Marco Arment
@marcoarment@mastodon.social
One of the most humbling experiences for me, as an imperfect meat-coder, has been seeing the flagship AI models *instantly* find and fix bugs in my code that had eluded me for months or years.
Seeing the stupidity (and quantity) of my own bugs makes it pretty hard to continue to believe that AI code is somehow untrustworthy or low-quality.
AI code has its own shortcomings, but bugginess is FAR lower than human code when you use good models, prompts, and architecture.