Well AI is going to write all your code and it is going to test it all and when you try to use it on your own it is going to break and almost nothing is going to work. So how is it possible that tests are being passed but user flows are a mess?

Well, this is why you need non-linear tests. Basically, figure out which will be the four five workflows that you wanna test for, and then spin off multiple variants of same workflows that are jumbled up in sequence of actions. For example, if you are making a video editor, then maybe a workflow can be to cut some footage and render it. And in that workflow, you might have to move some things, cut some things, redo, undo some things, adjust the edges a bit. So. These are like main actions, and what you are supposed to do is create tests where the action sequences are not same every time. What this allows you to do is to catch bugs which the tests you are AI normally writes can’t catch.

You have to spread a wide net of tests so that you can catch what is wrong. Whatever AI will write is like a single poke in a jute sack. Of course, it’s going to pass through. You have to layer things up by making them non linear

It’s like you are supposed to put pressure on a lot of points at once, at different orders, to see if your software survives.

So make the workflows, decompose all the actions and make a lattice of those actions and stress test the system properly multiple times in various orders.

Sometimes make unusual patterns of action so that you can be more confident about what your app can do and have proper logging so you can figure out what’s going wrong.

For example, if your app can have a starting point, three actions, and an ending point, then explore those five actions as many times as you can in different orders. Maybe open a file, save when you have not done anything. Does it give an error? Maybe do one thing and then try to save. Does it give an error? Maybe after saving, try to open it and try to do the three actions. So it’s like You have to smartly figure out which ones to go for and then test for those cases. But you cannot just have a single line of action and approve the whole workflow. It is under testing that makes you approve a workflow even though you have not sufficiently tested it.