They do. Most economically viable software already exists today, most people's job is to make .1% improvements, and features on top. For this work it's obviously pivotal that any change should not worsen the existing software, as that instantly nullifies the value of improvements.

Another kind of highly valuable software is a new design which digitizes some process, like the ticket management of a rail company. If you don't follow the internal processes and workings of the firm exactly, or you do not interface with existing systems 100% accurately, your software very soon becomes worthless.

Figuring out how to create a piece of code that solves the exact problem the customer has, while not breaking anything, and fitting into existing operations seems to be a big head scratcher still, and something humans still need to do, at least that was my experience so far with LLMs.