Tag
The article discusses how the AI engineering loop can be fully automated but argues that handing over the entire loop produces 'agent slop' due to imperfect evals. It recommends automating certain steps while keeping human judgment for nuance.