Why understanding, ownership and control matter more as AI takes on more of the work.
At the Milan Leadership Forum on 30 September 2026, Yuval Noah Harari called AIs “native bureaucrats.” The point is easy to miss and hard to forget. Organizations run on words: proposals, contracts, policies, the record of what was promised to whom. No person remembers all of it. An AI can.
That is where AI is most at home. It is also where people can quietly lose track of what is being said in their name.
Harari's deeper worry is that we may end up living inside systems we no longer understand. He was talking about finance, law and government. The same thing can happen inside a single company, one prompt at a time.
So the question I keep returning to while building Gyrium is a simple one.
Imagine an employee preparing an important proposal. They need information quickly, so they ask a chatbot.
Almost at once they have a convincing answer. The language is polished, the structure makes sense, and the result looks professional.
But where did the information come from? Does it reflect what the organization knows today? Is it consistent with what colleagues have promised elsewhere? And who is responsible for checking it?
The employee may not know. The organization may never find out.
Multiply that by hundreds of employees, thousands of prompts and countless documents. Facts diverge. Language shifts. Promises stop matching. Knowledge that should belong to the organization stays scattered across conversations, files and individual minds.
Nobody chooses this. Yet the organization slowly drifts away from its own understanding.
In 1983, the psychologist Lisanne Bainbridge described what she called the ironies of automation. The more of a process a machine takes over, the less practice people get. And the less practice they get, the worse they handle the moments that still need them: the unusual case, the moment the machine is wrong.
She was writing about industrial control rooms. Today the control room is the inbox. Recent research from Microsoft and Carnegie Mellon found that knowledge workers who placed more trust in AI reported applying less critical thinking to its output.
The convenience is real. So is the cost.
Truth. A convincing answer is not a correct one. Organizations need to know where their information comes from, whether it is current and who has checked it.
Consistency. Every communication shapes how an organization is understood. When hundreds of separate AI conversations produce different versions of the same story, that identity begins to fragment.
Cost. Without a shared foundation, people solve the same problems again and again. Work is written, corrected and rewritten. The apparent speed hides the duplication.
Reputation. Customers do not hold an AI accountable. They hold the organization accountable. Its credibility rests on everything it says being accurate and coherent.
Understanding. This is the one that matters most. If nobody can explain why something was produced or what it rests on, who really owns the work? It is Harari's warning at the scale of a single company.
The usual answer is to keep a human in the loop. That is not enough. A person approving every step does not scale, and an approval given without understanding is not control.
Control means the organization decides what is true. It sets the boundaries. It knows who answers for what. And it can see how anything was made. With those in place, the work can move fast without moving away from the organization.
The point is not to have people supervise every action. It is to give them the understanding and the authority to decide what the work should be, to notice when something is wrong, and to correct the source. A correction made in a conversation disappears with it. A correction made in the foundation holds for every piece that follows.
This is why I built Gyrium.
Gyrium is not another autonomous agent designed to take work away from people. It is a platform where an organization owns and curates what it knows, its brand and its imagery, and where that foundation is put to work in everything it communicates.
Go back to the employee and the proposal. In Gyrium, the same request starts from the organization's own facts. The facts used to build the proposal can be traced back to their sources. The wording is the organization's own. The person who sends it knows what it rests on. And what they correct goes back into the foundation, so the next proposal starts from a better place.
People establish what is true. They maintain the knowledge, set the standards and decide what goes out. Each piece of work adds to a shared understanding, and that understanding grows more valuable over time.
The future of AI should not be measured by how much human work it removes. It should be measured by how much more capable people and organizations become.
Can employees make better decisions? Can they understand their organization more deeply? Can they speak for it with more confidence? Can the organization keep and grow what it knows, and stay able to explain it?
Harari closed his talk with advice to leaders: find the time to take care of your mind. The systems we build should make that easier. Not harder.
Every project should leave the organization knowing more than it did before.
Henk van Dijke, founder. More about me
If what your organization knows is what it sells, the pilot is where to start.