As long as it is used by individual users, apart from hallucinations and inconsistent responses, it is quite good and sometimes arouses the “WOW!” effect.
But what happens if we try to put it into production, using the API, in a professional application, with true multithreading (not that of pyThon, to be clear) where tens, hundreds, thousands of interactions are necessary?
Well this happens: HTTP/1.1 429 Too Many Requests
Not to mention the variability of the answers, i.e. the inability to ALWAYS respect the directives (it does what it wants).
AI at the current state of the art is a tool, not a full autonomous agent solution, it is a force multiplier and productivity enhancer.
GIGO rules with AI, poorly defined requests lead to unintended consequences, the same as if you told a junior dev to “make me an app that looks like Instagram” if you were expecting a fully compliant, secure, industry standard result from a single input then you will be disappointed.
If you build the team to make a product and you enhance them with AI, you get two main benefits, you can do more with less, and faster.