Microsoft's 2026 Work Trend Index found that organisational factors account for 67% of the reported impact of AI at work, against 32% for individual mindset and behaviour, and that 10% of AI users sit in blocked agency: skilled, and unsupported. Read through the AQ A.C.E. model, that is an Environment constraint being misdiagnosed as a skills gap.
If you use AI every day and your job still feels the same, the most likely explanation is not you. It is the system you are working inside. Microsoft's 2026 Work Trend Index, a survey of 20,000 knowledge workers who already use AI at work across 10 markets, ran a model against 29 different factors to see which ones actually predict whether a person reports real value from AI. Organisational factors, meaning culture, manager support and talent practices, accounted for 67% of the signal. Individual factors, meaning your own mindset and behaviour, accounted for 32%. Roughly two to one, in favour of the conditions around you.
That is the first hard number I have seen put on something we have argued at AQai for years without being able to weigh it: adaptability is not only a property of the person. It is a property of the person and the place, together. And when the place is not set up for it, capable people run into a wall and quietly conclude the wall is their fault.
Because using a tool and being allowed to redesign your work with it are two different things, and only one of them changes a job. Most people have done the first. Far fewer have been given permission, cover and time to do the second.
The Work Trend Index puts a shape on this. Microsoft mapped every respondent on two axes: individual readiness, which combines how sophisticated their AI use is, how confidently they direct it and judge its output, and whether they report creating new value with it; and organisational readiness, which combines governance maturity, manager support, whether AI shows up in how people are evaluated, and the wider culture around AI use. Sorting people into that grid produces five groups, and the distribution is the story.
Only 19% land in what Microsoft calls the Frontier, where individual capability and organisational readiness are both high and reinforcing each other. 16% are stalled, low on both. 5% sit in unclaimed capacity, where the organisation is ready and the person has not caught up yet. Half of everyone, 50%, sits in an emergent middle where both are still forming. And 10% are in the group that matters most for this conversation, which Microsoft names blocked agency: people who have built real skill with AI and do not have the systems, permissions or support to apply it.
If you are reading this because you are frustrated, that 10% is probably where you are.
Blocked agency is the condition of being individually capable and organisationally constrained. You know what the tool could do. You can see the workflow that ought to replace the one you have. You are not permitted, resourced or rewarded to build it.
It shows up in a few recognisable ways. You use AI to produce the same deliverable faster, and the time you save gets absorbed rather than reinvested. You have proposed a different way of doing something and been told to focus on the plan. Nobody has told you what good AI-assisted work looks like here, so you set your own bar and carry the risk alone. Your manager does not use these tools in front of you. And nothing in how you are measured this quarter has any relationship to whether you changed how the work gets done.
None of that is a capability signal. All of it is a conditions signal.
It is Microsoft's name for the trap at the centre of all this: employees are ready to reinvent how they work, and the system around them, meaning the metrics, incentives and norms, keeps rewarding the old way. The same pressure that is driving AI adoption is holding it back.
Three findings from the survey make the squeeze concrete. 65% of AI users say they fear falling behind if they do not use AI to adapt quickly. 45% say it feels safer to focus on hitting current goals than to redesign their work with AI. And only 13% say they are rewarded for reinventing work with AI when the results do not immediately land.
Read those three together and the behaviour of a very large number of sensible people becomes obvious. You are told change is urgent. You are measured on this quarter. Reinvention that does not pay off inside the measurement window costs you. So you use AI to go faster at the thing you are already being marked on, and you leave the redesign alone. That is not resistance and it is not a lack of ambition. It is a rational response to the scoreboard you were handed.
The leadership picture underneath is thin in the same direction. Only 26% of AI users say their leadership is clearly and consistently aligned on AI. That is one in four. The other three quarters are being asked to transform their work by people who have not agreed with each other on what transformation means.
Their environment, far more than their enthusiasm. Microsoft identifies a group it calls Frontier Professionals, the most advanced AI users in the research, who use agents for multi-step work, routinely redesign workflows around what AI does well, and take part in setting shared standards for their team. They are 16% of the AI users surveyed, and 80% of them say they are producing work they could not have produced a year ago, against 58% of AI users overall.
The interesting part is what surrounds them. Compared with everyone else, Frontier Professionals are far more likely to say their manager openly uses AI (85% against 64%), sets quality standards for AI-assisted work (83% against 57%), creates space to experiment (84% against 61%), and encourages more ambitious redesign of the work (87% against 61%). They are twice as likely to say reinvention is rewarded regardless of the immediate outcome (26% against 11%).
Every one of those is something a manager does, not something a person is. A separate Microsoft study of 1,800 workers found the same pattern from the other direction: when managers actively modelled AI use themselves, their people reported a 17 point lift in the value they got from AI and a 22 point lift in critical thinking about their own AI use. Where managers created safety around experimenting, reported AI readiness and value rose by up to 20 points.
There is a personal discipline in this group worth naming, because it cuts against the caricature. Frontier Professionals are more likely than others to deliberately do some work without AI to keep their own skills sharp (43% against 30%), and to pause before starting to decide what should be done by a human and what by AI (53% against 33%). They are not outsourcing their thinking. 86% of all AI users in the survey said they treat AI output as a starting point rather than a final answer and stay responsible for the thinking, and this group holds that line hardest.
Three things are worth holding, because these numbers get quoted carelessly.
Microsoft sells Copilot. This is vendor research, carefully done and openly documented, but it is not disinterested. Treat the exact percentages as directional rather than precise.
Everything here is self-reported by the same person at the same moment, which Microsoft states plainly in its own methodology: the relationships shown are statistical associations, not causal effects. A supportive manager and a person who reports getting value from AI travel together. That does not prove which one produced the other.
And the survey only includes people who already use generative AI at work. Anyone who answered "never" was screened out. So this describes the experience of AI users, not the whole workforce, and the 10% in blocked agency is 10% of that narrower group.
None of that undoes the finding. It sizes it honestly.
AQ, or Adaptability Intelligence, is the measurable capacity to adapt to change. AQai measures it across three domains, Ability, Character and Environment, and what makes this dataset useful is that it maps onto all three at once while pointing hard at the one most organisations ignore. You can see how the three fit together in the AQ model.
Ability covers the mechanics of adapting, including Unlearn, Mental Flexibility, Mindset, Grit and Resilience. Unlearn matters here more than almost anything, because AI does not ask you to add a tool to your method, it asks you to drop a method that used to work and rebuild the job around a different one. We have written before that unlearning is the metric that predicts whether an AI rollout lands, and I still think that is right at the level of the individual.
What this dataset adds is the ceiling. Unlearning is necessary and it is not sufficient. You can have high Unlearn and high Mental Flexibility, do exactly the right thing with them, and produce nothing visible, because the organisation has nowhere to put the result. That is what blocked agency is: full Ability meeting an Environment that cannot absorb it.
Character covers the more stable patterns in how you engage with change, including emotional range, thinking style, motivation style and hope. Hope in the AQ model is not optimism. It is the belief that there is a route from here to somewhere better, and some agency over walking it.
The 45% who say it feels safer to hold to current goals than to redesign the work are not short of ambition. They are short of a visible route where redesign pays. Tell that group to be more open to AI and you are talking past the problem. Show them one colleague whose redesigned workflow was recognised even though the first attempt underperformed, and you have changed the calculation, because now the route exists.
Environment covers the conditions you are adapting inside, including work stress, team support and company support. This is the domain most models leave out entirely, treating adaptability as a purely individual trait, and it is exactly the domain the Work Trend Index just measured as carrying twice the weight of the individual one.
Read the numbers as an Environment profile and they are unambiguous. 26% leadership alignment is a company support reading. Managers who do not model AI, do not set a quality bar and do not make space to experiment are a team support reading. And a reward system where only 13% see reinvention recognised without immediate results is company support telling you, precisely and correctly, not to bother.
More than nothing, and less than everything. The honest position is that you cannot fix an Environment constraint on your own, and you can stop paying the cost of misreading it.
Start by naming it accurately. If you have been telling yourself you are behind on AI, check that against the evidence. If you can direct these tools, judge their output and see the redesign that should happen, you are not behind. You are blocked, and those two things need completely different responses.
Then make the redesign legible rather than invisible. Blocked agency stays blocked partly because the work happens in private. Pick one workflow, rebuild it, and write down what it used to cost and what it costs now. A specific before and after is a very different conversation from a general enthusiasm for AI, and it is the kind of thing a manager can actually act on.
Ask for the quality bar rather than waiting for it. In the absence of a stated standard for AI-assisted work, most people invent a private one and carry the risk of being wrong alone. Asking your manager what good looks like here is a small request that moves a genuine constraint, and it is one of the few Environment levers an individual can reach.
Protect the skills you are not using. The 43% figure is a habit worth copying: deliberately do some work without AI, so that the judgement you will be evaluated on later does not quietly erode while you wait for your organisation to catch up.
And be realistic about the last option. If you are demonstrably capable, in an organisation with no leadership alignment and no reward for changing how work is done, the environment may simply not be one you can adapt inside. That is a legitimate read, and it is worth making deliberately rather than by exhaustion.
That the bottleneck has moved, and most AI investment is still pointed at the old one. If organisational factors carry roughly twice the weight of individual ones in whether people get value from AI, then another round of tool training is aimed at the smaller half of the problem.
The three levers the data points at are unglamorous and specific. Align the leadership team on what AI is for, visibly, because one in four is not a mandate. Ask managers to use these tools in the open, set a quality bar for AI-assisted work and make room to experiment, because that cluster of behaviours is what separates the group whose jobs have changed from the group whose jobs have not. And change what you reward, because as long as reinvention only counts when it works first time, sensible people will keep choosing the safe quarter over the better system.
There is a diagnostic question underneath all of this that is worth asking before you spend anything: how much of what is stalling here is capacity, and how much is conditions? An organisation full of people in blocked agency will look, on every adoption dashboard, exactly like an organisation with a skills problem. It is not one. And the cost of getting that wrong is not just wasted training budget, it is the slow loss of the people who were ready. If you want a read on where your own people sit, AQme measures Ability, Character and Environment separately, which is the distinction this whole argument turns on.
Because using AI and being able to redesign your work with it are different things. Microsoft's 2026 Work Trend Index found that organisational factors account for 67% of the reported impact of AI at work against 32% for individual mindset and behaviour, and that only 13% of AI users say they are rewarded for reinventing work when the results do not immediately land. If nothing in your environment rewards redesign, AI tends to make the existing job faster rather than different.
Blocked agency is Microsoft's term for people who have built real capability with AI but lack the systems, permissions or support to apply it. It accounted for 10% of the AI users in the 2026 Work Trend Index. In AQ terms it is high Ability meeting a constrained Environment, and it is frequently misread by both the person and the organisation as a skills gap.
The 2026 Work Trend Index data points at culture and management practice more than individual skill, by roughly two to one. That does not make training useless, since capability still has to exist. It does mean that training alone, in an organisation with no leadership alignment and no reward for changing how work is done, is aimed at the smaller half of the problem.
Unlearning still matters, and it is not enough on its own. AI asks people to drop a method that used to work rather than add a tool to it, which is an Unlearn demand in the AQ Ability domain. What this research adds is the ceiling: a person with strong unlearning capacity in an environment that cannot absorb a redesigned workflow will produce no visible change.
The survey covered 20,000 knowledge workers who use AI at work across 10 markets, fielded by Edelman Data x Intelligence for Microsoft in early 2026. Three caveats matter. Microsoft sells AI products, so treat the exact percentages as directional. All measures are self-reported by the same respondent at the same moment, so Microsoft itself describes the relationships as statistical associations rather than causal effects. And people who never use AI at work were screened out, so the findings describe AI users rather than the whole workforce.
Copyright adaptai Ltd. All rights reserved. AQ® and AQai® are registered trademarks of adaptai Ltd.
.png){kind=small}
.png)
