Who owns the agent? Why a named owner keeps it switched on

Unclear ownership, every time

It shows up in every failure analysis and it matches what we have seen. The pilot was sponsored by innovation, built by a vendor, and handed to operations who never asked for it. Six months later nobody is checking the exception queue and the agent is quietly switched off after the first awkward output.

• The agent that reconciles supplier data belongs to the head of procurement.

• Their ops lead checks the exception queue every morning before coffee.

• That is not governance theatre. It is what keeps the thing switched on.

What the owner actually does

Signs off the three lists. Agrees the two error thresholds. Reads the weekly report. Decides when an item moves from propose-only to unattended. None of this is technical. All of it needs someone who knows what a mistake costs and has the authority to change the process.

We will not start a build until the owner is named. It is the cheapest risk control there is.

We will not start a build until the owner is named. It is the cheapest risk control there is.

We will not start a build until the owner is named. It is the cheapest risk control there is.

Why the AI team cannot own it

They can own the model, the infrastructure and the evaluation set. They cannot own the outcome, because they do not run the process and they do not feel the cost of a wrong answer. The moment the agent is measured on outcomes rather than accuracy, it has to belong to the business.

What we learned shipping banking platforms

Every lending platform we put live in India, Vietnam and China had a business owner whose name was on the go-live. The ones that stayed healthy were the ones where that owner kept showing up after launch. Agents are no different. The technology changed. The ownership rule did not.

We help you find the owner before we help you find the model.