Im not so sure on the AI coding thing, but who knows. Some human labour seems to be a requirement for that to work, because a lot of stuff has to actually be thought about; you need to reason and know what you client needs to do, and then build to software for it to do that thing.
This is because you actively need to ask/query particular things, from particular places/relays. And its the combination of what to ask for where and when that is non-trivial.
Perhaps after there are a number of working examples of this stuff out there, as well as lots of written content about the ideas behind it, that some AI has enough to latch onto and figure it out. But as it stands currently, its not going to figure it out on its own.
Login to reply
Replies (1)
I agree with your sentiment, but I also want to shed light on the fact that using LLMs, MCPs, templates, Agents and sub agents has accelerated my building time 3-5x times. Using the tools is a skill that needs to be cultivated, its not going to happen overnight.
Also doesn't mean we don't need to know the fundamentals; its actually more important than ever before to be able to code review, understand software architecture and be selectively critical.
Maybe I see it differently because I'm knee deep, but I sure feel a massive difference between Nov '24 and Nov '25. Its not like we're in the same decade anymore. Also I can't walk out my front door without hearing someone talk about tokens casually or LLMs; I live in AI valley.
However, I know for sure, whomever isn't using the tools, these people are most definitely going to be left behind.