Post

I am currently working on a project I was given to fix bugs in. It was developed at first manually in the pre-historic era of no-LLM but then the maintainers went full AI mode. LLMs can't do good software engineering yet. They had DATA in migrations and the base migration is a pg_dump.

Maybe I hang out with the wrong folks :-) but I see a LOT of hate for "LLM slop" code (from people who use AI!); you *have* to write your own README and probably should mention that you use AI contributions.

The apps you liked and used before are also using AI to improve; so unseating them is not harder per se, but it's not a competition between new killer apps and the old; the old are evolving and becoming better at the same time, with a head start. Anyway idk about this analysis man

Alternatively we’re seeing exactly how much of the software industry is defined (/constrained) by powerful network effects and moats that inhibit competitors and upstarts to such a degree that ~no amount of excess execution capability matters. Even small networks like Twitter just coast.

As writing Code becomes cheaper; aligning humans eats up a larger fraction of the total development cost. Making people who are good at this (empathetic managers) more valuable. And lowers the value of people who play well with machines but not people.