Each year Microsoft publishes the Work Trend Index, and each year it is the most widely read piece of research on how AI is changing work. This year's edition — Agents, human agency, and the opportunity for every organization — is genuinely good. It is senior, it is honest about the organisational nature of the challenge, and it resists most of the hype that surrounds the subject. I read all twenty-nine pages, the Frontier Firm playbook that sits behind it, and the how-to guides for leaders, and I came away agreeing with a great deal of it.
So this is not a takedown. It is the first of what I intend to make an annual habit: an independent reading of the report from the position Requisite Intelligence is built on. We sell no software and licence no platform, so we have no stake in the answer before we have heard the question. That independence is worth most precisely when a subject is moving this fast and the loudest voices have something to sell.
I want to start with what the report gets right, because it is a lot. Then I want to spend the rest of this piece on the four places where my research with UK SMEs, and the practice that has grown out of it, sees the world differently. One of those differences is foundational; the others follow from it.
Where the report is right
Dr Karim Lakhani's foreword could, in places, have been lifted from our own work. He draws the distinction between a business model — how a firm creates and captures value — and an operating model — how that value actually gets delivered, day to day, underneath. He argues that the consequential change underway is not the adoption of new tools but the emergence of a new operating model, and that the real challenge is therefore managerial and organisational rather than technical. He is right. His standout line is that “the path from adoption to advantage is neither linear nor automatic,” and that single sentence does more honest work than most of what is written on this subject.
Two phrases from the report are worth taking into everyday use. The first is the distinction between AI adoption and AI absorption — between a tool being available and a tool being woven into how work actually gets done. The second is the idea that every firm is a learning system: that the firms pulling ahead are the ones turning their own output into insight, capturing it, and feeding it back. Both are true, and both are useful. I have already started saying “absorption” where I used to say “adoption.”
The report is also right about the headline finding: organisational factors — culture, manager support, talent practices — account for roughly twice the reported impact of individual factors like mindset and skill. This is not a technology problem you can buy your way out of. It chimes almost exactly with what we found.
So the diagnosis is sound. My disagreement is with the prescription's underlying mental model — and it is one disagreement, repeated in four forms.
Readiness is not a gate
Here is the heart of it. The report maps every surveyed worker onto two axes — individual capability with AI, and the organisation's readiness to absorb it — and sorts them into five zones. It then describes the resulting picture as a “Transformation Paradox” that is “at its core a systems problem, and systems don't fix themselves — they have to be redesigned.” The logic runs: redesign the system, then the value flows.
That is a gate. It says readiness is a state you reach, a configuration you lock in, after which adoption can proceed.
The tell is that Microsoft does not actually believe this. Its own thought leadership — the WorkLab essay Three Things Frontier Firms Understand About AI — states plainly that “the frontier is a practice, not a place,” that Frontier Firms “operate in perpetual beta,” that there is “no finish line for AI adoption.” That is exactly right, and it directly contradicts the gated framing of the report. The house view knows it is a practice. The report's measurement model and its prescriptions still treat it as a gate.
Our research lands firmly with the WorkLab essay and against the report. The model that came out of twenty interviews across eighteen UK SMEs — what I call the Leadership Sensegiving–Readiness Heuristic, or in plainer terms, a repeating cycle of three stages — works like this. Leaders set the interpretive conditions: what they frame, model, permit, and bound. People then make their own sense of those signals through trust, emotion, identity, and the stories they tell each other. And readiness emerges out of that — but as a provisional, uneven state, not a finished one. Then it loops: what people experience feeds back into what leaders choose to frame next.
The crucial word is provisional. Readiness is not a gate you pass before adoption. It develops through use, and it can stall or reverse if trust, time, or boundaries collapse. A team that felt ready in March can be set back to scepticism by one badly handled hallucination in April.
This matters to a leader for a very practical reason. If you believe readiness is a gate, you spend your scarce attention on the things that look like getting ready — the strategy document, the policy, the platform decision — and you under-invest in the things that actually move people: visible modelling, routines for sharing what works, and guardrails that make experimentation safe rather than forbidden. The report's own data shows the cost of this. Only 13% of AI users say they feel rewarded for reinventing how they work, regardless of whether it pays off immediately. That is not a gate that has failed to open. That is a loop that is not turning.
You don't get ready, then adopt. You get ready by adopting — carefully.
The “paradox” is really a spectrum
The report's central construct is the Transformation Paradox: employees ready to reinvent how they work, held back by systems — metrics, incentives, norms — that reinforce the old ways. The lived tension it describes is real. The people already using AI genuinely are running ahead of the organisations around them.
But “paradox” is the wrong word, and the report's own numbers show why. A paradox would be a real contradiction — high individual capability trapped inside low organisational readiness. The report has a name for exactly that group: “blocked agency.” It is 10% of AI users. The mirror image — high organisational readiness but low individual capability, “unclaimed capacity” — is 5%. The other 85% is not paradoxical at all.
So the honest description is not a paradox but a spectrum — a continuum of sensemaking maturity. And a spectrum is a far more useful object for a leader than a paradox is. A paradox tells you that you are stuck in a contradiction. A spectrum tells you where you currently sit and what the next move is. The work is not to escape a trap; it is to advance one position along a curve, and then the next.
Calling it a paradox makes it sound like a trap. It is a position on a curve — and curves have a next step.
The report lets leadership off the hook
This is the one I feel most strongly about, because it is where the report's enterprise assumptions quietly fail the firms most of my clients actually run.
The report makes a clean split: “leaders set strategy at the top,” and “it's managers who operationalise it.” Then it produces its most striking evidence — and that evidence is all about managers. When managers actively model AI use, employees report a 17-point lift in the value they get from AI and a 22-point lift in critical thinking. When managers create psychological safety around experimentation, readiness rises by up to 20 points. The data on modelling is excellent. But the people at the top have been quietly excused from doing any of it. They set the strategy; someone else is seen to use the tools.
In a large enterprise, you can almost get away with that division of labour. In an SME — the founder-led, sixty-person firm where the leader is the strategy — you cannot. Our research was unambiguous on this point: in smaller organisations, proximity magnifies leadership cues. Silence is not neutral. When the boss says AI matters but is never seen to touch it, the silence becomes the signal — that this is risky, or premature, or not really meant. The higher-performing firms in our study were not the ones with the best strategy decks. They were the ones where leaders demonstrated real workflows in front of their people.
The report itself half-admits the problem: only 26% of AI users say their leadership is clearly and consistently aligned on AI. The fix is not another memo. It is the person at the top being visibly, regularly, imperfectly in the work.
A strategy memo nobody sees the boss act on is just a memo. Curiosity is contagious; distance breeds scepticism.
Saving time is not transformation
Lakhani's move from business model to operating model is the right one, and the report's emphasis on redesigning how work gets done — how it is divided, where judgement sits, how expertise is codified — is exactly where most of the near-term value lives. But the report largely stops there, and that is its quiet ceiling.
Productivity gained at the edge does not automatically become enterprise advantage. If you save an analyst three hours a week and do nothing deliberate with those hours, you have not transformed anything — you have shifted a little cost from a person to a machine. The value only appears when the time saved is reinvested into better decisions, redesigned processes, or work that simply could not be done before. The earlier research on AI and productivity is sobering here: time saved that is not deliberately redirected tends to evaporate.
And there is a larger prize the report barely touches: not just redesigning the operating model, but rethinking the business model itself — the value you offer and the way you create and capture it — now that AI changes what is possible. That shift is harder to see and harder to capture, which is presumably why the report leaves it alone. But it is where the durable advantage will eventually sit.
Even the operating-model gains, though, depend on something the report frames too narrowly. Its Frontier Firm playbook rightly tells leaders to “build your diffusion engine” — but it frames diffusion as an engineering exercise, an operating model built to scale. Our research says diffusion is a social act before it is a technical one. Individual power-users do not become organisational capability on their own. It takes champions, show-and-tell sessions, shared prompt libraries, and guardrails that preserve trust — the human routines through which a local win becomes a shared practice. Microsoft will help you build the engine. Someone still has to win the hearts that run it.
Saving time is not transformation. What you do with the time — and whether you redesign the work, or eventually the business — is.
A note for the board: how to read this report well
If you are going to cite the Work Trend Index to your own leadership team — and it is worth citing — read it with five things in mind. None of these undermines the report. They simply keep you from reading more into it than it can bear.
-
i
It surveys the already-converted. The findings come from 20,000 workers who already use AI at least occasionally. Non-users were screened out before the count began. So every figure describes the world as seen by people who have already crossed the line. It tells you nothing about the colleagues who have not.
-
ii
“Frontier Professionals” are the elite of that group, not the norm. They are 16% of AI users — and they skew heavily towards tech (35%) and financial services (12%), towards millennials, and towards larger organisations (over half are in firms of 500-plus employees). These are the people building multi-agent workflows and posting about it on LinkedIn. They are real, and they are instructive, but they are loud, few, and disproportionately big-company. If you run an SME, do not benchmark your firm against this vanguard and conclude you are behind.
-
iii
It is a Copilot-centric view. The data leans on Microsoft 365 Copilot usage. Firms whose people have gravitated to ChatGPT or Claude may be moving differently, and faster, in ways this report does not capture.
-
iv
Mind the word “Frontier.” The report uses it three ways with three different denominators — a zone (19% of AI users), a capability group (“Frontier Professionals,” 16% of AI users), and a share of total workers (16%) in its country tables. They are not the same population. Read the footnotes before you quote a number.
-
v
The national differences are large and under-explained. Frontier-professional share swings widely by market: Brazil sits at 27%, against 16% in the UK and 17% in the US, and single digits to low teens across France, the Netherlands, Italy, Germany and Japan. It is tempting to assume an English-language survey is flattering the anglophone markets — but the report states the survey was run online and translated into local languages, so that explanation does not hold. The spread is real; its causes, whether cultural, economic or structural, are not something the report explains. Don't over-read any single market's number.
The verdict is not “ignore it.” The verdict is: it is a trustworthy portrait of the frontier, not a census of the field. Read it as a picture of where the leading edge already is, and use your own judgement about how far your organisation sits behind it — and why.
What it all comes down to
Strip away the four disagreements and you find the report and I agree on the destination. The firm that wins is the one that becomes a genuine learning system — one that compounds its own hard-won, hard-to-copy intelligence and adapts faster than its competitors. That is Jim Collins's old insight in Good to Great, restated for the AI era: the durable advantage was never the technology itself, but the rate at which an organisation can absorb and apply it.
Where we part company is on the map, not the destination. The report draws the journey as an architecture problem — build the right organisation, pass through the gate, capture the value. Our research draws it as a practice — run the loop, continuously, with leaders inside it rather than above it, knowing the readiness it produces is always provisional and can always slip back.
So the question the report leaves leaders with — are you built to capture it? — is the wrong question, or at least an incomplete one. It sounds like something you answer once. The better question, the one we will keep asking, is: are you running the loop — and can the person at the top be seen running it?