You have Databricks. Your executives now want agents. Between those two facts sits a handful of decisions most teams make by accident: where agent assets live, who can write to production, how you know an agent is good enough to ship, and what the second agent inherits from the first. Get them wrong and every agent is a bespoke build. Al Ghurair is working through exactly this. Phil Chan, SVP of AI, opens with what he inherited: multiple agents spread across separate vendor stacks, no shared registry and no evaluation baseline. This fireside chat will leave you knowing which of these to settle before agent one, which are expensive to retrofit, how to make governance something your CI/CD pipeline enforces rather than a review board, and where simple RAG and agent loops start to run out.