[noufalibrahim]

One point where, I think, the analogy fails is context.

If one wants to modify a code base, it's necessary to be able to, sort of, load the program into ones head and then work off a mental model. The "slowness" of traditional development and the tooling around it gave people enough time to do this and over time, get really good at a navigating and changing a code base.

With LLMs being able to generate huge amounts of code in a short time, this is missing. The LLM doesn't fully know what it generated and the nuances. The developer doesn't have the time to absorb all that so at the end of the day, you have something running which nobody (including the original AI author) really understands. That's risky.

Of course, there are ways to mitigate and handle this I don't know if the original analogy is missing this.