I can tell you anecdotally as someone who only does green field implementations which are proof of concepts/MVPs to train clients and to hopefully get a larger implementation down the road where we will need to bring in a team, before LLMs got “good enough”, I would have had to staff 1-3 junior engineers to do the grunt work while I did a lot of the higher customer facing, architecture, team lead type work, now I can do it all myself in the same amount of time.
The code that even ChatGPT generates is usually just as well structured using good software practices as I would have done - and I’ve been in the field professionally since 1996.
The only work I do manually is moving functions out into their own libraries and telling it to use my libraries to reduce duplicate code.
It’s going to hurt junior devs the most. I still have to do the higher level design work and tease out requirements.