LLM coding, if done right, is an extremely involved process, perhaps more involved than coding it yourself. The professional result is that the code can be produced with significantly higher quality than if I were to code it myself.
It takes significant human effort to thoroughly understand, define, and refine specifications to recursively satisfy all important feedback from multiple LLMs. Only then can the coding begin, and the actual coding then takes just a couple of minutes. Following this, the code has to be thoroughly reviewed, perhaps also slightly refactored. Lastly, sometimes some follow-up issues have to be created to improve the feature further before it is shippable.
As with anything else, the quality of the result is proportional to the level of effort you put in.