Agent reliability is an infrastructure problem, not a prompting problem

I spent two weeks instrumenting the agents that do QA work on my personal trading research repo. Every tool call is logged. Every run is… Continue reading on Medium »