Blog

AI in the workplace of a small team: who starts, who checks, what you still see as the owner

What does AI in the workplace look like when it isn't just a chat window but actually takes over work? Who kicks off which workflow, who looks over it, and what does the owner get to see of it? We show this here with a real example, a tax advisory firm with two AI employees, from the perspective of the AI employees and of the office team. We also describe what an owner should see at this level, what changes for the team, and which worries deserve to be put honestly on the table.

Two kinds of AI in the office

In many offices today, AI at work means a chat tool. Someone on the team opens it, asks a question, gets a text back and keeps working with it. That's helpful. But that alone doesn't make the workflow run on its own.

That's why we distinguish between two kinds. Thinking partners are AI employees that someone works with in conversation, steering every step themselves. Process employees take over a fixed workflow, start on their own and send a report at the end. This article is about the second kind, because only that kind changes who on the team does what, and when.

How to get started with AI in a small business in the first place, including with chat tools, Kevin has described on his blog: Bringing AI to a Small Business.

A day at the tax advisory firm, from two perspectives

The following day is put together from what has been set up at the firm. It isn't a log of one particular day. Both processes are at Autonomy Level 2, Autopilot, as of September 2026.

From the perspective of the AI employees

The AI employee for filing starts every day on a fixed schedule and retrieves the new digital assessment notices. That can be up to 30 to 40 a day. It reads each notice, recognizes the type of assessment, fills in the filing fields and files it in the firm's document management system. For this, it knows around 45 rules for the different types of assessment, types of tax and types of letter.

The second AI employee checks the firm's inbox every 30 minutes. It spots the emails with paper notices and hands them over to the first, which files them in the same way.

If the filing employee comes across a notice it can't match with certainty, it doesn't file it on a hunch. It sends the notice to the office team by email and lists it as open. At the end of the run, it sends a completion email that shows what's done and what's still open.

From the perspective of the office team

Before, the office team filed every digital notice by hand, about 2 minutes per notice. On top of that came up to 2 hours for the notices on paper, depending on volume.

Today, the office team only gets what the AI employee can't match with certainty. The office team decides the case, just as it did before. That's followed by a feedback session in which the new variant is trained in. Next time, the AI employee knows it.

So the office team's work has shifted. There's less typing and filing. Deciding, training and handling the exceptions have stayed. The expertise about which notice goes where now also sits in written-down rules.

And what the owner sees: the target picture

This section doesn't describe a day at the firm. It describes how Level 2 is meant to work for the owner in our program. The goal of the 90 days: your process starts on its own and reports back with a completion email.

For you as the owner, a day at this level is above all a day with reports. After every run, the report shows what's done, what's still open and where there's an exception, without anyone having to look through the filing system. Without a report, you end up following up yourself, and you're back to managing every step. That's why the goal is: you read the report and only step in where you're actually needed.

Who in your business gets the reports, you decide for each process. That can be the owner, the person on the team who looks after the AI employees, or both.

Who starts, who checks: depending on the level

The day above applies to Level 2. At the other levels, everyday work on the team looks different. The level applies per process, not to the whole business.

Level 1, Assisted. The AI employee does its task when someone starts it, and your team checks every run. In day-to-day work, that means someone on the team kicks it off and looks at every result before anything moves on. That's how every process starts with us. How this works in detail is described in Autonomy Level 1, Assisted: when your team checks every run.

Level 2, Autopilot. It starts on its own on a schedule or trigger, runs through to the final report and hands anything uncertain to a person. The team gets the exceptions, the owner gets the report. More on this in Autonomy Level 2, Autopilot: the process starts on its own and reports back at the end.

Level 3, Audited. An AI manager reviews its work and bundles the reports, so you only see the exceptions. In the program, that's a goal after the first process: one morning report instead of many separate emails. In Kevin's own business, this reviewing role is still on trial.

What goes into a good report

In everyday work, the report is the place where the team, the owner and the AI employees meet. That's why it's worth settling its structure once. We recommend three parts, in this order:

  1. What needs you. Cases waiting for a decision or an approval. If there's nothing here, it says "none". Then you know someone thought about it.
  2. What's open. Cases that went to a person, with that person's name and where things stand.
  3. What's done. One short line with the number of cases handled.

That way, anyone can stop reading after the first part if there's nothing in it. The team sees in the second part what's sitting with them. And the owner sees in the third that the process ran. If the report doesn't arrive at all, that's a message too: the run didn't finish, and someone takes a look.

A report nobody reads doesn't help. If you notice that you're only skimming the reports, that's a sign there are too many of them. Then the next step is a reviewing role that bundles them.

What changes for the team

Approving instead of typing. The goal on our home page is: your team only approves. The work shifts from getting things done to deciding the cases that need a decision.

The variants stay with the team. Every new variant the AI employee doesn't know lands with a person. Whoever decides it then teaches it to the AI employee. The knowledge stays on the team and gets written down along the way.

A new task: looking after the AI employees. One person on the team looks after the AI employees together with the owner. That person joins the weekly 1:1, knows the emergencies and keeps the rules up to date. How to hand over a role like this, including authority and responsibility, is covered in Delegating tasks without them coming back to you as questions.

Covering for each other gets easier. When a process runs at Level 2, the routine work no longer piles up just because someone is on vacation or sick. The escalation path itself still needs a deputy: who gets the uncertain cases when the responsible person isn't there? Until someone answers, the case stays where it is and appears in the report.

The worry on the team: will this replace me?

This question comes up, and it deserves an honest answer rather than a reassuring line.

An AI employee takes over routine work: what comes up often, follows fixed rules and simply has to get done. What stays with the team are the cases that need a decision, the approvals and the knowledge about the variants. At the tax advisory firm, the office team still decides the uncertain cases and teaches the AI employee new variants.

What we can't promise is how you use the time that frees up in your business. That's your decision as the owner. We recommend telling the team early and involving the team from the start. In our program, the process map is drawn up area by area together with the team, and the person who does the process today sits next to the AI employee during training.

Whoever does the process today knows the variants best. That's why, with us, they sit next to the AI employee while it's being trained.

What this looks like for us

The tax advisory firm above is our client case. Two AI employees, two processes, both at Level 2 as of September 2026. It took 2 weeks there from the first round of watching to sign-off. That's what happened in this one case, not a time promise for your business. It was signed off after four clean runs, and after that it was monitored side by side for a while.

We describe the whole case with all its steps in What a German tax advisory firm lets run on its own today.

In Kevin's own business, 14 AI employees are at work: content, SEO, video editing, bookkeeping prep, leads and email. Each one has a personnel file, a fixed role and clear limits. An AI manager reviews their work; that role is still on trial.

Kevin's everyday work with them: he reads reports, makes decisions and approves anything that goes outside the business. The AI employees may read and write drafts without him. None of the processes there has passed the test of a week without him yet either.

Frequently asked questions

Does my team need to be good with technology?

Nobody needs to know how to program. What matters more is that the person who does a process today can explain how they do it, and that someone on the team looks after the AI employees with you. Where the AI employees run and who sets up the access, we settle in the first sessions.

How much of my team's time does this take at the start?

More at the start than later. The team shows how it does the process today, sits next to the AI employee during training and checks the first runs. In our program, you and the person on your team come to the 1:1 every week for 60 minutes and complete the agreed tasks before the next session. How much that is depends on the process.

Who may start or stop an AI employee?

You decide that, for each AI employee separately. We write it on a page of rules, together with its role, limits, escalation path and report. A template for this is in An AI governance framework for a business of ten: everything on one page.

What happens if the AI employee fails?

Then the completion email doesn't arrive, and that's the alert. Every process profile, our word for a process description, says who takes the process over by hand. First someone looks, and only then is it restarted, so that nothing gets handled twice.

Your next step

In a small team, AI in the workplace means: routine work starts on its own, exceptions go to a person, and the owner reads reports instead of following up. If you want to picture what that would look like on your team, bring a workflow that costs time every day. On the call, we'll walk through it together and work out whether Owner Autonomy is a fit for your business. That's also where we go over how it works and what it costs.