Harness Engineering and Continual Learning in LangSmith

Short Description: A practical session on building evaluation and feedback loops that help agents improve over time. We'll look at how teams can use LangSmith to connect traces, datasets, evaluations, human feedback, and production observations into a continual learning loop. Full Abstract: This session will explore harness engineering as a practical way to build better agents. I'll talk about how LangSmith can help teams inspect real behavior, turn traces into datasets, run evals, capture feedback, and use that loop to continually improve their systems. The goal is to give people a simple mental model for moving from ad hoc debugging toward a more systematic way of learning from production and making agents better over time.