AI governance in SMEs: who checks the AI when you no longer do?
AI governance is everything in a business that sets out what AI is used for, who is responsible for each AI system, which data it may see, who checks its work and who can stop it. For a business with a team, AI governance and AI agent oversight come down to a few questions per process, and the most important one arrives quietly: what if an AI employee makes a mistake and nobody notices? As long as your team checks every run, it sees every mistake. Once several processes run on their own, nobody checks every run. That's what Autonomy Level 3, Audited, is for: a reviewing role takes over the checking, and you only get what needs you. Below, we show the rules per level as a governance framework, then the reviewing role, its report and the emergency brake.
We call AI systems that take over processes AI employees; others say AI agents. Both mean an AI system with a fixed task that does work in your business.
The four review questions per level as a governance framework
For every autonomy level, we ask three review questions: Who starts the run? Who checks? What do you still see? A governance framework adds a fourth: what stops the process if something goes wrong? Together, the four answers are the rules a process runs by at its level.
- Level 1, Assisted. Start: someone on your team. Check: your team, every run. You see: every run, yourself or through your team. Brake: without a start, nothing runs; anything unclear goes to a person via the escalation path.
- Level 2, Autopilot. Start: the process itself, on a schedule or trigger. Check: nobody checks every run; rules and the report keep it on track, anything uncertain goes to a person on your team. You see: a completion email after every run. Brake: the escalation path; an approval point before every step with no way back; if the completion email doesn't arrive, a person takes over.
- Level 3, Audited. Start: the process itself. Check: an AI manager that cross-checks reports and logs. You see: a morning report with the exceptions. Brake: as at Level 2; the manager reports, a person stops the process.
- Level 4, Autonomous. Start: the process itself. Check: the AI manager; at the approval points, you or your deputy. You see: goals and exceptions, even when you're away for a week. Brake: as at Level 3, with a deputy for every approval point.
These rules apply per process: one process can be at Level 3 while another is still at Level 1, each with its own answers. Level 4 describes what the week-away test checks. None of our processes has reached it yet, not even in Kevin's own business, where the reviewing role from Level 3 is also still on trial.
How to write these answers down on one page per AI employee, with role, hard limits, escalation path, report, review and emergency brake, is covered in An AI governance framework for a business of ten: everything on one page.
The team working with the AI employees is part of the framework too. Anyone using AI systems in a business has to take measures under the European AI Act to support the development of AI literacy among their staff (Art. 4 AI Act, newly worded since July 2026). The law doesn't require a specific level. For tax advisory firms in Germany, their professional law applies on top; what that means is described in What a German tax advisory firm lets run on its own today.
What else a business without a legal department can put in place is covered in AI compliance in a small business: what a business without a legal department should put in place. For your individual case, ask your own legal counsel.
What Level 3 means
On our home page, Level 3 is one sentence: an AI manager reviews its work and bundles the reports, so you only see the exceptions.
The first three review questions in more detail:
- Who starts the run? The process itself, on a schedule or trigger, as at Level 2.
- Who checks? A reviewing role we call the AI manager. It reads and cross-checks the AI employees' reports and logs.
- What do you still see? A morning report with what needs you at the top: approvals, exceptions, processes that didn't run. The rest you can read, but don't have to.
How all four levels fit together is covered in The four autonomy levels: how a process in your business runs on its own.
The reviewing role: review, report, escalate
The AI manager is itself a process employee. It's hired like any other, with a job description and the same five points: trigger, start, work, end, report.
Our Owner Autonomy program material describes its work with three verbs.
Review. It reads every report and log the AI employees produce. It checks whether every expected run happened; a missing sign of life is a finding. It checks handoffs: if one AI employee hands ten cases to another, ten have to arrive there. And if a report says "done", it looks for the evidence in the log. Contradictions are written down as findings, not smoothed over.
Report. The report comes to you once a day, no more often, or you're back to the pile of emails.
Escalate. Anything that can't wait until tomorrow goes straight to the person responsible. If a process's sign of life is missing at the set time, the alert goes to the person on your team who then takes it over by hand. If a step awaits approval, the request goes to whoever approves it. If the same exception shows up several times in a week, it goes into the report under "Unusual".
Just as important is what it doesn't do:
- It doesn't do the actual work. The filing employee files, not the manager.
- It doesn't approve anything. Approving stays with a person.
- It doesn't change other employees' results. It reports what isn't right.
That's why we recommend read-only access to reports and logs, and emails only to defined recipients. Behind this is a rule for every reviewing role: whoever creates something doesn't check it themselves. A role that checks and, in the same move, corrects or approves ends up checking its own work.
What goes into an exception report
For the manager's report, we recommend a fixed structure. What needs you goes at the top, what's for information only at the bottom. That way, you can stop reading after the first section if it says "none".
- Needs you today: approvals and decisions only a person can make. If there's nothing here, it says "none". Then you know someone thought about it.
- Didn't run: every process whose sign of life is missing, with what has already been set in motion.
- Exceptions: where an AI employee was unsure and asked a person, who it went to and whether it's resolved.
- Done: one line per process, numbers only.
- Unusual: what keeps repeating. The same exception three times in a week is a sign that a rule is missing.
Every line points to its source, the completion email or the log entry, so you can look it up without asking. And the manager can't invent a number without it showing when you look.
The emergency brake
"What if it goes wrong after all?" is the question behind every worry about control. By emergency brake, we mean three things, each described separately in the program material. Together, they make sure a process stops before a mistake has consequences.
First: the AI employee stops on its own when unsure. If it doesn't know a variant, information is missing or the match isn't clear, it doesn't handle the case. It writes to a person and lists the case as open in its report. That's the escalation path, and it applies at every level.
Second: every step with no way back has an approval point in front of it. A payment, a submission to a public authority, an email to a client: here the process stops until a person explicitly says yes, every time, for every single case. No answer means no. If a request sits unanswered, nothing happens, and it's back at the top of the next morning's report. Every approval point has a deputy, so the process doesn't stall when you're on vacation.
Third: if a process fails, it goes back to a person. Every process profile says who takes it over by hand. If the completion email doesn't show up, that's the alert. Someone looks first and restarts only afterwards, so nothing gets handled twice.
The manager pulls none of these brakes itself; it makes sure the right person hears about it in time.
Why Level 3 only comes after Level 2
You might think a reviewing role matters most at the beginning, when the AI employee is still unsure. In our sequence, it comes afterwards, for three reasons.
Leadership needs something to lead. When a single process runs on its own, a completion email is exactly right. An extra role reading that one email to you would be effort without benefit. The need only arises once several processes run and the emails pile up.
The manager can only review what's written down. It needs reports, logs and handoffs, which every process at Level 2 delivers anyway. Without them, it doesn't review; it guesses. Putting a reviewing role on top of messy workflows only creates a second source of claims.
At Level 1, someone already checks. There, your team sees every run. A second check by an AI adds little there.
That's why, in the Owner Autonomy program, leadership only comes once several processes are running. The goal then: an AI manager bundles the reports into a morning report and flags exceptions.
An AI manager, too, is trained like any other AI employee. We recommend reading the completion emails yourself for another week and comparing them with its report. Every difference gets trained in. It's only signed off after four clean reports in a row.
AI agents out of control? The real worry: mistakes nobody sees
If you search for AI agents out of control, you quickly land on security gaps, attacks and headlines. That's not our topic, and this article isn't IT security advice either. The worry behind it, named above: mistakes nobody sees.
A loud mistake gets noticed. If a run breaks off, someone notices. The dangerous ones are the quiet mistakes: a document in the wrong folder, a variant misread, a report that says "done" while a case was left unhandled. Mistakes like these often only come up when a client calls.
At the first two levels, this is well covered. At Level 1, Assisted, your team checks every run. At Level 2, Autopilot, the process starts on its own, runs through to the final report and hands anything uncertain to a person.
After every run, you read a completion email: what's done, what's still open, where there was an exception.
That works well with one or two processes. Every additional process brings another email, plus immediate alerts and escalation emails to the team. At some point, you only skim the reports. Then you're the bottleneck again, just more quietly: you no longer have to think, but you still have to read everything. And what you skim, you don't check.
This is where the reviewing role described above comes in. The completion email before it is described in Autonomy Level 2, Autopilot: the process starts on its own and reports back at the end.
What this looks like for us
Kevin has a reviewing role like this in his own business: an AI manager reviews the work of the other AI employees, and that role is still on trial.
In his business, that role is filled by Kai. Kevin describes him publicly on his AI workforce page, and we describe him no further than that.
Kevin gives Kai a leadership assignment with a goal, a review standard and boundaries. Kai then reads what the other AI employees leave behind: reports, logs and the version history of the files. He holds a report's claims against the actual state. If evidence is missing, he writes "not verified" instead of quietly smoothing over the gap. At the end, he condenses everything into one document with the decisions that are Kevin's to make. He decides nothing on substance, approves nothing and doesn't change other employees' work.
Kevin himself gives an example from the tests so far: Kai spotted an incomplete correction pass, an open permission question and an unverified change across the work of several employees. At the same time, he confirmed that another rollout was fully live.
An incomplete correction pass is exactly such a quiet mistake: it only shows up when someone holds the work against the actual state.
That's why we write "on trial" and nothing more. The reviewing role, too, needs someone at the start who reads its reports. Kevin tries new levels in his own business first, before they come to yours.
Runs without me
Every two weeks, one process that runs without the owner: what came in, what the AI did, where a person decided. Two minutes to read.
Worksheet
AI Rulebook on One Page
What your AI employee may do, what it must never do, who keeps an eye on it and what is checked before release: eight fields to fill in and an acceptance checklist. Comes as a PDF by email once you confirm.
Strategy session
In a 20-minute strategy session, you tell us where things get stuck. We'll tell you honestly whether an AI employee is a good fit for you.
Book a strategy sessionFrequently asked questions
Can an AI really check another AI's work?
Yes, if the check is tied to evidence. Holding a report against a log, comparing numbers between two handoffs, noticing a missing sign of life: these are tasks with a clear-cut result. What matters is that creator and checker are separate, and that the reviewing role doesn't correct or approve anything itself.
Does the reviewing role replace IT security?
No. Level 3 answers whether the work was done correctly. Securing access, protecting systems and fending off attacks is a separate job for someone who knows the field. We keep access as narrow as possible: read-only where that's enough, and write access only where the process needs it.
When does Level 3 pay off?
When several processes run on their own and you notice that you skim their reports instead of reading them. Before that, a reviewing role is usually more effort than benefit.
What happens if the manager itself makes a mistake?
The same as with any AI employee. At the start, a person reads along, every difference gets trained in, and every line of its report points to a source you can look up. That's why, in Kevin's business, the reviewing role is still on trial too.
Does your business need this yet?
Whether a reviewing role is due for you depends less on the number of AI employees than on a state: are you still reading, or are you already skimming? On his blog, Kevin lists four signs of that, plus three simpler measures that work first: Does Your Business Need an AI Lead Role?
After Level 3 comes the last test, whether a process keeps running for a week without you: Autonomy Level 4, Autonomous: the week-away test.
Your next step
Control over AI employees comes from reports that someone holds against evidence, from approval points in front of every step with no way back, and from a person who knows when to take over. If you want to know where your processes stand today and what would be next for you, we'll look at it together on a call.