Building a worker-evaluator retry loop in LangGraph (and where it bites you on Lambda)

Most agent demos stop at "the model called a tool and gave an answer." That answer is often wrong, and nothing in the loop notices. I wanted an agent that checks its own work against a success criteri