It's not conceived as an LLM dependency, but really more similar to the way I currently use LLMs as an engineer: as a consultant. I don't think current models are up to that task on their own, certainly not for someone without significant programming knowledge - but the sense I got, especially as the talk touched on still needing access to a skilled dev for a while at least, was that the author also understands this. Too, in sketching out so broad a vision as this, I wouldn't necessarily expect any author to treat what's possible today as a hard limit, any more than anyone else working in tech does or should.
I also don't understand why you'd consider LLMs as a blocker to local-only. You can run a 7B model on a $600 refurbished M1 Mac mini today. It won't be the fastest or the cleverest, and you probably will need an hour or so from
someone who knows how to set up Ollama and Open WebUI, but it is absolutely usable after that with no Internet connection at all - I have a Mac mini so configured on my desk right now, actually. Or for an order of magnitude more outlay, you can get a Studio capable of supporting multiple concurrent users with 70B models producing tokens faster than you can read them. Again, sure, it won't be GPT-4o, and it won't be up to the standard of the sorts of tools described in the talk - not this year, anyway. In 2026 or 2027, though? Again, this talk looks to a possible future, not just today. And with a setup based on open models running locally as I've described, there is the countervailing advantage that once you have it, you own it.
As for the term 'barefoot developer' itself - it's a cavil, I grant, but I still want to quibble, because as a child in rural Mississippi with an Apple IIc and a couple of books, I was a barefoot developer, writing my own programs to solve my own problems, and often enough going to play in the woods by the road cut when I got stuck. (It helped! These days I'm more likely to chase wasps with a camera, but it still helps, when I can force the time. For what it's worth, though, I do wear shoes these days.)
Now I'm what Amazon would call an L5, earning as highly as almost anyone in this generation of my family - and not for a lot of student debt, because I never went to college. Whatever I am as a professional, I am in large part because I've been discovering and developing my craft since my single-digit days - and was free to do so without my parents having to pay anyone a damned subscription, besides.
True, the technology has grown far away from the kind of locally and individually empowering model this talk describes, and that I recall from my own childhood and early career days - that proto-SPA I mentioned nearby, for example, was hosted on a machine that everyone who ever worked on it could physically touch without leaving the office we all shared. And this in 2011! It really has not been that long a time since a model very much like what this talk describes was actually present in reality.
I see no reason in principle why that can't be true again. Certainly I'm convinced that it should.