Transitioning AI Systems to Customers

An engagement with an AI system does not simply stop. Every working engagement ends at some point, and the way it ends is called the handoff. At the handoff, responsibility for the system changes hands, and the work of keeping the AI system running day to day moves from the FDE to the customer's internal team.

This step matters because it is where everything built during the engagement is put to the test. Before reading further, pause and consider the question below, then continue to see the answer.

The important word in this definition is operation. The handoff is not described as a final meeting or a stack of documents being passed across a desk. What moves is the ongoing work of running the system. Before the transfer point, the FDE is the one operating the system. After it, the customer's internal team is the one operating it, and the FDE is no longer the operator.

Notice what does not change. The system itself is the same system. What changes is the group responsible for keeping it working. That is why the transfer point is the moment to look at the system closely: the question shifts from what the FDE can do to whether the system can keep working without the FDE operating it.

The FDE and the customer's internal team hold different roles across the two phases. The source material states clearly that day-to-day operation moves from the FDE to the customer's internal team, and that the FDE moves on to another engagement or to the next project at the same customer. The table below sets out how these roles differ on either side of the transfer point.

The handoff is the closing step of the engagement. Work on the system happens first, with the FDE operating it. The handoff then marks the end of that working period, and the FDE's role in day-to-day operation ends with it. Read the sequence below from left to right to trace the order of events.

The chapter describes loops that run during an engagement. The source material explains that the handoff tests whether those loops were running cleanly enough that the system survives the FDE's departure. In other words, the handoff does not add a new requirement at the last moment. It reveals the state that the earlier loops produced.

The logic runs in both directions. If the loops were running cleanly, the system survives the departure of the FDE. If they were not, the handoff is where that shows. This is why the handoff should be read as evidence about the whole engagement, not only as the last task before the FDE leaves.