Why the playbooks leaders keep reaching for are actively unhelpful, and what to do instead.

Primary reader: Board member, CEO, CHRO, Transformation lead

Primary use case: Strategy offsite pre-read, executive briefing, governance framework anchor

For the past four decades, organisational change has followed a recognisable grammar. You diagnose the current state, define the future state, and map the path between them. You manage resistance along the way. That grammar has carried mergers, ERP rollouts, digital transformations, and cultural resets. Its underlying assumption has always held: there is a future state worth naming, and the change programme is how you get there.

AI breaks that assumption.

The capabilities organisations will depend on three years from now have not yet been built. The roles that will matter most have not yet been named. Tooling, talent profiles, workflow design, the organisational shape itself, all are moving faster than any planning cycle can accommodate. Leaders are being asked to define a future state that will be obsolete before the transformation office has finished writing it.

Having spent the last three years in conversation with leaders across more than thirty countries, I have seen a consistent pattern. The organisations handling AI best are the ones that have quietly stopped treating it as a change management exercise. The organisations struggling most are the ones still trying to run the old playbook at AI speed. This piece is about why that is, and what the alternative looks like.

The imported assumption

Change management as a discipline assumes the destination is known, the journey can be planned, and resistance is the primary obstacle. Every major framework, from Kotter's eight steps to ADKAR to Lewin's unfreeze-change-refreeze, rests on those three assumptions, and they are being imported wholesale into AI programmes, usually without examination.

None of the three holds.

The destination is not known. Planning horizons for anything touching AI have collapsed to something between one and three quarters, and even that is generous. The capabilities organisations will be building against three years from now are being demonstrated on a public roadmap no one can see. Any five-year workforce plan written against that moving target is, in practical terms, not a plan. It is an aspiration with a budget attached.

The journey cannot be planned either. Planning presumes a stable relationship between inputs and outputs. AI breaks that relationship unpredictably. A capability that required six months of training in January can be available to anyone with a £20 subscription by July. A role that was automation-proof in 2023 is automation-adjacent by 2025. The CIPD's Labour Market Outlook, published in November 2025 and drawn from more than 2,000 UK HR decision-makers, found that one in six employers now expect AI to reduce their headcount within a year, with junior, clerical, and administrative roles the most exposed. A quarter of those employers expect to lose more than 10% of their workforce. That is not the shape of an orderly transition.

And resistance is not the primary obstacle. In the AI case, the primary obstacle is uncertainty, which is a different phenomenon with different dynamics. Resistance responds to communication, incentives, and change champions. Uncertainty does not. Uncertainty responds to transparency, sense-making, and the willingness of leaders to say, on the record, that they do not yet know. Most change programmes still treat uncertainty as resistance, which is why so many AI initiatives produce employee cynicism rather than engagement.

Three imported assumptions, none of them fit for purpose. The playbook is sound. The playbook is wrong for the problem.

The script

The most damaging consequence of treating AI as change management is the pressure it creates on leaders to perform a confidence they do not possess.

Consider what a CEO is expected to say in 2026. We have a clear AI strategy. We are embedding AI across every function. We have a roadmap for skills transformation, a target operating model for the AI-enabled enterprise, a governance framework, a centre of excellence. Each of those statements is required by the grammar of change management. Each of them, in most organisations, is not quite true. The strategy is a direction, not a plan. The embedding is uneven. The roadmap is provisional.

Leaders know this. Workforces know it too. There is a script for what the CEO is supposed to say, and everyone in the room can recite it along with the speaker. The distance between what the script says and what the room knows to be true is what erodes trust, slowly, across every layer of the organisation.

Take a recent conversation I had with the HR lead of a large financial services firm. She described the tension in her sales team. As the AI tools had grown more capable and more visible, her sellers had moved from curiosity into fear. The more clearly they understood what the tools could do, the more certain they became that the tools were coming for their jobs. Leadership had responded with the expected script: we are augmenting not replacing, these tools will free you up for more strategic work, your role is safe.

The sellers were unconvinced, and they had reason to be. None of those statements could be verified, and some were probably false. They could feel the difference between honest uncertainty and corporate reassurance. The reassurance made the fear worse, not better. I have watched this same scene play out, with small variations of vocabulary, in sixteen organisations over the last year. The script does not work. It has not worked for some time. And still the script is what leaders reach for, because the alternative, standing in front of a workforce and admitting what nobody has prepared you to admit, feels like professional self-harm.

It is not. It is the only move that compounds trust rather than eroding it.

The script plays out at every level. Boards demand AI strategies, CEOs produce them, the strategies get cascaded. The workforce, which can read the trade press as easily as anyone else, watches the strategy get overtaken by developments the document did not anticipate. Leaders then face a choice between updating the strategy quarterly, which makes them look unstable, and defending a strategy they know is already out of date, which makes them look dishonest. Most choose the latter, and the script continues.

The alternative is not an absence of strategy. It is a different register. Leaders who have adopted that register speak in provisional terms. They name what they know, what they are watching, and what they are prepared to adjust. They do not promise a destination. They describe commitments, and the conditions under which those commitments would change. This sounds weaker in a boardroom than the traditional language of strategic clarity. In almost every organisation I have seen, it is more durable.

What the evidence is telling us

Three findings from the last year are worth holding together, because each of them contradicts a prediction made confidently in 2023 or 2024.

The first concerns what AI actually does to expertise. A large randomised study published in 2025 and analysed by MIT Sloan, covering nearly 5,000 developers across Microsoft, Accenture, and an anonymous Fortune 100 manufacturer, found that GitHub Copilot raised completed tasks by an average of 26%. The more interesting result lay inside that average. Less experienced and shorter-tenured developers saw productivity gains of 27 to 39%. More senior developers saw gains of 8 to 13%. This is a software-development finding, and the authors are careful not to generalise beyond it. I will risk the generalisation anyway, because I think the pattern holds more widely than the methodology allows them to say: AI lifts performance from the bottom upward while leaving the ceiling roughly where it was. The consequence is not that AI is an equaliser. It is that the scarcest skill in the next five years will not be using AI. It will be knowing which tasks it should and should not be used for, which is a judgement the tools cannot yet make for you.

The second finding concerns organisational shape. For two years, the confident prediction has been that professional services firms, banks, and consultancies would move from pyramid to diamond: fewer graduates, more mid-career staff, fewer leaders. The logic was that AI would automate the work graduates typically did. In practice, the picture is far messier. In February 2026, IBM's chief HR officer Nickle LaMoreaux announced at the Charter Leading with AI Summit in New York that the firm would triple its entry-level hiring in the United States in 2026, explicitly including the roles AI was meant to replace. Set against this, the same CIPD research cited above shows UK employers moving the other way, expecting junior roles to fall first. The pyramid has not become a diamond. It is reshaping in different ways in different sectors, in different countries, in different firms. A single transformation narrative does not survive contact with the data.

The third finding is the most telling. Within any single organisation, AI adoption is deeply uneven. Sales and recruiting functions, where KPIs are tight and measurable, are typically furthest along. HR operations, finance, and much of the legal function trail by several quarters. Specialist technical functions in regulated industries often sit further back still, sometimes by design. A CHRO I work with described her organisation as containing divisions that operate like AI-native startups and divisions that run much as they did in 2019, housed in the same building, answering to the same executive committee, working from the same strategy deck. This is not a failure of rollout. It is a feature of how AI adoption actually moves through a real company.

Performance lifted unevenly. Organisations changing shape in ways nobody predicted. Adoption proceeding at different speeds in different corners. A single, centrally-driven transformation plan is, in this context, the wrong instrument for the work.

Value drift

There is a second kind of drift, under-discussed but probably more important than headcount forecasts, and it sits inside the work itself.

A February 2026 article in The Conversation, written by researchers studying AI use in Australian and New Zealand organisations, introduced the idea of value drift. Generative AI, they argue, does not simply automate calculations. It automates plausible language. It writes the summary, the rationale, the email, the policy draft, the performance feedback. And because the output sounds reasonable, the values those texts encode shift incrementally without anyone noticing. A manager uses AI to draft performance feedback, and the prose is smoother, but the judgement is harder to locate. A policy team uses AI to produce a balanced justification for a contested decision, and the trade-offs disappear into the fluency. Over time, the meaning of good work quietly changes.

This is the part of AI transformation that no change management framework will reach, because it is not a change programme. It is an accumulation. And it is the strongest possible argument for why AI transformation cannot be run as a time-boxed initiative. The thing you most need to watch cannot be captured in a milestone.

The geographic divergence

One further pattern matters, because it shapes how leaders calibrate their own pace.

North America, broadly, is moving on AI with a conviction that Europe and the UK are not matching. American CEOs worry about under-investing. European CEOs worry about over-investing. The boardroom question in New York is whether we are moving fast enough. The boardroom question in London, Paris, and Frankfurt is whether we are moving too fast and will regret it.

This is not really about technology. It is about risk culture, regulatory context, and the different relationships between enterprise and capital markets on either side of the Atlantic. But it has real consequences. European leaders frequently describe a kind of asymmetric anxiety, worrying that they are falling behind American competitors while simultaneously worrying that matching the American pace would be reckless. The practical effect is hesitation, which is its own form of decision.

The right response is neither to mimic the American tempo nor retreat into European caution. Many of the firms pushing hardest on AI will make expensive mistakes that will be publicly reported over the next two years. Many of the firms holding back will discover, too late, that windows they let close do not reopen. The right response is a sector-specific view rooted in your actual operating context, which is the harder answer and the more durable one.

The practice that replaces the plan

So if AI transformation is not a change management problem, what is it? The framing I have found most useful, and the one I now use in most client conversations, is that AI transformation is a capability-building problem with a moving target. It resembles the development of organisational athletic fitness more than the execution of a programme.

Athletic fitness is not delivered in a six-month initiative. It is built through repeated, well-chosen practice, sustained over years, with periodic recalibration as the athlete's body and goals change. The organisation that is fit for AI in 2028 will not be the one that ran the best transformation programme in 2026. It will be the one that built habits in 2026 and still had them in 2028, even as the tools underneath those habits changed several times.

The practice I have seen work can be distilled into three questions, asked by the senior team of the organisation every quarter, with the expectation that the answers will change.

What is breaking now that was not breaking last quarter?

This is the detection question. Something in the organisation will have shifted. A process that ran smoothly in February is now backed up. A team that was performing is now confused because the tool it relied on has been deprecated, replaced, or outgrown. A capability that used to sit with a specialist is available to a generalist with a subscription, and the specialist is asking what her role now is. The question forces leaders to look at the actual surface of the organisation rather than the reported one. Most organisations find out about these fractures only when they surface in exit interviews, customer complaints, or missed numbers, by which point the cost of correction has already multiplied. Asking quarterly does not prevent the fractures. It shortens the time between them forming and being noticed.

Who is now doing work that we thought required someone else?

This is the reshaping question. The map of an organisation, who does what at what level for what salary, is being redrawn each quarter by people who are not waiting for permission. A graduate producing outputs at the level a three-year associate used to produce. A finance partner whose judgement calls are being approximated by a senior analyst with the right tools. A marketing coordinator whose content output has tripled. These shifts are not in the organisational design deck. They are in the work itself. The question lets leaders see the reshaping while it happens, rather than a year later when the pyramid has quietly turned into something else and nobody is sure how to re-grade bands, structure promotions, or price roles in the external market.

What are we still doing that nobody needs us to do?

This is the subtraction question, and it is the hardest of the three. Organisations are extraordinarily good at adding capabilities, tools, processes, meetings, and reports. They are spectacularly bad at removing any of them. A report that mattered in 2022 is still being produced in 2026 because nobody has asked whether it still matters. A weekly meeting that earned its place in the calendar five reorganisations ago is still running. A licence bought in the first wave of enthusiasm is still being paid for, unused. Every quarter, something on that list has shifted from useful to wasteful, and nobody will notice unless you ask. The question requires active leadership cover, because the honest answers tend to implicate the person who introduced the thing in the first place, and that person is frequently in the room.

Three questions, asked every quarter. That is the minimum. The best leaders I have seen ask them more often, and they ask them of people two and three levels below the executive team, because those are the people who see the answers first.

It is deliberately simple. A more elaborate framework would imply the problem is elaborate, and the problem is not elaborate. The problem is that the environment changes faster than the organisation's sensing mechanisms, and a short list of recurring questions is a better sensing mechanism than a long document reviewed annually. Organisations that adopt this practice develop AI muscle rather than AI strategy. A strategy document has a half-life of about six months. Muscle compounds.

The leadership posture

Everything I have described above asks something of leaders that their training has not prepared them for. Not better strategy. Not more data. A different posture entirely.

The traditional leader announces direction, mobilises the organisation, and delivers outcomes against a plan. The AI-era leader does something quieter. He names what is known, what is being watched, and what would cause him to change course. He is comfortable saying, on the record, that the organisation's answer to a particular question will probably be different in three months than it is today. In a 2026 boardroom that still rewards the performance of certainty, this kind of talk can sound like weakness. It is not. It is calibration, and it is the only posture that survives contact with an environment that keeps moving.

The leaders who will run the most successful organisations of the next decade, in my view, are not the ones with the best AI strategies. They are the ones who have accepted that AI does not reward strategy in the traditional sense at all. It rewards a repeated, disciplined practice of paying attention, telling the truth about what you see, and adjusting. The leaders I have watched learn this have, almost without exception, told me the same thing: it was easier than they feared. Workforces do not punish honesty. They punish the gap between stated and actual.

The central mistake of the current moment is the belief that AI transformation is a problem to be solved. It is not. It is a condition to be lived with, through posture and practice rather than plans and playbooks. Organisations that accept this earliest will have the deepest advantage. The ones that continue to commission transformation programmes with defined end states will spend the next five years producing decks their workforces have already stopped reading.

_______________________________________________

Rahim Hirji is the founder of The SuperSkills Intelligence Company and author of SuperSkills: The Seven Human Skills for the AI Age (Kogan Page, July 2026). He advises organisations on human capability development in the AI age.