1. The problem Car-delivery operations teams manage a chain of drivers, ferries, pickup sites, and customer promises. One missed connection can trigger hours of checking, replanning, and coordination. The difficult part is gathering enough reliable information to make the next decision. 2. What we built BahnOS is a control room where agents investigate and people make the judgement calls. It surfaces deliveries needing attention and brings together what happened, the supporting evidence, the proposed action, and its consequences. 3. How it works—and what’s AI about it Our AI agent, BahnBot, uses a language-model tool loop to read delivery records, check booking status, investigate alternatives, and prepare proposals. For example, when a ferry booking times out, it checks whether the provider holds a reservation before proposing another purchase.
People approve consequential changes, and the agent never contacts paying customers on its own. The prototype includes six recorded model investigations using synthetic data and mocked services. We also inject failures into copies of orders to test warning coverage without changing the originals. 4. What’s next Pilot BahnOS with operations staff and measure time to a decision and unnecessary warnings. Expand pickup-readiness and vehicle-warning coverage, then test proposed rule changes against historical cases before asking a person to adopt them.
