Ask what the agents may do alone, what they hold for a person, who decides on held work, how reviewers are measured, what happens when an agent gets something wrong, and how your own authority is granted and kept. These questions show an interviewer you understand governed agents. They also tell you whether the job is real oversight or rubber-stamping.
Why do these questions matter to you?
Two jobs can carry the same title and be very different. In one, you review a manageable flow of uncertain cases, your decisions are recorded, and your corrections change how the agent works. In the other, you approve hundreds of items an hour, nobody reads your corrections, and you carry the blame when something goes wrong. Your questions in an interview are how you find out which one you are walking into.
They also signal competence. A candidate who asks "at what depth does the agent act on refunds?" is talking like someone who has done the work.
Which questions should you ask?
| Question | What a good answer sounds like | Warning sign |
|---|---|---|
| What may the agents do without a person? | Specific actions, with limits | "Pretty much everything" |
| What gets held for a reviewer, and why? | Named thresholds and categories | Nobody quite knows |
| Who decides on held work, and how fast? | A named role, a service level | "Whoever is free" |
| How is a reviewer's accuracy measured? | Sampling, calibration, a standard | Measured by volume only |
| What happens when an agent gets something wrong? | It is recorded, the cause is fixed, authority may be reduced | It is quietly corrected |
| How does an agent earn more authority? | Evidence from real cases, re-checked over time | A setting someone changed |
| Is every decision recorded, and can I see the record? | An append-only record, reviewers can read it | Logs only engineers see |
| Do approval rights depend on a current certificate? | Yes, or a plan to adopt it | No idea what is meant |
| How are agent sources kept current? | A named knowledge owner | Nobody owns it |
| What training do reviewers get? | Structured training and a check | "You'll pick it up" |
You will not ask all ten. Pick three or four that match the role.
What do the answers reveal?
Listen for names and numbers. "Refunds over the customer's limit are held for a team lead, who decides within four hours" describes a real control. "The AI flags things sometimes" does not.
Listen also for how the employer talks about mistakes. A good one treats agent errors as information: the case is fixed, the cause is found, and if the agent's certification fails it drops to a lower level of authority until it earns it back. A poor one treats the reviewer as the place blame lands.
Finally, listen for your own authority. If you will approve payments or decisions affecting customers, ask how your right to approve is granted. Some businesses make approval rights depend on a current certificate. That protects you as well as them.
An example
Take a candidate we will call Marcus, interviewing for an operations reviewer role at a payments firm. He asks what the agent holds for review. The manager explains that payments to new payees above a set amount are held, and that any payee details arriving by email are treated as external and can never authorise a payment on their own. Marcus asks how reviewer accuracy is checked. The manager describes weekly calibration on a sample.
Marcus now knows the controls are real, and the manager knows Marcus understands them. Both learned something useful in two questions. For a ready list you can adapt, see the questions to ask in an AI operations interview template, and for the other side of the table, what AI operations managers look for when they hire reviewers.
What should you avoid asking?
Avoid questions that suggest you want the agent to do the work for you, such as "will the AI handle most of it?" Avoid asking about pay before you understand the role. And avoid technical questions about models unless the job is technical; they rarely tell you how the work runs.
If you already work somewhere that is adding agents, the same questions work internally. Asking your employer to train you for AI agent work explains how to raise them with a manager. The guide to AI agent jobs and careers covers the wider picture.
Questions people ask
Will these questions seem too technical in an interview?
No. They are about how work is decided and checked, which any operations manager will recognise. Phrase them plainly and they come across as prepared.
What if the interviewer cannot answer?
That is information. It may mean the agents are new and the role will help shape the controls, which can be a good opportunity if you are told so openly.
Should I mention my certificate when asking?
Only if it is relevant to the answer. Asking whether approval rights depend on a current certificate is a natural moment to say you hold one.