On the one hand, OpenAI seems to have shown a lot, but in essence — nothing new.

Apps in ChatGPT are essentially the same GPTs but with an iframe. Previously, authorization was through links, now it’s directly in the chat. GPTs turned out to be useless for both users and businesses. It’s unclear why Apps would become more useful. I hope they at least remove the warning window about requests going to a third-party site.

AgentKit is an answer to similar products from Microsoft, Google, and Amazon. And here it’s unclear — every week we get news about a new model release that outperforms all previous ones. Why would I use a platform with obvious vendor lock-in for production if the company that owns it doesn’t make it a core focus of their business and, therefore, doesn’t guarantee its support and development? It could be considered a backend for Apps, but that’s not available now either. What remains is prototyping.

I’ll have to try it, of course, but I’m not particularly impressed this time. Quite the opposite: they showed nothing new or cooler than the competitors. Perhaps this is a conscious strategy to not lose focus on the main thing and just occasionally show something new, but those Claude-powered apps look much more breakthrough than ChatGPT Apps.