The Meeting Where the Question Settled Itself
The scene has repeated itself in mid-market boardrooms with remarkable consistency over the past two years. A leadership team convenes for its regular strategy session. An AI initiative appears on the agenda, and not a careless one: the proposal is competently scoped, with defined use cases, a phased budget, a named owner, and vendors who have been properly vetted. The work deserves a serious hearing, and it receives one.
Then a senior operator, often the executive closest to daily operations, offers a single observation: the team is not ready for this.
What follows is worth examining, because it is not debate. It is agreement, arriving faster than deliberation could account for. The CFO concurs. The head of human resources notes, accurately, how much has already been asked of the workforce. Within minutes the initiative has been moved to the next fiscal year, and the room has passed on to more tractable business.
If you have led one of these meetings, you will recognize what accompanied that decision: relief, and not only among the skeptics. The chief executive felt it too. Most CEOs who deferred AI adoption on these grounds have carried a private discomfort about that relief ever since, suspecting it of being something less flattering than judgment.
An organization’s tolerance for change is a finite resource, invisible on the balance sheet and ruinously expensive to overdraw.
That suspicion deserves to be retired. The relief was pattern recognition, exercised by people whose careers have taught them what transformation costs. Every seasoned executive has watched a well-funded initiative fail, not because the technology disappointed but because the organization had no remaining capacity to absorb it. An organization’s tolerance for change is a finite resource, invisible on the balance sheet and ruinously expensive to overdraw. When the change budget is spent, initiatives do not merely stall; they consume the credibility of their sponsors and raise the price of the next change that matters.
The operations lead who said the team was not ready was reporting a fact. The executives who agreed were protecting an asset. The CEO who accepted the deferral was doing what boards retain CEOs to do, which is to decline investments whose costs are visible and whose returns are not.
The sentence was true when it was spoken. In most companies it is still true today. What has changed is not the accuracy of the observation. What has changed is what the observation implies about the correct next move.
The Conditions That Made Waiting the Correct Call
Decisions deserve to be judged against the information available when they were made. By that standard, the decision to wait was not merely defensible in 2023 and 2024. It was correct.
Consider the conditions. The tools of that period performed impressively in demonstrations and erratically in production, and the distance between those two settings was absorbed entirely by the employees asked to close it. Early adopters reported a consistent pattern: pilots that fascinated, rollouts that disrupted, and returns that proved difficult to isolate from everything else happening in the business. A technology that breaks workflows faster than it improves them is not an investment. It is a tax on attention.
There were, moreover, no playbooks worth the name. An executive who called five peers about sequencing, staffing, or governance received five incompatible answers. Vendors were selling capability rather than adoption, and grew noticeably vague when pressed on what the organization would look like on the other side of implementation. The consulting industry, still assembling its own understanding, was in no position to supply the missing map.
The workforce arithmetic was equally clear. Mid-market employees had absorbed, in close succession, a pandemic, a rebuilt model of distributed work, a hiring surge, a hiring correction, and in many companies a major systems migration that ran past its schedule. Change fatigue is dismissed in some quarters as a soft concern. Operators know better. An exhausted organization does not resist transformation; it simply fails to metabolize it, quietly, at great expense.
And there was the training problem, arguably the decisive factor. The specific techniques of that era carried a shelf life measured in months. Capabilities shifted so quickly that a curriculum was often obsolete before its rollout concluded. Executives were being asked to spend their scarcest currency, organizational attention, acquiring skills with a known expiration date. Declining that trade was not timidity. It was discipline.
Assembled honestly, the calculation produced a clear answer. For a mid-market company without a research budget or a bench of engineers to absorb early failures, deferral was the rational allocation of finite capacity. The CEOs who made that call were reading the conditions correctly.
The record should state as much without qualification, because what follows depends on it. The calculation was sound. The question now is whether its inputs still hold.
Three Conditions Have Flipped, and Only One Involves the Tools
The first change is technological, though not in the way it is usually described. The tools became more reliable, but more consequentially they changed category. Systems that once sat beside an employee and accelerated individual tasks now pursue defined objectives through sequences of actions, with a person supervising rather than operating. We have described this elsewhere as the shift from subscribing to AI to hiring it, and its practical effect is to relocate the constraint: the binding limit is no longer access to capability but the management capacity to supervise it.
The second change is that adoption support matured. The failure modes of 2023 are now documented. Sequencing patterns exist. Governance questions have been asked often enough that the answers are borrowable, and the peer who could offer nothing useful two years ago can now describe precisely what happened in his operations group and what it cost. The missing map that justified waiting has largely been drawn.
The third change is the least comfortable, and the most important.
The competitors whose teams now appear ready did not begin with ready teams. Their people were skeptical. Their first deployments produced little. In their leadership meetings, someone senior said the team was not ready, and that person was correct on the day the words were spoken. Those companies started anyway, at modest scale, and the readiness now visible from the outside is the accumulated residue of two years of doing the work badly, then adequately, then well.
Readiness, it turns out, is an output of adoption, not a precondition for it.
Readiness, it turns out, is an output of adoption, not a precondition for it.
That finding reverses the logic of waiting. The deferral rested on an assumption shared so widely at the time that it required no defense: that waiting was neutral, that a team would be no less prepared next year, and likely more so as the tools simplified. The tools did simplify. But ease of use was never the binding constraint on organizational readiness. Fluency was, and fluency accrues only to organizations in practice. Because the capability curve is now advancing faster than any workforce passively absorbs, the gap between practicing companies and waiting companies no longer narrows with time. It compounds, quietly, in the wrong direction for the company that waits.
Readiness Is Designed, Not Awaited
The reframe at the center of this argument is a short one. Readiness is not a condition an organization reaches through the passage of time. It is a product, and it is manufactured through structured exposure.
The components are unglamorous. Sponsored experimentation: a senior executive owns the permission to try, bounded by explicit rules about which data may be used and which never may. Visible executive use: the CEO and the leadership team working with the tools on their own responsibilities, imperfectly and in view of others, because nothing an executive says about readiness carries a fraction of the weight of being seen learning. Protected time: hours on the calendar, defended when the quarter tightens, because readiness is purchased in time rather than enthusiasm. And small, scoped wins: bounded problems with checkable results, chosen so that the first success becomes a story a skeptic can repeat accurately.
Most companies, it should be said, already operate a readiness engine. It is simply running unsponsored. In most organizations of any size, employees are using these tools now, on personal accounts and personal initiative, applied to the work they are paid to do. They have not announced it, because no policy exists that makes announcing it safe, and because the productivity gain belongs to them for exactly as long as it remains invisible. The question facing the chief executive was therefore never whether to introduce AI into the company. It arrived unbidden, as email once did. The question is whether adoption remains unsponsored, where lessons fail to compound, discoveries never travel beyond one desk, and corporate data exits through unmonitored channels, or whether leadership sponsors and structures what is already underway.
This is also why the instinctive response, commissioning a training program and awaiting its effect, tends to disappoint. As we have argued in these pages, the scarce and expensive variable was never the answer; it is adoption itself, the slow and irreducibly human work of getting an organization to do the thing. Training built in advance of practice teaches yesterday’s interface to people who have not yet formed the questions. Training built from an organization’s own sponsored experiments teaches what its own people discovered. The sequence matters, and it runs opposite to instinct.
What the Transition Honestly Looks Like
There is a question that was not asked aloud in that leadership meeting, and it is generally not asked aloud anywhere, because one version of it sounds heartless and the other sounds afraid. What happens to the workforce, and to the executives themselves? The question deserves a more honest answer than the market has offered.
The honest answer begins with an admission. Some work goes away. Pure-execution tasks with clean inputs and defined outputs, first-pass document review, routine reconciliation, tier-one triage, standard reporting, the first draft of nearly everything, will contract, and reassurances to the contrary are managing the listener’s feelings rather than informing his decisions.
But roles are bundles of tasks, and bundles reorganize faster than they disappear. As the execution portion contracts, what remains concentrates in judgment on ambiguous cases, oversight of output that now arrives faster than it can be casually checked, exception handling, relationships that carry obligation, and accountability that must bear a human name. On the timeline that matters to a mid-market company, most roles transform rather than vanish. Some will not survive the transformation, and leadership will see which ones well before the people in them can. The useful response to that foresight is specific, early planning for those individuals, with more notice and more help than the market will extend to them.
Managers face the sharper change. They are beginning to supervise output from AI workers alongside output from people, and span-of-control assumptions bend under the combination. A manager responsible for six people and nine autonomous agents holds a different job than a manager of fifteen people, and no one trained her for it. It is, however, a discipline rather than a talent: sampling quality instead of reviewing everything, stating an acceptable error rate aloud, and knowing which decisions are never delegated to anything that cannot be held accountable. Disciplines are learnable, and this one is learned in practice.
The failure to avoid is not caution but presumption, in either direction. The chief executive who froze entry-level hiring on the strength of headlines and a demonstration, without piloting a single workflow, committed the same category of error as the executive who dismissed the technology entirely.
What the CEO owes the leadership team is a sayable truth: we will not pretend to predict the end state, we will build the capacity to keep adapting, and the people who engage with this work will become more valuable here, not less. That commitment can be made honestly. The person who can direct and audit machine-produced work is worth more than the person who produced it by hand, inside the company first and in the wider market soon after.
The Design Task of the Next Twenty-Four Months
The chief executive’s task, properly stated, is not to forecast where the workforce lands. It is to design the transition, and transition design is concrete work with three parts.
The first is sequencing. Some function must go first, and the correct choice is rarely the loudest advocate or the largest budget line. The better candidates share three traits: the work is already documented, the output is checkable by someone who knows what correct looks like, and the function’s leader holds credibility with peers. Two of the three suffice. The purpose of the first function is not its savings but its story, the internally credible account of what happened that makes the second and third functions possible.
The second is defining what humans own permanently, in writing, before anyone asks. Judgment under ambiguity. Relationships that carry obligation. Accountability that must have a name attached when something goes wrong. Publishing that list accomplishes more for organizational readiness than any curriculum, because people commit to change far more readily once they can see precisely what is not being taken from them.
The third is cadence. Every conclusion in this domain now carries a half-life, including the conclusions in this article. The sequencing choices, the ownership list, and the pace of expansion should be revisited quarterly, on the calendar, with the same seriousness given the financial forecast. A decision that is never revisited is a decision that expires without notifying anyone.
Then there is one question worth carrying into the next leadership meeting and allowing to sit unanswered for a while: which three roles in this company would become dramatically more valuable if the person in them had an AI team underneath them, and what would it take to make that true by next quarter?
Note what the question does. It is not an instrument of reduction, and a leadership team hears that immediately. It points at the strongest people in the company and asks what they could carry if the routine execution beneath them were handled. It converts a deferred abstraction into three names and a deadline, which is the form in which executives have always done their best work.
The deferral was right. The start is right. Neither verdict requires the other to be wrong.
The deferral was right. The start is right. Neither verdict requires the other to be wrong. Readiness was never a condition that arrives with time; it is a thing organizations design, and design is work a capable leadership team already knows how to do.
