This is actually one thing I think will be great as AI coding agents get better. Companies whose main expertise is hardware might start producing better software.
There are so many little bugs in consumer-facing apps that hit the ‘sweet’ spot of being incredible little annoyances that just aren’t worth putting an engineer on for a week to fix, but which are totally worth having an engineer throw an agent onto them.
How? Coding agents are trained on every copy of every tutorial that skips error checking and implements the least resistance path.
I find that the code AI likes to write actually checks for “errors” too often when often you wouldn’t even want to do that. You don’t need to check every dictionary access and come up with a default value for example
I mean I would hope at least one person actually reviews the code before it goes out, but yeah we all know what hope does :)
This is actually one thing I think will be great as AI coding agents get better. Companies whose main expertise is code reviewing might start producing better software.
Yeah, like fixing a annoyance while introducing one or two SEV-1 for sure is going to be great progress.