Seeing a lot of hot takes on the ongoing #anthropic source leak hilarity bemoaning just how terrible the code is.

Folks, you're missing the point. Their idea is that code is now a low-level implementation detail, it's not for reading. You only program with prompts, and nobody cares how hard it is for agents to read their own code. That humans still need to review it is considered a temporary inconvenience, which they are solving (for now) in the usual way: hiring more bodies. (cont.)

Eventually the AI industry believes "we will develop similar practices around unread code"[1].

They don't know how, they just believe.

It might even work out, given the amount of talent thrown at it. But right now nobody knows.

But what I'm saying is that "lol this code's a mess" is not such a slam dunk you might hope.

[1]: https://www.heavybit.com/library/article/write-only-code

Write-Only Code | Heavybit

AI may be accelerating software into a world of "write-only" code that forces a rethink of culture, processes, and tools. Heavybit’s Joseph Ruscio explains.

Heavybit