Why is the customer still doing the orchestration in AI customer service?

In many AI customer service deployments, the customer ends up doing the orchestration: routing between channels, repeating their problem, and choosing the right path. This happens when each AI agent is built in isolation, no system passes context between them, and the human escalation has no memory of what came before.

A customer messages a chatbot. The chatbot does not understand. They are told to call. They call. A voice AI answers with no record of the chat. They explain the problem again. The voice AI sends them to a human queue. The human asks them to explain the problem. By the third explanation, the customer has done all the routing work the system was supposed to do for them.

What people in the field are saying

DCX Newsletter has a piece, "The customer is still doing the orchestration," that names the pattern. It is one of the clearest "AI changed the surface, not the work" critiques in the CX press right now.

How does this happen?

Each AI agent is built and owned by a different team. The chatbot belongs to support. The voice agent belongs to the contact centre. The in-app assistant belongs to product. No system carries context between them, because the seams between owners are nobody's job. The customer is the only one who knows the whole story, so the customer carries it.

Why does no one own the seams?

Org charts have boxes for the agents. They do not have a box for the arrows between agents. Routing, handover, and context-passing belong to no team by default. Without an explicit owner of the handoffs, every AI deployment quietly pushes the gluing work onto the customer.

What does fixing it look like?

Two things. First, give one owner explicit responsibility for the handoffs between agents and to humans. Their job is the seams. Second, instrument the handoffs: every transfer between an AI and another AI, or between an AI and a human, should pass the conversation context with it. A handover with no context is the customer starting over.

What should a CX leader test for?

Take a sample of multi-channel customer journeys and replay them. Count how many times the customer has to repeat their problem. That number is the orchestration work the system is doing. The higher it is, the more of the work the customer is doing on the system's behalf.

Related: own the handoffs in multi-agent setups, the glossary explainer on escalation, and use case 2: checking an order or account status.