The great AI paradox does not exist.

I · A disappointing discovery

McKinsey has discovered a great paradox. Companies expected enormous transformations from artificial intelligence, invested accordingly and then failed to see the corresponding effect on their bottom line.

I have disappointing news.

There is no paradox.

When an organisation buys something and the expected value does not appear, we have not encountered a rupture in logic. We may simply have encountered an investment that did not work. The technology may be immature, the use case weak, the implementation expensive or the original expectations nonsense. These possibilities are merely awkward, because each requires someone to reconsider the decision.

II · The perfect circle

McKinsey offers a more convenient explanation. Value is missing because workflows, leadership and operating models have not changed sufficiently. Companies invested in AI but did not become genuinely agentic.

The absence of evidence becomes evidence that we need a larger dose.

In medicine, we might test the treatment before doubling it. Here, the patient is redesigned to fit the medicine. When investment does not pay, the answer is more AI, more governance and more radical transformation. Push harder. Become more agentic. This is not an explanation of failed returns. It is a mechanism for protecting the original assumption from failed returns.

III · From possibility to necessity

AI agents can execute tasks, call tools and sometimes perform bounded work faster than people. From this reasonable possibility, McKinsey leaps to necessity: workflows must change fundamentally, operating models must shift, roles must be reshaped and leaders must transform themselves.

But there is no such law.

An agent may automate one step behind an otherwise unchanged service. It may justify a wider redesign. It may also be too unreliable, expensive or unnecessary to use at all. The category agentic AI tells us none of this. The use case does.

A particular implementation may change permissions, responsibilities or work. But saying that agentic technology requires a new organisational paradigm is technological theology. A possibility has been promoted into destiny, and a management choice has dressed itself as a law of history.

IV · The consequence becomes the cause

Once an organisation chooses agents that produce uncertain outputs or act with delegated authority, consequences follow. Results must be evaluated, permissions constrained, actions logged and privacy, security, legality and accountability assured.

McKinsey presents these as capabilities demanded by the agentic era. But the era made no decision. Management did: The organisation selected the use case, autonomy and acceptable risk. Any resulting need for supervision, structures or controls follows from that selection. It is not mysteriously emitted by AI itself.

The consequence is presented as the cause. Responsibility evaporates somewhere between the first and third line.

V · The free transformation

The same manoeuvre makes costs disappear. Governance, integration, training, redesign, human review and error correction become transformation enablers. Economically, they remain costs of choosing and operating the system.

Calling a cost a capability does not remove it from the income statement.

If an agent saves one thousand hours but requires nine hundred hours of checking and recovery, gross productivity is irrelevant. If it creates mistakes or work elsewhere, local speed is not organisational value. And if people cannot verify the machines they supervise, moving them “above the loop” merely moves the inspection desk upstairs. The calculation must include acquisition, integration, inference, oversight, failures, maintenance and poor quality. Only the remainder is value.

If nothing remains, the conclusion is not that the organisation lacks agentic maturity. The conclusion may be that the tool is a burden.

VI · History has joined the sales team

Weak business cases borrow authority from inevitability. Technology advances, jobs change and AI users replace non-users. The agentic era approaches, so leaders must reorganise before history notices their hesitation.

But development, availability, adoption and value are different things. Technology can advance without being suitable, spread without being economical and be adopted without producing value. It can also create value without demanding a leadership revolution.

History has been hired as a salesperson, and fear has been assigned to customer success.

Inevitability conceals choice. If adoption is destiny, no one must justify it. If transformation is unavoidable, its cost need not compete. If employees fear replacement, enthusiasm can be mistaken for consent. This is not foresight. It is a sales funnel in which the future has already signed the contract.

VII · Begin with the promise

An organisation exists not to become agentic but to keep promises: create value, meet quality, safety and legal requirements, deliver on time and control total cost. These are the boundary conditions. AI does not set a single one of them.

Beginning with an agentic organisation of the future puts the answer before the question. It makes the tool the destination and then searches for reasons to travel there.

Only then can we know what should change—technically, procedurally, organisationally or not at all. Its legitimacy comes from improved performance against the promise, not from the word AI.

VIII · What survives the rhetoric

McKinsey is strongest when it forgets the agentic destiny. Examining a whole workflow can expose bottlenecks that task-level productivity hides.

Asking where human judgment belongs, recognising machine-produced rubbish, preserving junior learning and favouring reversible experiments are reasonable ideas. None of them proves the need for an agentic organisation. They prove that technology should be evaluated inside a real system. Sometimes a workflow should be redesigned. Sometimes one task should be automated. Sometimes people should do the work. And sometimes the most mature AI decision is not to use AI.

That is not resistance to change. It is resistance to changing without a reason.

IX · No paradox, only responsibility

Companies expected transformation, invested and did not receive the return. There is no paradox—only a business case that has not closed and decisions that should be examined. Perhaps they chose the wrong problems, misunderstood the work, measured activity instead of value or trusted inadequate tools. Perhaps the revolution exists mainly in presentations by those who benefit when transformation never ends.

The answer might still be an agentic workflow. But it must survive comparison, evidence and total cost. It cannot be smuggled into the premises and rediscovered as the conclusion.

AI does not absolve management of choosing. The future accepts no accountability for an investment. Technology gains no authority to reorganise a company merely because someone gave the decade a dramatic name.