In an owner-led company there is no AI governance department. There is a simpler question that used to answer itself: whose name is under this? A note on what advanced technology quietly removes – and how to put it back.

In every company I have worked in, there was one question nobody needed to ask: who signed this?
An offer, a price list, a report for the board, a letter to a client. Somebody’s name was on it. If it was wrong, everybody knew whom to call – and the person who signed knew it too. That knowledge did a lot of quiet work. It made people read things twice.
AI may be removing that question without anybody deciding to remove it.
A summary arrives in a minute. A market overview, a draft contract, an answer to a customer.
It looks finished – well formatted, confident, complete. And because it looks finished, the old friction disappears: the moment when somebody stopped and asked, “is this right, and am I willing to put my name under it?”
Governance without a governance department
Large companies answer this with structures: AI committees, policies, processes, controls, risk registers, a named officer. An owner-led / short-decision-chain company has only a few of those.
It has the owner and a few people who know the business.
I do not think that is a weakness. Fewer people means the question can be answered quickly in one sentence. But it still has to be asked – and in my experience it is asked surprisingly rarely.
When I talk about AI governance with owners, I try to keep it to three questions.
They are the same three I use for any technology: from where, through what, to where.
What is it for?
Not “what can it do”. What is it for, in this business, in one process you can measure?
Turning a possibility into daily practice is painstaking work. Most of the value of advanced technology is not lost in the model. It is lost in the distance between “this could help us” and “this is how we work on a Tuesday”. A tool bought for everyone, for everything, usually ends up being used by nobody for anything important.
What does it work from?
Every AI answer is built on data, and most companies are less ready here than they think.
Before the clever part there is the unglamorous part: data that is scattered, half-documented, or simply outdated.
In the health-data work I am involved in – many hundreds of practices’ live data – most of the time does not go into technology. It goes into preparing the data, clarifying who may use what, and understanding what the users actually want to know, and why. The analytics are the easy bit.
An AI system working from data nobody has looked at will give you answers nobody should trust. Fluently.
Who answers for it?
This is the question the technology removes or blurs, so it is the one you have to put back on purpose.
The person who answers for an AI-prepared result must be able to tell when it is wrong.
That is not a matter of knowing the tool.
It is a matter of knowing the subject – the market, the client, the numbers – well enough to notice that something is off.
A report can be beautifully written and built on figures that are years out of date.
Only someone who knows the field will see it.
So the real AI skill in a small company is not prompting.
It is judgement – and the habit of using it, even when the answer looks finished.
What the law now says – and does not say
Since February 2025 the EU AI Act has required companies that use AI systems to address the AI literacy of their staff. This summer the Digital Omnibus softened the wording: since 27 July 2026, companies have to take measures to support a sufficient level of AI literacy, rather than ensure it, and no specific level is prescribed (White & Case; European Commission).
Some will read this as relief. I read it as a hand-back. The law no longer tells you how much is enough.
You have to decide for yourself and for your business – and you have to be able to explain your decision when something goes wrong.
From where, through what, to where
From where: what data, in what state.
Through what: which process, measured how.
To where: what result – and whose name is under it.
Advanced technology does not remove responsibility.
It moves it – to whoever is willing to sign for the answer.
A question to you:
In your company, if an AI-prepared answer turns out to be wrong tomorrow – whose name is under it?
Gábor
I help owner-led / short-decision-chain tech companies turn advanced technology into revenue. If that’s you, write me: priegl.gabor@gmail.com
Leave a Reply