Most organizations did not sit down and decide how AI would fit into their business.
It arrived in pieces. A tool here. A pilot there. A vendor demo that became a department habit. By the time anyone asked the bigger question, AI was already inside the walls, doing real work, in ways nobody had specifically authorized.
This is not unusual. Most technology adoption happens this way. What is unusual is the size of the decision that got skipped, and the fact that nobody can say who made it.
The Decision That Was Never Assigned
Somewhere in the rollout of every AI tool is an implicit answer to a question almost no organization has formally assigned to anyone: who decides whether a given task is performed by a person, by AI, or by some combination, and on what basis?
In most organizations, that decision is currently being made by default, at the level of individual managers, one workflow at a time. A manager adopts a tool because it saves time. A team starts routing certain requests through AI first because it is faster. None of these choices were made by someone with the authority or the visibility to ask what the cumulative effect would be.
This is not a failure of intention. It is a failure of governance. Each individual adoption decision looks like a small efficiency gain, made by someone close to the work. None of them, on their own, looks like an organizational decision about how work gets done. But an organization that has accumulated enough of these small decisions has, in effect, made a large one, without anyone having the authority to make it, the visibility to see it happening, or the accountability for the outcome.
Two Models, and Who Is Accountable Under Each
Call the first model human-first. AI exists to make people more capable. It removes friction, surfaces information, and handles the parts of a job that get in the way of the parts that matter. The person remains accountable for the outcome, and AI is a tool they direct.
Call the second model AI-first. AI performs the core of the work, and people exist to manage, train, correct, and oversee it. Accountability shifts: the person is now accountable not for the output itself, but for the quality of oversight applied to a system producing it.
These are not just different philosophies. They are different accountability structures, and they require different things from the people in them. A human-first role can be staffed and evaluated the way roles have always been staffed and evaluated. An AI-first role requires a different skill set, oversight, correction, judgment under uncertainty, and a different way of measuring whether someone is doing their job well.
Most organizations are drifting from the first model toward the second without updating the accountability structure to match. People are increasingly accountable for overseeing AI output, but their roles, their training, and their performance metrics were built for a world where they produced the output themselves.
Why This Is Not Just a Strategy Slide
It would be easy to treat this as the kind of question that gets answered in a strategy offsite and then filed away. It is not that kind of question, for two reasons.
First, the decision is not made at the strategy level. It is made at the level of individual workflows, by individual managers, deciding individually whether a given task goes to a person or to a system. Strategy can set a direction, but the direction only becomes real if those thousands of small decisions actually move toward it, and if someone has the job of checking whether they do. An organization can have a human-first strategy on paper and an AI-first reality on the ground, because nobody was assigned to connect the two.
This is not a distant concern. Recent analysis projects that over the next two to three years, roughly half of jobs in the US will be reshaped by AI, with most roles remaining but changing substantially in what they require. That reshaping is happening role by role, decision by decision, inside organizations that have not assigned anyone to track it.
Second, the consequences of an unassigned decision show up in places that look unrelated to the decision itself.
When an employee raises a religious objection to how AI is used in their role, the organization is suddenly asked who decided that this task would be done this way, and whether that decision can be revisited. An organization that cannot answer the first question has no path to answering the second. The accommodation request becomes hard not because the employee’s concern is unusual, but because no one was ever assigned to own the answer.
When the costs of an AI rollout turn out to be different than expected, talent premiums that did not pencil out, roles that get eliminated and then rehired at higher cost, performance metrics that reward the wrong things, the organization is discovering, after the fact, the financial consequence of decisions nobody was accountable for at the time they were made.
Both of these are downstream of the same upstream gap. The decision was never assigned. It was made anyway, by accumulation, by whoever happened to be closest to the workflow that day.
What Assigning the Decision Looks Like
Assigning this decision does not mean creating a new committee or adding a line to someone’s title. It means being able to answer, for any given role or task, three things: which model applies, who decided that, and who is accountable for revisiting it if circumstances change.
For a task currently run human-first: is that the result of a decision, or the absence of one? If no one has evaluated whether AI could take on more of this task, the organization does not actually know whether it is human-first by design or merely by default.
For a task currently run AI-first: who is accountable for the oversight role, and were they given the training and time to do it well? If the answer is “whoever happens to be assigned to that team,” the organization has created an AI-first task without creating the accountability structure that task requires.
Most organizations, if they ran this exercise honestly across their workflows, would find a large number of tasks where the honest answer to “who decided this” is no one, exactly. That is not a philosophical gap. It is a governance gap, and it is fixable in the same way other governance gaps are fixable: by deciding who owns the decision, documenting how it gets made, and building in a mechanism for revisiting it.
The Starting Point
This is not an argument for slowing down AI adoption, or for choosing one model over the other as a matter of principle. Both models can work, and most organizations will run both simultaneously across different parts of the business.
It is an argument that someone needs to own the decision of which model applies where, because right now, in most organizations, no one does.
Everything else, the accommodation request that has no clean answer, the cost overrun that nobody saw coming, the manager who cannot say whether a task requires AI or merely uses it, traces back to the same starting point. A decision about how work gets done was made. The question is whether anyone was accountable for making it.
The Inside Advantage 
Leave a comment