Vibe Check: What should an agent remember?

​Chris Gambill, principal architect at Gambill Data, joins with Scott Haines, Staff Developer Advocate at Databricks, to get into where agent harnesses actually are right now and what it takes to make AI tooling and data durable once it leaves the demo stage. The interesting part is everything past the happy path, and we will let the conversation roam across it. Agents that run for a long time, wait on a slow tool call or a data load or a multi-day approval, and have to survive a process dying mid-task without losing state. What state is even worth keeping once persistent memory grows unbounded, and how much an agent should carry from one session to the next. How a system reflects on past runs to consolidate what it remembers. Expect a live demo and a real walkthrough. This is a plan, not a script, so we will follow whatever is most worth showing by the time we go live.