The AI Opportunity Gap Is Real. It’s Growing. And It Is Not About Access to Tools.

In the first part of this series, I argued that discernment, the ability to recognise when an AI-generated answer is wrong, is becoming one of the most valuable capabilities inside an organisation.

The question this piece addresses is simpler and harder: who is actually being given the opportunity to develop it?

The AI opportunity gap is real. It is not primarily a gap in access to tools. It is a gap in permission. And I believe that gap starts earlier than most leaders realise, often in school.

Access is not fluency

AI tools are widely available, though inside large organisations, rollout is rarely instant or even. Some teams get access before others, and cost or licensing tier still gates who gets which tool.

The organisations building genuine AI capability are not necessarily the ones that moved fastest on procurement. They are the ones where a meaningful proportion of the workforce has developed real fluency: the ability to apply AI to actual work, assess its output, understand its limits, and know when human judgment needs to take over.

That fluency does not develop through training programmes. It develops through use. Through trying an approach that fails, understanding why, and trying again. That process requires time without an immediate deliverable attached and room to experiment without every unsuccessful attempt being treated as wasted effort.

The people developing real AI fluency right now are those with discretionary time to explore. Others work in roles where output is measured closely and learning is not, where heavy workloads, rigid processes, shift patterns, and responsibilities outside work leave no margin for the kind of low-stakes experimentation that builds genuine judgment.

BambooHR’s 2025 research found that 72% of C-suite executives use AI daily, compared with 18% of individual contributors. Separately, only 32% of employees have received formal AI training, despite 72% wanting to improve their AI skills. That is not a gap explained by access to tools or by individual motivation. It is a gap explained by what organisations signal about whose time belongs to learning and whose time belongs to delivery.

Permission is not distributed equally

The asymmetry in how AI experimentation is socially coded across organisational levels is one of the least discussed dimensions of this problem.

A senior manager who spends an afternoon exploring an AI tool is seen as curious and forward-looking. An individual contributor who does the same may be seen as distracted. For a warehouse operative, field technician, or call centre worker, that time is simply classified as time off task. The social coding is not subtle and it is not accidental. It reflects embedded assumptions about whose development is an investment and whose is a cost.

This matters because judgment requires failure. People develop real AI fluency when they have had enough experience of AI getting things wrong to understand where its limits actually are. Formal training can create awareness. It cannot create the judgment that comes from working through failure in conditions where it is safe to do so.

The permission problem does not stop at knowledge workers. It runs deeper, into a structural gap that most organisations have not yet named.

The gap in authorship

Fluency assumes that the worker has some relationship with the AI output: the ability to read it, question it, and decide what to do with it. For a growing part of the workforce, that assumption does not hold.

Consider a warehouse supervisor working with AI-generated routing recommendations. A field technician receiving automated diagnostic guidance. A customer service representative following an AI-scripted workflow. These workers are not being asked to develop AI fluency. They are being handed AI-assisted systems as fait accompli, with no agency over what those systems optimise for, how exceptions are handled, or what happens when the person closest to the situation believes the recommendation is wrong.

Fluency is the ability to use AI effectively and judge its output. Authorship is the ability to influence how AI is applied to your work. Without authorship, the system makes the call and the human carries it out. The employee may see that a recommendation is flawed but the organisation has not created a mechanism for that judgment to change the outcome. That is not a skills problem. It is a governance and workforce design problem, and most organisations have not named it yet.

The asymmetry of risk across levels compounds this. For knowledge workers, failing to develop AI fluency risks career stagnation. For workers whose tasks are being directly automated or tightly directed by AI systems, the exposure is more severe: fluency was never offered as a hedge. The question for those workers is not how to use AI better but whether the organisation has any pathway for them to remain relevant and influential as their work is reshaped around them.

The compounding risk

Right now, the internal AI divide looks like a capability gap. Some people inside your organisation are developing genuine AI judgment. Most are not. That gap will show up in productivity, decision quality, and execution speed within two to three years.

But the compounding starts earlier than most leaders realise, and from a source outside their direct control. This divide does not start in the workplace. It starts in school. The OECD and European Commission published a shared AI literacy framework for schools as recently as June 2026, which tells you something important: until now, there was no common definition of what students should be able to do with AI by the time they leave school. The framework was developed in direct response to fast-rising AI use among young people, with the European Commission noting that 68% of teenagers are already using AI, often without guidance on how the systems work or where they fail. Meanwhile, China and the UAE have mandated AI education starting with the 2025 to 2026 school year. Most other countries have not. Students whose schools treat AI as a tool to learn with, not just a rule to enforce against, arrive at their first job already fluent. Students who do not get that chance arrive already behind. The organisational gap this piece describes is a second act. The first act happened in the classroom.

Over a five to seven year horizon it becomes something more structural. Workers who were never given a pathway to develop real AI fluency become a stranded cohort. Not because they failed to develop but because the organisation never created the conditions for them to do so. Meanwhile, the organisations that invested in cross-level AI literacy, not just for senior knowledge workers but across every level of the workforce, have built a talent advantage that cannot be bought or replicated quickly.

According to PwC’s 2026 AI Jobs Barometer, the pace of skills change in AI-exposed roles is now running at roughly twice the rate of non-AI-exposed roles. The window for organic catch-up is closing faster than most workforce planning cycles account for. There is also a growing divide between organisation types: employees in smaller or slower-moving firms are falling behind counterparts in tech-forward organisations at an accelerating rate. The gap is not only opening between individuals. It is opening between organisations. The organisations that are behind on this in 2028 will not be behind because they lacked the budget. They will be behind because they did not make the decision in time.

What the organisations getting this right are doing differently

The organisations closing the internal AI gap fastest are not the ones running the most training programmes. They are the ones whose leaders have made a deliberate decision to treat AI experimentation as legitimate work, not a distraction from it.

That means protected time for exploration without a required deliverable. It means treating a failed AI experiment as a learning outcome rather than a performance issue. It means extending permission explicitly and visibly to the people whose jobs are being reshaped most directly by AI, not just to the technical teams and senior leaders who would find space to experiment anyway. And it means creating genuine mechanisms for those people to influence how AI systems are used in their work, not just to operate them.

At Qlik we have moved beyond making courses available. We recently launched a dedicated internal microsite giving every employee a direct route into practical AI application. Each function has designated AI Champions, people whose role is to help colleagues find use cases that work in their specific context, share workflows that are delivering results, and escalate questions that need more than a quick answer. The principle is simple: fluency develops through use, not through completion rates.

The organisations building the most durable AI capability are treating it as infrastructure, something developed deliberately across the whole workforce, not as a procurement decision or a training exercise. The organisations that treat it as the latter will spend the next five years watching the internal capability gap widen even as their AI spend increases.

Three things leaders should do now

The internal AI divide will not close on its own. Closing it does not require new technology or additional budget. It requires decisions about what counts as a legitimate use of people’s time.

• First, make permission explicit. Give people protected time to test AI against real work, including work that may not produce an immediate result. Extend that permission visibly to the people whose roles are being reshaped most directly by AI, not just to those who would find discretionary time anyway.

• Second, treat learning as an outcome. If every AI experiment is required to deliver a measurable productivity gain, people will avoid the uncertain, exploratory work through which real judgment develops. Discovering where AI fails, where it introduces risk, and where it requires human oversight is valuable work and should be recognised as such.

• Third, give employees authorship, not just access. Create clear mechanisms for the people closest to the work to question AI-generated recommendations, flag where systems are producing wrong or harmful outputs, and influence how those systems develop over time. Access without authorship produces dependency. Authorship produces judgment.

The gap between the people inside your organisation who are developing real AI capability and those who are not will be one of the most consequential competitive variables of the next five years. It will not be measured in tools, licences, or training completion rates. It will be measured in the decisions your organisation can and cannot make well.

AI fluency develops where people have permission to experiment. AI is used responsibly where those same people have authority to question it.

Every month this goes unaddressed, the people inside your organisation who are developing real AI judgment pull further ahead of those who are not being given the chance to.

That distance is recoverable now. In year six, it will not be.