AI governance for non-lawyers: what every product and ops team needs before they ship agents

The agent is already in a staging environment. It can read the ticket queue, draft a reply and, if someone flips a flag, send that reply without a person in the loop. The team is pleased. The legal review is booked for a month when someone has time. The flag is the sort of thing that gets flipped on a Thursday because a backlog is ugly.

This is the ordinary path, not a scandal. Product and operations teams move faster than policy, and they should know enough to pause themselves. Governance for non-lawyers is that pause, written as questions a squad can answer before an agent is allowed to act. The longer regime, NIST, ISO 42001, the EU AI Act, sector rules, is how the organisation files the answers. It is not the script for the standup.

Draft, recommend, or act

The first distinction is the one teams blur because the demo looks the same. A draft waits for a person. A recommendation pushes a person toward a choice. An action changes the world: it sends, pays, grants, denies, files or tells a customer something they will hold you to.

Most of the value in the first months is in drafts. Most of the harm is in actions that were described as drafts until the flag changed. Write, on one line, which of the three this system is. If it can act, name the action in verbs a customer would recognise. Send the refund. Close the ticket. Email the client. If you cannot say the verb, you are not ready to ship it.

Questions before the flag

A squad can answer these without a lawyer in the room. They should still show the answers to someone who owns the risk, and they should not ship on a shrug.

1. What data does it read, and is that data allowed in this tool under the rule you already have, not the rule you hope to get.
2. Who can see the output, including vendors, logs and the person who maintains the prompt.
3. What can it do without a person, and what is the blast radius of one wrong action on a busy day.
4. How does a human stop it, and how fast, including after hours.
5. What would you do if the output were wrong in public tomorrow morning. Name the owner of that call.
6. Which people are affected who did not choose to be in the experiment, such as customers, staff in another team, or applicants.
7. What will you review after two weeks of real use, and what result means you turn it off.

Seven answers, on one page, dated, with names. If a question is still we will sort that out, the agent stays in draft mode. Draft mode means a person sends. It does not mean a person is theoretically available.

The human in the loop, specified

Human in the loop is a phrase that hides design choices. Specify the loop. Does the person see every action, a sample, or only the failures. Do they have enough context to judge, or only a suggested button. Are they staffed for the volume, or is the loop a fiction that collapses at peak.

A loop that nobody has time to use is not a control. It is a sentence in a slide. If the volume is too high for a full review, you are choosing a sample and a blast radius. Say so. Cap what the agent may do until the sample looks boring. Boring is what you want from an action that touches customers.

Where lawyers still matter

Non-lawyers can stop a reckless ship. They cannot invent the company's position on privacy, employment, consumer promises or a regulator's expectation. Take the one page to the people who own those risks before an agent acts on anyone outside the squad. Ask them to mark the page, not to write you a framework from a blank document. You will get a faster answer, and they will get a concrete system instead of a hypothetical.

The AI governance workshop is for the leaders who have to turn those marked pages into an operating rhythm: what gets approved, who owns a use, and how a rule stays short enough that a squad will follow it. The workshop does not replace the one page. It stops the one page from being a private note that dies in a ticket.

A rule the standup can remember

Give product and operations a rule that fits in a breath. Drafts can move with a named checker. Actions that affect a person who is not in the room need the one page signed and a way to stop the system the same day. New tools follow the data rule you already published, not a personal account.

Then audit a sample of agents the way you would audit a sample of access rights. You do not need every prompt. You need the ones that act. If the one page is missing, the flag goes back. A rule that lives only in a policy folder will lose to a backlog. A rule that a squad has to answer before they ship will slow the actions and leave the drafts alone.