A customer is due a refund. The support agent sends 10x the amount that was due. Who is at fault? Was it the team that set the rules, the runtime that managed the loop, the model that chose the action, or the tool that carried it out?

In this agentic whodunit, we'll follow the evidence through one failed run and uncover where the mistake happened. Solving the case raises a harder question. Who owns the outcome when an agent acts on your behalf?

We'll examine the agentic loop as the unit of ownership, from the runtime and model to the tools, data, and control plane around them. You'll learn what teams need at each layer to inspect what happened, constrain what can happen, and change the system when it no longer meets their needs. Along the way, we'll see how open source infrastructure gives teams the access they need to own the full path from decision to action.