Ask HN: Hype aside, what practical issues come up when building LLM-based apps?

Prototyping an application with an LLM is pretty straightforward task these days: try out a few prompts and paste in some context and see if it seems to work.

After trying to build something more substantial (a script that takes a text description and attempts to scrape that info out of a collection of PDFs/websites), I realized there are a number of annoyances with getting this out of the prototype stage:

* Parsing prompt responses to ensure they match my expected schema (e.g. I want XPATH selectors, sometimes the model hallucinates a DOM id)

* Hacks to avoid long context windows (especially if the context isn't easily vector-searchable, e.g. a DOM tree)

* Retry logic

* Measuring how well the system is doing over multiple examples

AI Twitter is full of examples of how LLMs, AutoGPT, etc. are cure-alls, but what are some of the practical issues that actually come up when you try to build on top of these yourselves?