Home/Our expertise/Article

Article · AI in production

Where the vet decides: the boundary inside the Yessa copilot.

Yessa is our own practice software for Dutch vets, in daily use in the consulting room. Its copilot works alongside the vet during the consult. The most important design decision was not what it can do, but where it stops. Because it is our product, we answered that question with our own money before any client had to.

Enigmatry engineeringAI in productionPublished 23 September 2026Rotterdam · Belgrade · Novi Sad

What it does during a consult

The animal is on the table, the owner is talking, the vet is examining. The copilot listens and captures: what was said, what was found, what was decided. It structures that into the record the practice needs afterwards, and it supports the vet’s decisions with what is already known: the animal’s history, the earlier findings, the things worth checking before the owner leaves. It prepares. It does not conclude.

Where it stops

The boundary sits at the diagnosis and the treatment. The model may propose, structure and remind. The vet decides, and the vet’s decision is what goes in the record. That is not a limitation we accepted reluctantly. It is the design. A confident wrong suggestion, delivered in front of a worried owner, costs the vet trust she cannot get back in that room. So the copilot is built to be right about what it is sure of, and to be quiet or explicit about what it is not.

The vet sees what the copilot captured, in her words and the owner’s, before it becomes the record. She can correct, remove or override anything. Nothing is written on her behalf without her seeing it.

The questions we had to answer first

What happens when the model is confidently wrong with a client present? What does the vet need to see in order to catch it? What is captured when the recording is incomplete, and how is that shown? Which decision may the model prepare, and which may it not touch? Each of these got an answer in the product, and each answer was tested on real consults in a practice that runs on our software. That is the advantage of building AI into your own product first: the hard questions arrive with your own name on them.

What we took to the clients

The same rule now sits under the copilots we built for others: a fiscal assistant where every answer traces back to its ruling, an AI guide for teachers that stays inside what it is supposed to know. Show the source first, then the answer. Leave the decision with the professional. The specifics differ per profession. The boundary does not. If you are thinking about a copilot for your own professionals, the first conversation is about exactly this: where the person decides.

AI in production

Tell us the workflow. We will say where the person decides.

A copilot that people use mid-workday starts with one question: which answer may it give on its own? We answer that with you, from Yessa, TEX and Kai.

Rotterdam · since 2006 · around a hundred engineers