Lazer structured the engagement as three one-week phases: discovery and data preparation, build, then iteration with feedback. The first week focused on the unglamorous groundwork that makes agent projects succeed: mapping where the data actually lived, how it was queried, and which workflows fired when. The team found that the single screen brand users most wanted to automate depended on nearly 20 separate data workflows. Calling all of them directly was not feasible, so Lazer simplified the SQL those workflows shared into a form an agent could invoke reliably, without changing the numbers the business already trusted. On that foundation, the build week assembled agent flows using the Retool Agents framework on the company's existing Databricks data layer, creating agents that run workflows and surface insights to brand teams on demand. Claude supported prompt engineering during the build. Alongside the agents, Lazer produced two enablement assets, an agent development guide and a POC agent-and-tools breakdown, documenting the gotchas so the client's own team could build subsequent agents without further help.