We definitely learned the exact same lesson. Especially if your LLM responses need to be fast and cheap, then you need short prompts and small non-reasoning models. A lot of information out there assumes you are willing to wait 30 seconds for huge models to burn cash, but if you are building an interactive product at a reasonable price-point, you are going to use less capable models.

I think the unfortunate next conclusion is that this isn't a great primary UI for a lot of applications. Users don't like typing full sentences and guessing the capabilities of a product when they can just click a button instead, and the LLM no longer has an opportunity to add value besides translating. You are probably better served by a traditional UI that constructs the underlying request, and then optionally you can also add on an LLM input that can construct requests or fill in the UI.