Why it matters
Both talks are about instrumenting and evaluating agents in production — the harness work that decides whether agent spend is measured or merely hoped about.
The tokenmaxxing angle
Turning production traces into durable memory, and judging agent output with an eval harness, are both levers for cutting the redundant token spend that bad agent loops quietly generate.
From the organizers
Speakers listed: Palash Shah of LangChain on a self-improving agent built on LangSmith Engine, plus Jeff Barg, Vyshu Khota and Soroush Khadem of Clay on their agentic eval harness; talks begin 6:30pm.