I concur, open-source will be more reputation based and no doubt, in the future, LLMs can also act as a quality gate.

I work a lot with quants (who can program but are more focused on making money than on clean-code) and Opus 4.5 and Kimi 2.5 are extremely good at giving them architecture guidance. They tend to overcomplicate some things but the result is usually miles better than what they produced without LLMs.

as we're doing anecdotes: I work with quants too

their LLM "assisted" work seems to be the roughly the same quality (i.e. bad), but now there's much more of it

not an improvement