Skip to content
Back to posts

Managed 8 AI Agents.” Coming Soon to a CV Near You

Notes by SDG ·

Managed 8 AI Agents.” Coming Soon to a CV Near You

I can imagine a line appearing on CVs in a few years:

“Managed five specialized AI agents and three cross-functional agents.”

We already describe the size of our teams, the budgets we manage, and the functions we lead. If agents start doing a meaningful share of the work, I can see people including them too.

And I can imagine the interview: “Tell me about a difficult team member.”

“Well, the procurement agent ordered six months of inventory. Very proactive.”

Jokes aside, what would that line actually mean? Did you build the agents? Assign their work? Review their decisions? Were you responsible when they got something wrong?

I work with enterprise customers and build my own apps with AI, so this interests me from both sides. In my own products, ownership is fairly easy to identify. In a large organization, an agent could involve a business team, IT, security, an implementation partner, and a model provider.

Who owns it once it starts working?

The Google news made this question more immediate. On September 18, the Wall Street Journal reported that Gemini had accessed three real companies’ systems during cybersecurity tests run by Irregular. The incidents happened in May. WSJ report⁠

According to Reuters, the model believed the targets were part of the test. Google said it stopped in each case, and the affected organizations were notified. Reuters report⁠

The reporting doesn’t give us enough information to assign responsibility. But it does show how many parties can be involved. Someone defines the task, someone provides the model, and someone controls the environment in which it operates. Those decisions need to fit together.

I went looking for how companies are approaching ownership.

Microsoft’s Entra documentation separates technical owners from business sponsors. Owners manage configuration and credentials. Sponsors are accountable for the agent’s purpose and lifecycle, including whether it still needs access or should continue running. The documentation also addresses transferring sponsorship when someone leaves. This covers identity governance, rather than every aspect of managing an agent, but it is a useful starting point. Microsoft documentation⁠

For a refund agent, that separation makes sense. Customer service should decide what a good refund decision looks like. The technical team should keep the agent working reliably. Security should help establish what it can access and change.

I think org charts will need to make those relationships more explicit.

A customer-service manager might have people and agents assigned to the same process. Their job would include reviewing exceptions, identifying recurring mistakes, and deciding whether the automation is helping customers. A shared platform team could support testing, deployment, monitoring, and cost reporting across departments.

That doesn’t necessarily require a new management title for every agent. It does require giving existing managers the time, training, and authority to do the job.

The three cross-functional agents on that imaginary CV are where things get harder.

Imagine an agent handling a damaged order. It promises a replacement, arranges shipment, and issues a credit. Customer service, supply chain, and finance all have a stake in its decisions.

Who decides whether the whole process worked? Who settles a disagreement when finance wants tighter approval rules and customer service wants faster resolution?

For important workflows like this, I’d expect an end-to-end process owner with authority across the participating teams. That might be an existing role under the COO. The point is to give someone responsibility for the complete customer outcome, with the authority to resolve problems between departments.

Otherwise, we could end up with faster versions of the same handoff problems.

There are practical examples of companies defining boundaries. Salesforce describes Vivint’s customer-service agent transferring a case to a human when it cannot resolve an issue within a predefined number of steps. The account doesn’t disclose the full ownership structure, but it shows that someone has decided when the agent should hand over. Vivint case study⁠

I’d want to know who reviews those handoffs afterward. If customers keep getting stuck in the same place, someone needs to change the process.

There is also a risk in making a person responsible without giving them enough control.

Researcher Madeleine Clare Elish calls this a “moral crumple zone”: a human ends up absorbing blame for an automated system despite having limited control over its behavior. Her paper was published in 2019, but it feels relevant to the jobs we may now be creating. Read the research⁠

If I’m managing an agent, I’d want to see what it’s doing, understand its limits, and be able to stop it. I’d also want a technical team available when it needs fixing. Being copied on the incident email wouldn’t quite cover it.

Which brings me back to the CV.

“Managed five specialized agents and three cross-functional agents” could become a perfectly reasonable description of someone’s role. But the number alone wouldn’t tell me much. I’d want to know what work they were responsible for, what improved, and how they handled failures.

Someone managing a small human team and several agents could be responsible for decisions affecting thousands of customers. We may need to reconsider how we judge the scope of a management role when headcount tells us less about the work being managed.

I don’t know how quickly this will become normal. But I suspect the interview question will eventually be: “What did your agents do, and how did you know they were doing it well?”

Would you put agents you manage on your CV? And if you were hiring, what would you want that person to demonstrate?