I agree, but to the extent I’m worried about the future of the profession, it’s because I don’t know how much more the models are going to improve. Today’s models desperately need a real SWE in the loop to do most serious engineering. Will that still be the case in 5 years? I really don’t know
I think integrating libraries into a product will be essentially free. Novel solutions that require an understanding of the problem space to such an extent that unorthodox solutions seem viable is still the limit in my experience. LLMs, at least in my experience, need lots of guidance when the solution you want is not something that has already been done. If you look at it from a business perspective- most software shops have a small proportion of their software that is truly unique and their USP - the rest is not necessarily boilerplate, but definitely not bespoke and unique. This overhead will now be cheaper. Similarly, software vendors that dominate despite not having any bespoke, unique software component will soon have a tougher time. Why would someone pay for atlassian products now? All of their products are not good software- they have market dominance. Replicating their features will be trivial. Anyone who will be sufficiently fed up with paying will DIY the solution.
I feel like once it doesn't need an engineer in the loop, does it need ANYBODY in the loop? at that point the whole company can probably be autonomous.
That seems like it would be such a paradigm shift that the second and third order effects are so crazy and nobody can predict what that world will actually look like
IMO models are getting smarter at software, but not at software engineering.
I can tell you as a job searcher companies are rushing to replace us more completely.
Most of the "legit" jobs coming through my inbox are for grading LLM software tasks. A lot for long horizon tasks.
It'll happen when it happens, but till then I think I'll not be a party to that lol.