Post

i agree that we shouldn't be lowering our standards to meet whatever llms produce. but if you aren't giving your agents a proper harness then you're doing it wrong. it's trained on median data so it's going to produce median output. learning how to get an agent to produce good code takes a while