Yesterday a friend of mine sent me Geoffrey’s video. I think, but am not really sure, that he tries to sell the future of software development. Or what is it one needs to be a SWE in 2026. I find this very intriguing at first. He spoke of Ralphing, which I haven’t heard up to that point. And again, without consideration I started thinking “is this guy right?”. Read some more about Ralph later on.
Then I slept on it and realized this is another FOMO driven cult personality. Now, I’m not arguing against the capabilities of LLM. They have changed the way I work. However, the more I think about Geoff’s video. The more I realize it has the classic divisive rhetoric of “US vs THEM” with that FOMO sprinkled on top.

Oh boy! You better learn LLM. Learn all the LLM tricks to not shoot yourself in the foot because it’s still just a statistical model. If you don’t. Boy, you’re going to be left behind! Boy! I swear, just half a year and it’s going to replace all the old school developers (whatever that might be).
He doesn’t forget to sprinkle in the HUUUUGE potential to make money. As do all the AI grifters (did you notice he’s selling a course, or will be soon anyway). Just like NFT grifters before. Just like crypto grifters before. How anyone can not see this is crazy to me.

I’m not an AI skeptic. But what LLMs are is not AI. Not yet. We keep saying it’s AI. But it is just a million monkeys hitting on typewriters and filtering on sentences they write that make sense. Yes, it can write Shakespeare. Yes, it can write Nomad. But with complexity, the time it takes to get to that end result depends on the length (LOC) of the result. He argues the cost per hour is below minimum wage. I won’t disagree with that. But has he compared how long it took LLM to write nomad (reverse engineer it to the spec, implement the spec)? Has he actually shown it? Has he had to make changes?
I am still not convinced. I love LLMs for small, tedious scripts and one shot applications that don’t require a robust lifecycle. But doing an MVP was never the hard part. Maintaining it, addressing user feedback, operations reliability. Those are the hard parts. Can LLM help you with them? Sure. Can it do it in reasonable time without human input? I don’t think so.