The scene repeats itself in companies of every size. Somebody declares in a meeting that artificial intelligence has to be implemented and ten minutes later the table is comparing platforms, models and demos. Nobody has said what result needs to change. The conversation advances with energy and does not move from where it stands: the resource is discussed before the work has been designed.
With people we never make the mistake in that order: no serious company hires somebody and afterward decides what it could do with them. First there is a need, a result, a role; afterward the holder is selected. Many AI initiatives invert the exact sequence: they buy the technology and walk the organization looking for tasks that justify the purchase.
The discipline that corrects that order is called Work Resource Management, WRM, the management of work resources. It is the backbone of the book AI Employee, and it starts from an idea that reorders everything: the primary object of management is not the person nor artificial intelligence. It is the work that must produce a result.
The solution before the problem
The book tells it with a composite case. A mid-sized distributor sees the demo of a conversational platform at a trade show; three months later it is bought, integrated into the website and answering questions. The real bottleneck remains intact: quotes take days because the pricing exceptions live in the head of the commercial manager, who resolves them one by one over WhatsApp. The company automated the flashiest part and left untouched the part that was bleeding.
Nobody did anything stupid: every step looked reasonable. The error was in the order, because the solution arrived before the diagnosis. And the ending is not an explosion: the distributor keeps paying the license, cancelling feels like admitting the mistake, and the quotes still take days. The real cost was spending the transformation budget and the internal credibility on the part that did not hurt. Bad decisions rarely explode. They almost always stay.
Try it with your own operation. Think about your last software purchase or AI initiative: what indicator was it supposed to modify? What concrete responsibility was it supposed to improve? Who asked for it, the process or the enthusiasm? If you cannot answer with precision, you probably started with the solution. That error was not born with AI, but its speed makes it more expensive.
Building the wrong thing fast is not progress. It is accelerated waste.
Managing the work before the worker
The separation that changes the conversation is this one: the work is not the person who executes it today. A role exists because the organization needs to produce certain results with a determined frequency, quality and risk. That María has spent eight years doing it magnificently is information about María, not about the nature of the role. That a language model can do a part of it is information about the model, not a sentence on the destiny of the role.
To manage the work you have to be able to describe it without naming its holder: the purpose, the responsibilities it contains, the decisions it demands, the authority needed, the performance indicators, the dependencies and the place its exceptions escalate to. Only when that architecture becomes visible does it make sense to ask who should hold it.
The order changes the question that governs the implementation: it is no longer "where can I put AI?", but "what result needs to be produced and how must the work be organized to get it?". Once that question is answered, technology stops being a bet in search of a justification.
If it finds clarity, it multiplies it. If it finds confusion, it accelerates it. Role design is not bureaucracy prior to AI: it is the difference between multiplying a healthy process and accelerating a sick one.
Four answers, not two
The commercial conversation admits two possible holders: a person or AI. WRM recognizes four. A person, when contextual judgment, trust, negotiation or personal responsibility predominate. An artificial resource, when recurrence, volume and auditability dominate. Deterministic software, the script of all time, when the flow is completely enumerable and the exceptions are rare: spending a probabilistic model on fixed-rule work is buying risk without buying benefit. And an assisted person, when the role has to stay with them but part of their functions is better executed by an artificial resource that prepares the ground for them.
WRM does not fall in love with the resource. It commits to the result. The neutrality is about the type of resource, never about the people affected: each responsibility must be executed by the configuration that produces the best result with acceptable levels of cost, quality, speed, risk and accountability. The assignment is argued, responsibility by responsibility, with the HWFA, the assessment instrument of the book.
Break the role down before automating it
A role is almost never a single thing. Treating it as a single piece leads to clumsy decisions: automate everything or automate nothing. The useful unit of analysis is not the role. It is the responsibility.
The book demonstrates it with its guiding case: a collections coordinator. If you ask what the role exists for and the answer is a list of activities, you have not reached the purpose yet. Ask what results would stop existing if the role disappeared. There it appears: turning invoicing into cash while protecting the relationship with the customer.
Separate it into responsibilities and look at them one by one. Monitoring of balances and aging: daily, repetitive, cheap and correctable error. Early reminders: high volume, standardizable tone, minor consequences. Negotiation of agreements: low frequency, contextual judgment, commitments that bind the company, expensive error. Vulnerable or sensitive customers: little volume and maximum delicacy, because a clumsy move does not cost an invoice but the relationship. Escalation of disputes: knowing when a case stopped being collections and became a commercial, legal or human problem.
Five responsibilities, one single role, and completely different profiles of frequency, judgment and cost of error. Whoever asks "do we automate collections?" is asking a question too coarse to be answered well. Whoever asks responsibility by responsibility discovers precise combinations: perhaps the first two ask for an artificial resource, the third and the fourth remain human with artificial preparation of the context, and the fifth demands an escalation design that today does not even exist on paper.
The exercise usually reveals something uncomfortable: many companies did not design their human roles well either. The responsibilities grew out of habit and the ownership gets shared out until nobody truly answers. A machine cannot clarify by itself what the organization never decided. Before automating, you have to manage.
This is not Taylorism
The objection always arrives, so it is best to answer it head on. Designing the role first is not treating people as pieces that fill pre-designed boxes. The declaration of the role governs the order of its birth, not the rest of its life: the holders, the human ones above all, remodel their roles, and a framework that denied that evolution would deserve the criticism.
What the hybrid organization demands is that every remodeling be declared. Where all the colleagues are human, a role that drifted from its description is learned in the hallway: everybody knows that "that is actually handled by Marta". An artificial colleague does not walk down the hallway; it can only read what is declared. The declared role is the interface. Taylor fixed the person to the box. This discipline versions the box.
The principle and its consequences
The WRM framework has a founding premise: first there is an organizational need; from it a role is born with purpose, ownership, results, KPIs, authority, limits and escalation; only then is it decided which resource should hold it. And its first rule closes the circle: every role must exist before its holder, and no role should survive its purpose. Birth and death, in the same sentence.
From there the rest of the method follows: the assignment is argued with an instrument; when the holder is artificial, the role is formalized in a role contract; and autonomy is earned with evidence. The role before the resource is not a slogan. It is the discipline that prevents confusing novelty with direction.
Frequently asked questions
Work Resource Management, the management of work resources, is the discipline that designs, assigns, governs and optimizes the work independently of whether the resource that executes it is human or artificial. Its founding idea is to separate the role from the holder: first there is an organizational need, from it a role is born with purpose, results, KPIs, authority and escalation, and only afterward is it decided which resource should hold it. It is the framework on which the book AI Employee is built: it manages the work that people and systems execute.
Because the inverse order produces the most common failure pattern in enterprise AI: buying a platform and afterward looking for tasks that justify the purchase. The predictable result is automating the most eye-catching part of the process and leaving the bottleneck intact. When the role is designed first, the question changes from "where can I put AI?" to "what result needs to be produced?", and technology goes from being a bet to being a resource selected with judgment.
First mentally delete the name of whoever holds it and ask what results would stop existing if the role disappeared: that gives you the purpose, not the list of activities. Then separate the role into responsibilities and evaluate each one by its frequency, variability, need for judgment and cost of error. A collections coordinator contains five responsibilities with completely different profiles. That separation makes it possible to decide which part is worth automating, which one remains human and which one needs artificial assistance.
No, and the difference is in what the declaration of the role governs. Taylor fixed the person to a pre-designed box; WRM versions the box. The declaration governs the order of the birth, not the rest of the life: human holders remodel their roles and the framework recognizes it. What the hybrid organization demands is that every remodeling be declared, because an artificial colleague does not learn in the hallway that "that is actually handled by Marta". It can only read what is declared. The declared role is the interface.
To understand what exactly an AI Employee is and what separates it from an agent, read What is an AI Employee?.
Want the full method? Read AI Employee. For executive AI consulting or keynotes and workshops.
Go deeper
Want to bring your team to the next belt?
Book a discovery call or explore the full book.