Who answers for it when your agent writes something wrong into a customer record?

Guillermo Rauch runs Vercel. Almost 1,000 people there use one internal company agent. Here's what he called the real work of running it:

"You have to be really really really thoughtful about it. That's literally our new job."

Not the model. Not the prompts. Deciding what the agent can read, what it can change, and who says yes before anything ships.

Now, the part that should bother you.

Schellman surveyed 525 US enterprise practitioners in July 2026. 74% think their organization could pass an AI compliance audit today. 27% actually run a mature governance program. That 47-point gap is a job nobody got given.

So here's the five-second test.

Something at your company touches a customer record. An agent, a bot, an automation, a Zap somebody built in 2023 and forgot. When it writes the wrong thing, who answers for it? Not who fixes it. Who answers for it?

If you need longer than five seconds, you just found the gap. Most of us do.

A field note from our own stack this year. A draft sync was set to update 14 contact records. The agent's summary said exactly that: 14 records, contact details refreshed. The diff said something else. The contact owner field was changing on all 14 because a mapping pointed to the wrong column. Nothing was broken, so no error was fired. Forty seconds of reading the actual changes killed the sync.

That's the whole job, performed once.

I wrote up the job description, who should hold it, and the smallest version a 10-person team can run this quarter: https://www.promptmetrics.dev/blog/governance-is-the-new-job

Your turn. Drop the role, not the person's name.

Who owns this at your company right now? And if the honest answer is "nobody yet," post that.