For most of the modern era, leaders have been able to ignore the fundamentals of how their organisations actually work. Operating models were left to accumulate rather than being deliberately designed. Nobody owned them. Coordination across functions was assumed rather than built. And for decades, many organisations got away with it.
They got away with it because people absorbed the cost. Wherever the formal model left a gap, a person filled it with the workaround that was never written down, the judgement that lived in one head, or the informal network that moved information the official channels could not. A parallel system of human effort carried the load the system should have carried, and it did the job well enough that leadership never had to look at what was underneath. Growth forgave the gap. When markets are expanding and margins are healthy, an organisation can carry an extraordinary amount of structural debt without anyone feeling the cost.
That era is over. What matters is understanding why, because the reason is widely misread. AI has not introduced a new set of requirements for how organisations need to work. Every requirement it appears to impose was already there. What AI has removed is the ability to neglect them and still perform.
Three things have always been true about operating models. They were true before the first enterprise software licence was sold, and they will be true after the current wave of technology has been replaced.
The first: operating models need to be designed, and designed as a whole. Every organisation has one, whether or not anyone chose it. A model that was never designed is still a model; it simply emerged from a thousand local decisions, each sensible on its own, none made with the whole in view. That is how an organisation ends up with a system optimised for nothing, where strategy fails at the handoffs, and where a symptom surfaces in one function while its cause sits in another. Coherence across the whole never arrives by accident. It is the difference between a system you chose and a system that happened to you.
The second: operating models need a long-term owner. Functions own their domains and defend them competently. The seams between them, the handoffs, the coordination points, the decisions that cross boundaries, belong to nobody by default. A model without an owner does not hold its shape. It drifts, one local optimisation at a time, until the system as a whole serves no one’s intent. Ownership is what turns a design from a document into a living arrangement that stays coherent as conditions change.
The third: operating models need incentives aligned to the whole. Structure shapes behaviour, and behaviour shapes outcomes. When the measurement system rewards local performance, people optimise locally and the enterprise result suffers even as every individual does exactly what the metrics ask of them. This is why undesigned models persist. Nobody has a reason to repair the whole when they are paid to perfect their part. It is also the fundamental AI treats most brutally, because AI accelerates whatever behaviour a system rewards. Point it at a model that rewards the wrong thing and it will pursue the wrong thing faster, more consistently, and at greater scale than any human ever could. A misaligned incentive does not survive AI. It compounds under it.
None of this is new. These truths have been available to every leadership team for as long as organisations have existed, and most have been able to treat them as optional. The undesigned model still shipped product, usually on time. The unowned model still hit most of its numbers. The misaligned model still grew. The fundamentals were fundamentals in name only, because ignoring them carried no intolerable price.
AI has changed this. AI lands directly on the structure it finds, and it accelerates whatever that structure does. Where decision rights are unclear, it produces insight nobody has the authority to act on. Where data is fragmented, it produces confident answers built on partial truth. Where the real process lives in someone’s head and the documented process is a fiction, it optimises the fiction and breaks the reality. The technology performs as intended. The environment determines what that produces.
And there is another risk that AI introduces. The human effort that has been quietly absorbing the cost of undesigned models, the workarounds, the judgement, the informal coordination, is precisely what many AI programmes are built to streamline or remove. Take the slack out of a system that was running on slack and what remains is the system as it actually is: unowned, undesigned, incoherent at the seams. The organisations discovering this now are not discovering a technology problem. They are meeting, for the first time and at speed, the operating model they always had.
This is why the AI investment story keeps producing the same disappointment. The pilots work, because pilots run in controlled conditions that supply the design, ownership, and alignment the wider organisation lacks. Then the pilot meets the real model, and the returns evaporate. Leadership looks at the technology for an explanation and finds a functioning system, because the explanation was never in the technology. It was in the suddenly unavoidable cost of years of deferred fundamentals.
There is something genuinely clarifying in this, and even something hopeful. Nothing new is being asked of leaders. The requirements in front of them are the ones that were in front of every leadership team before them: design the model deliberately and as a whole, give it an owner with a long horizon, and align what you measure and reward with the outcome you actually want. What has changed is that the work now carries consequence in both directions. Neglect is more expensive than it has ever been, and the fundamentals, done well, return more than they ever have.
For most leaders, then, AI will function as an enforcement mechanism. The fundamentals they were able to treat as optional are the fundamentals that now determine the return.
Yet enforcement presumes something leaders rarely have which is a clear view of the model being enforced against. Only the people who run it hold that view. It exists in the judgement they apply, the workarounds they have built, the coordination they perform without being asked. They are the only ones who can show you where the model as run diverges from the model as drawn, and that divergence is where things will go wrong. No technology can automatically document it. It surfaces only when people are treated, from the first conversation, as the holders of operational truth rather than as the subjects of a change being done to them.
Durability works the same way. A design endures only as long as the understanding beneath it, and that understanding is something only people carry. Built function by function, it produces better-organised silos and a structure that decays the moment the few who understood how the whole fitted together move on. Built alongside the people who do the work, it holds, because the understanding is shared rather than isolated in a handful of heads.
The one part of the work AI cannot do is the part that now decides whether it works at all. That is seeing the operating model clearly, with the people who do the work, and designing it with them. That was always the foundation for success. AI has only made it impossible to skip.