Most companies hired a chef, but are still cooking themselves.

Looking at how AI has entered organizations in the past four years, we often use the analogy of a kitchen chef. The same chef was hired to run every company kitchen. This chef has read every cookbook ever written, and could describe any dish perfectly. The downside? The chef could actually not make the dish.

Therefore, for two years, we asked the chef questions and we did the cooking ourselves. Useful, but not necessarily transformational. Yet, most companies are still cooking today and continue to approach AI like a search engine.

A chatbot used to produce text for a person to act on. Nowadays, an AI agent takes a purpose, works out the steps to achieve it, pulls what it needs from an organization's systems… and acts. In doing so, it can perfectly prepare a quote, process a claim, place an order, ... it can cook the dish.

Four in ten companies above one billion in revenue are scaling AI agents nowadays, up from roughly a quarter a year earlier (McKinsey, State of AI 2026). What most organizations struggle with today is the mise en place: writing down the recipe well enough, so the chef knows how to cook the dish. During our latest Craftzing executive roundtable, we explored what a proper mise en place looks like to move AI from generating knowledge to performing the action.

Stop automating existing tasks. Start redesigning the work.

Think of an AI agent as a new colleague. Just like any other capable new hire, an agent also needs to know at least three things before taking real responsibility:

• how your business actually works;

• what is true right now;

• and what happened the last time a similar case came along.

Most companies have written down the first, although often poorly. In many cases, the other two live in people's heads and in systems that have never spoken to each other before. AI adoption, and delegating tasks to an AI agent, already stalls there. What exists only in someone's head cannot be handed over to an agent.

Writing the work down, however, is not the same as redesigning it. Eight in ten respondents say AI has made them personally more productive, while only 37% of companies see any contribution to EBIT. Among the small group capturing real value, nearly three quarters fundamentally redesigned their workflows (McKinsey, State of AI 2026).

The kitchen shows why. A chef who cuts vegetables five times faster does not get dinner out any sooner if the plates are still being prepared, checked and sent the way they always were. One task got faster. The process around it did not.

What is the largest irreversible action you would let an agent take without a human in the loop?

The moment a system acts in your name, the risk changes. A chef who talks nonsense in his kitchen is embarrassing. A kitchen chef who cooks nonsense and sends it to table four is a liability. A model that gives a poor answer costs ten minutes. An agent that acts on one sends it to a customer, a supplier or a regulator.

Agents also change your cost base in a way budgets rarely show. They trade the cost of executing a task for one of verifying it, and only one of the two appears in your plan. Checking a task is a responsibility of your most experienced people. Budget for it, and delegation scales. Overlook it, and your senior staff become the bottleneck in front of a system that never tires.

So the question that belongs on the board agenda is: what is the largest irreversible action you would let an agent take without a human in the loop? A payment, a price, a contract renewal, a customer commitment? The answer sets the pace for everything else.

Don't just pilot agents. Own how the work gets done.

Those choices need an owner. If you'd ask who is supposed to build and maintain all this, from the explicit processes to the oversight of a growing group of agents, the honest answer would probably point to a team that does not yet exist.

IT owns the systems. The business owns the work. In between sits the space where agents either become an operating capability or stay a collection of pilots. That space needs a named owner with real authority: someone who decides what gets written down, what agents may do, and when a pilot becomes part of how the company runs.

In our first article, we discussed who owns what people learn by using AI. In this second piece, we zoom in on who owns the way the work itself is done. Both are leadership appointments rather than IT tickets.

Read more on how our co-founder, Thomas Van Orshaegen, thinks about how organizations can move AI from individual productivity to company-wide intelligence.

The chef is rented. The kitchen is yours.

AI models are interchangeable: they improve for everyone on the same day and are rented by the token. Your kitchen is not. Your processes, your rules and what you know about your customers are yours alone. Written down in a form a machine can use, they become an advantage no competitor buys from the same provider.

That means leadership must be making three choices:

• Redesign the work: writing down a process that should not survive only teaches agents to repeat it.

• Decide what agents 'may' do, not only what they 'can' do.

• Appoint someone accountable for the way agents and people work together.

The agentic AI project of the next twelve months is probably not an AI project at all. It is the work of making your organization's kitchen legible, which makes the board's question a simple one: how much of your business could a capable newcomer run tomorrow, using only what has been written down? Every honest gap is work you can start on now, and it pays off whether your next colleague is a person or an agent.

Curious how much of your business is ready to be handed over? At Craftzing, we help leadership teams turn AI ambition into work that agents can actually take on: from the strategic choice, through the enterprise architecture and the way of working, to the technology that supports it.

--

AI Roundtable Series

In the third and last article of our AI Roundtable series, our CEO Roeland Tegenbos zooms in on keeping control of cost, data and accountability as AI moves into the heart of the operation.

By Kevin Françoisse