The person who already owns the process the agent works in. If the head of payments operations answers for payments today, they answer for the payments agent tomorrow. Technology, risk and data protection support that owner, but do not replace them. Record the ownership in writing, with the authority to narrow or stop the agent, before it goes live.
Why not the team that built it?
Because building and owning are different jobs. The build team knows how the agent works; the process owner knows whether its work is right. When something goes wrong, the questions are about the process: was that refund correct, should that case have been escalated, which customers need a letter? Only the process owner can answer those quickly and with authority.
There is also a principle at stake. Accountability for an outcome should not move because the work is now done by software. If the head of claims answered for claim payments when people made them, they answer for them when an agent prepares them. Our view on why operations, not IT, should own agents in production develops this further.
What does the owner answer for, and what do others answer for?
| Role | Answers for | Records |
|---|---|---|
| Process owner | What the agent does: scope, depth, outcomes, stopping it | Signed acceptance, depth decisions, review notes |
| Platform owner | The platform: access, versions, connections, uptime | Change log, access grants |
| Risk | Whether bounds and controls match the firm's risk appetite | Sign-off on bounds and depth 4 decisions |
| Data protection | Lawful processing of personal data | Screening and any impact assessment |
| Sponsor | Funding and whether the agent continues | Business case decisions |
Only one row answers for outcomes. If several people share it, nobody does.
How is ownership assigned and recorded?
- Sponsor: names the owner when the business case is approved; see how to write a business case for AI agents. Evidence: the approved case naming the owner.
- Owner: accepts in writing, including authority to narrow or stop the agent without further approval. Evidence: a signed acceptance.
- Platform: records the owner against the agent in the inventory, and on every depth decision. Evidence: the inventory entry.
- Owner: names a deputy for absence. Evidence: the deputy on the rota.
- Sponsor: reassigns ownership formally if the owner moves role, with a handover. Evidence: a dated transfer.
In regimes that assign named individual accountability to senior people, such as the UK's Senior Managers and Certification Regime, the agent's owner should sit within a senior manager's area of responsibility. That senior manager answers for the business area; the agent is part of how it is run.
What if an agent spans several teams?
Pick one owner anyway. Say a bank's agent prepares onboarding files for business customers, which involves the sales team, the onboarding team and financial crime. The onboarding operations manager owns it, because the agent's work is the onboarding file. Sales and financial crime are consulted on scope and escalation rules, and financial crime analysts review flagged files, but the onboarding manager decides depth and answers for the agent's output.
Where an agent's capabilities clearly belong to different processes, the better answer is often two agents with two owners. The broader method is in deploying an AI agent in a business.
How does this differ by market?
The principle holds everywhere, but local supervision shapes the detail: who must be informed, which roles carry personal accountability, and how outsourcing to a platform is governed. Our sector briefs on AI agents in business banking operations in Nigeria and in India show how ownership sits under those markets' supervisors.
The wider sequence is in our guide to deploying an AI agent.
Where Pryme fits
In Pryme, authority is granted per agent and per customer, and every verdict is filed with the bound that decided it. That gives an owner what they need to answer for the agent: a record of what it did, the depth it holds and why, and the ability to lower it at once.
Questions people ask
Can a committee own an agent?
A committee can oversee agents, but each agent needs one named owner. Committees are good at setting policy and poor at deciding at three in the afternoon whether to stop an agent.
What if the process owner does not want the agent?
Then it is the wrong process to start with, or the owner needs to be persuaded on evidence. Imposing an agent on an unwilling owner usually produces weak review and slow decisions.
Does the owner need technical knowledge?
No more than they need to run any system in their area. They need to read evidence, understand the agent's depth and bounds, and know how to lower them.