I'm going to throw out my own ignorant theory.

AIs that I find useful are still just LLMs and LLMs power comes from having a massive amount of text to work with to string together word math and come up with something ok. That's a lot of data that comes together to get things ... kinda right... sometimes.

I don't think there's that data set for "use an app" yet.

We've seen from "AI plays games" efforts that there have been some pretty spectacular failures. It seems like "use app" is a different problem.

Are llm testers doing anything traditional scripts with for loops can't?

They should totally be able to. If there's "vibe coding" there should be "vibe testing." We're working on just such a product (https://actory.ai); right now it only does websites but just imagine when we turn it on mobile/apps, etc. How cool would that be?

Because for meaningful tests of an app (assuming b2c or b2b for end users) you are supposed to be or imitate a human being.

Current AI is not even designed to do that. It is just a very sophisticated auto-complete.

It is sophisticated enough to fool some VCs that you can chop your round peg into square hole. But there is no ground to expect a scalable solution.

They are; we're working on agents for web application testing over at qa.tech.

Anecdotally, I know someone who tried to have ChatGPT generate unit tests and it was an abject failure.

I worked with a team that did this for the Facebook app.

https://engineering.fb.com/2018/05/02/developer-tools/sapien...