well one thing I am doing day-to-day now (and also is part of the role) is looking at routine tasks real people are doing and botting it, mostly via livebook + OpenAI API/function calling ad-hoc… and some parts of engineering, especially the more boilerplate stuff with react/typescript is pretty straightforward to automate!
Like give the model some constraints or an example and tell it to create more code → testsuites → docs, and then run a few “how to improve” loops autonomously. I think the sheer learning corpus size for some programming areas is huge and it works well for specific languages, ideally if you can provide a kind of sandbox to run the generated code and feed results/errors back via function calling API.
Still, we want humans here, for the more interesting/creative part, but of course we should let machines do the work whenever possible 
