A related post on the topic: https://martin.kleppmann.com/2025/12/08/ai-formal-verificati...

AI can really only be as good as the data it’s trained on. It’s good at images because it’s trained on billions of them. Lines of code, probably 100s of millions, but as you combine those codes into concepts, split by language, framework, formatting etc all you loose the numbers game. It can’t tell you how to make a good enterprise app because almost nobody knows how to make a good enterprise app, just ask Oracle… ba-da-bum!

Sure. You're free to throttle your AI speed or whatever. But don't impose that on me.

That has been said about the world in general. Guess what?

So, we're giving up on the Singularity, then?...Good.

It’s called TDD, ya write a bunch a little tests to make sure your code is doing what it needs to do and not what it’s not. In short, little blocks of easily verifiable code to verify your code.

But seriously, what is this article even? It feels like we are reinventing the wheel or maybe just humble AI hype?

AI: urgh, sick of these escort missions

> He would just spot-check the correctness of AI’s work and quickly spin up local deployments to verify it’s indeed working.

I'm not really sure how exactly he get the project done, but "spot-check" and "quickly spin up local deployments to verify" is somehow makes me somewhat unconformable.

For me, it's either unit-tests that hits at least 100% coverage, or when unit-test is inapplicable, a line-by-line letter-by-letter verification. Otherwise your "spot-check" means no shit to me.

> AI always thinks and learns faster than us, this is undeniable now.

Huh? The LLMs we're using today don't learn at all. I don't even mean that in a philosophical sense— I mean they come "pre-baked" with whatever "knowledge" they have, and that's it.