Map the work before buying the tool.
Luke Girgis describes breaking roles into tasks and tracing the workflow from order to delivery. His experience points to a useful starting question: where does work stall, repeat, or require an avoidable handoff?
Choose one recurring process. Write down its steps, owner, time spent, and most common failure.
Observe your customers first.
Latané Conant starts with customer conversations. The opportunity is to identify friction before deciding what an AI agent should do. Customer effort and satisfaction give a project a purpose beyond simply deploying technology.
Review five recent customer interactions. List repeated questions and the point where each customer needed extra help.
Define what a good result looks like.
Gal Vered describes giving agents context, a plan, and acceptance criteria. For software, he stresses verifying actual behavior. Apply that discipline to any delegated task: the deliverable must be something a person can inspect.
Write three acceptance criteria for your chosen workflow. Test on a small sample and compare the result with your current process.
Your next move
A one-page pilot brief: one workflow, one owner, a baseline, and a clear definition of success.
Editorial learning path based on the published episode descriptions. Listen to the linked conversations for the guests’ full context.