Who owns the agent.

Ask a leadership team who owns their CRM and you will get an answer. Ask who owns the AI agent that answers customers on WhatsApp at midnight and the room goes quiet. IT set it up. Service uses it. Marketing wrote some of the content. Nobody is accountable when it gives a wrong answer.

That gap is the real risk of agents, more than the technology. An agent with no owner drifts. Its knowledge ages, its limits are never revisited, and its results are never compared with the people it works beside.

Why do agents need an owner at all?

Because an agent does work. HubSpot now describes Agent Hub as “one place to see and manage every agent running across their business”, with agents “all working together and using shared context” (HubSpot). Once agents prospect, resolve tickets, draft content and chase invoices, they are part of your workforce. Work without an owner is work nobody improves.

The owner should be the leader who owns the outcome the agent serves, not the person who configured it. The head of service owns the Customer Agent. The head of sales owns the Prospecting Agent. Finance owns the Revenue Agent. Technical teams operate; business leaders own.

What are the roles in the operating model?

We use five roles. Most organisations already have the people; they lack the definitions.

• Business owner. Accountable for the agent's results. Decides its job, approves its limits and decides when it expands or stops.
• Agent operator. Usually RevOps or the CRM team. Configures the agent, maintains its knowledge and connections, and runs the reporting.
• Frontline team. The people the agent works beside. They review outputs, take handoffs and flag poor answers.
• Risk and compliance. Sets the rules on data, consent and regulated topics, and is consulted before the agent's scope changes.
• Leadership team. Informed through a monthly view of what agents are doing and what they are worth.

What limits should every agent have?

Limits are the agent's job description. Write them before go-live, in plain language, and have the business owner and risk sign them off.

1. Scope. The intents, segments, channels and markets the agent covers. Everything else goes to a person.
2. Actions. What it may read, what it may write back to the CRM and what it may trigger. Reading a record and changing a deal stage are different levels of trust.
3. Autonomy. Whether a person approves each output, a sample, or none. Autonomy is earned by segment, not granted by default.
4. Red lines. Topics the agent never handles: refunds, complaints, health, legal commitments, anything regulated in your market.
5. Spend. Agents in HubSpot consume HubSpot Credits. Set a budget per agent and an alert when it moves. Our HubSpot Credits answer explains how they work.

How should agents be reviewed?

Review is how an agent gets better, and how you catch problems before customers do. We use three rhythms.

• Weekly. The operator and frontline lead review handoff reasons, poor answers and edited drafts. Each one points to missing knowledge, a weak limit or a bad intent choice.
• Monthly. The business owner reviews results against the agent's goal and decides whether to widen scope, add autonomy or hold.
• Quarterly. Risk and compliance review limits, data access and any regulatory changes, such as Malaysia's PDPA amendments.

Keep the reviews short and the decisions written down. A change log for each agent is cheap and saves arguments later.

How do you measure an agent in Agent Hub?

HubSpot says Agent Hub lets teams “see the live status and performance for every active agent” and “manage agent outcomes organized by go-to-market goal” (HubSpot). That is the right frame: measure each agent against the business goal it serves, not against activity.

For every agent, we report four measures to its owner:

• Outcome. The business result: resolved conversations, meetings booked, invoices paid.
• Quality. Customer feedback, edit rates on drafts, or accuracy against a test set.
• Handoffs. How often it passes work to people, and why.
• Cost. Credits consumed per outcome, compared with the cost of the same work done by people.

HubSpot reports strong averages for its agents, such as 39% faster ticket resolution with Customer Agent (HubSpot). Treat those as direction. Your owner should judge the agent on your own baseline, measured before go-live. Our piece on five metrics that prove AI is paying goes further.

How do you put the model in place?

Start with the agents you already run, including the ones nobody mentions. For each, write a one-page agent charter: its job, its owner and operator, its limits, its review rhythm and its four measures. Agree it in a short meeting with the business owner and risk. Any agent that cannot be given an owner in that meeting should be paused until it can.

Then make the charter a gate. No new agent goes live in Agent Hub without one, and no agent's scope or autonomy changes without the owner signing the update. It is a small amount of paperwork that prevents a large amount of confusion.

What changes for the people?

Agents change jobs more than they remove them. The frontline team handles harder cases and becomes the agent's coach. RevOps gains a new discipline: running agents as a product. Managers start reviewing agent output in the same meetings where they review people's work.