UI that consumers prefer over web browsing to purchase products. How should merchants plug into that UI? This protocol is an easy way for merchants to do that. And once merchants are plugged in, consumers still need to be able to pay, and merchants need to get paid. Stripe makes that easy.
Consumers will only use the UI if it's better. Merchants will plug into the UI to make more money. So everyone wins-- consumers, merchants, OpenAI, and Stripe. And since the protocol is open, other chatbots and other payment processors can implement it too. Who loses?
(I agree that at scale these things tend to accrue to top players and you get all kinds of weird unsavory consequences. But I'd argue that's a critique of our regulatory apparatus, not of the companies building products and services.)
