It can do several QA tasks like unit testing, but it can't automate the testing of a website because it doesn't have an understanding of the whole system. This, the lack of strategy and so on, is what shows that we are faaaar away from AGI. And maybe an AGI is a guy that perfectly knows the manual, but still has to deal with bugs and all kinds of weirdness, fuzzy states and signals, delays and the halting problem, etc.
AI software for intellectual tasks. Robots for physical tasks.
Isn't this obvious?