The LLMs have no understanding whatsoever just token IDs and their probabilistic relationships with each other. I do not see them replacing 'shittiest' programmer whatsoever. Anyone writing games/software with LLM is the one who actually can code and would have produced same or better output by surely, spending more time than with an LLM.
But if you're clueless average desktop user, you cannot expect LLMs to churn out a full working end to end solution, not yet and probably not ever with the current probabilistic graph fitting foundations.
So, LLMs are surely very useful, marvellous and a handy tool in capable hands, will remain so but that Armageddon is not happening yet that you start calling for six month ban. Imagine, six months. What after six months? No answer.