I’ve seen the code current tools produce if you’re not careful, or if you’re in a domain where training data is scarce. I could see a world where a couple of years from now companies need to bring outside people to fix vibe coded software that managed to gain traction. Hard to tell.
Right now I'm creating clinical trial visualizations for biotech firms. There's some degree of complexity because I have to understand the data schema, the specifics of the clinical trial, and the goals of the scientists. But I firmly believe that AI will be able to handle most of that within 5 years (it may be slower in biotech because of the regulatory requirements).
But I also firmly believe that there is more demand for (good) software today than there are programmers to satisfy it. If programmers become 10x more efficient with AI, that might mean that there will be 10x more programs that need writing.