I've found GPT to be useful for prototyping software. It isn't a polished product, but saves lots of typing. The down side of GPT is the LLM is out of date. Nearly 2+ years old. Software frameworks evolve daily. It is very hard to keep up. After GPT gives a first approximation, there is still refinement necessary. In the software world we enjoy some automation from tools like SONAR that apply strict rulesets and scan security databases to ensure the code is correct and safe. Less subjective than a human editor and ruthlessly strict. Even so, there is still some professional tweaking e.g. applying CSS to a web page to achieve a pleasing layout, typography and colors. There are automated tools that hammer on the CSS to ensure compliance with ADA standards.
The party isn't over until the code goes to the test group and gets hammered by the test protocols. Hacking is always a concern, so I use tools like ZAP, BURP and Fortify to look of avenues of hacking. After all the fine details are resolved, the final approval comes from the customer.