How a Small Company in Eindhoven Works with an AI Team

People sometimes ask me how a small company keeps up with partners in more than 15 countries. Part of the answer is a team you will not find on our photo wall: thirteen AI agents, each with one job. But the agents are not the most interesting part. The process around them is.

I have worked in personal safety for twelve years. In that time I have seen what goes wrong in this business: an alarm that is handled differently on a Friday night, a help page that no longer matches the product, an answer that depends on who happens to pick up. My background is in industrial engineering and management, and both taught me the same lesson. Mistakes rarely come from one person, or one agent. They come from a process that lets them through.

So when we built our AI team, we did not start with the agents. We started with the process.

Most of the process is plain code

The backbone of our AI team is not AI at all. It is code: predictable, tested and the same every time. Code keeps the board with every task on it. Code tracks the status of every task and every agent. Code decides when an agent is woken, and with which task. Code checks every step against strict guardrails before the work moves on.

The agents only work inside that process, in the steps where they are strongest: reading, comparing, writing, testing the way a user would. Everything around those steps, the order, the hand-overs, the checks, is fixed in code. That is what makes the whole thing reliable and controlled.

Every task is a ticket

Every task is a ticket with one owner, on one board, in order of priority. A ticket says what is needed, why, and what done looks like. When a task is finished, it moves to review, and it is checked before the next step starts. When something blocks a task, only that task waits, and the rest of the work goes on.

Fixed hand-overs between agents

The agents work together along fixed lines. Take a release. Tess, our AI tester, works through the platform and the app the way a user would, on a test environment, and writes up what changed, with screenshots. She even has her own set of physical devices that she controls to perform tests. David uses her findings and the release notes to update our help pages, and Art turns her screenshots into clear help images. Lina, who writes our posts, may only describe what Tess has seen working.

Manon, our AI team manager, plans the order and checks every delivery. Luuk keeps the system itself running and improves it when a step can be done better or more efficient. Changes to code go on their own branch and are reviewed, with tests, before they are merged.

Each hand-over is a standard: the same template, the same checks, every time.

Guardrails, built into the process

• Your data stays on our own hardware. Work that involves customer data or other confidential information runs only on servers we own and manage ourselves. In Europe. With AI models that run locally.
• Rules are checked by code. Before anything is written to a ticket, our local system checks it and refuses anything that is not needed or could contain confidential information. Such as mail addresses, phone numbers and secrets.
• One rulebook. Every public text about our AI team is checked against one fact base. If a claim is not in it, it does not go out.
• A person decides. Nothing an agent writes reaches a customer, a prospect or the public before one of our people has read and approved it.

What this gives our partners

For our partners the result is not a different Cuebly. It is the same people, with more time for them, and work that is done the same careful way at the end of a long week as on a Monday morning. That is what twelve years in this business taught me matters most.

Meet the whole team on our About page. And read how we test every release in Tested before you see it.

Jarno Heijting, founder of Cuebly