Calling AI a “superpower” may sound exaggerated, but in some ways it is becoming a reasonable description. Two people can have the same job title, similar experience and work the same number of hours, yet they produce very different amounts of work because one has learned how to use AI effectively, while the other has not.
One person may still write reports from scratch, read every document manually, organise their own notes, prepare meeting papers, search for information and draft routine correspondence in the traditional way. Another may use AI to summarise documents, compare sources, identify themes, draft first versions, restructure reports, prepare briefing notes and analyse information. They may also create reusable prompts, templates and automate workflows to remove repetitive work.
From the outside, the second person may look extraordinarily efficient. That creates an important management problem. If somebody completes in two hours what previously took a day, what exactly are we observing? Are they more intelligent, more experienced, better organised, working harder, or simply much better at using AI? The answer may be some combination of all of these, but managers may have no reliable way of knowing.
Two classes of worker?
AI competence is not evenly distributed. Some employees are experimenting constantly. They test new tools, follow developments, learn from others and continually refine the way they work. Others may have tried ChatGPT once or twice and concluded that it was not particularly useful, felt uncomfortable using it, or simply failed to see the benefits.
The result is that people doing broadly similar work may now have very different levels of productive capacity. That does not necessarily mean that the AI-enabled employee is producing better work. Productivity is not simply about volume. AI can also produce errors, poor judgement and superficially impressive output that does not stand up to scrutiny.
However, when AI is used well, the productivity difference can be substantial. The danger is that organisations may interpret that difference as a difference in employee quality rather than a difference in capability, working methods or access to training.
Why would people share what they know?
We often assume that useful knowledge will spread across an organisation, often being expressed as sharing best practice. Somebody discovers a better way of doing something, informs colleagues, and everyone benefits. But AI may create a more complicated incentive.
Suppose I develop a workflow that saves me five hours every week. If I share it, my colleagues become more productive and the organisation benefits, but I also lose an advantage that makes my own performance stand out. The rational organisational behaviour and the rational individual behaviour may therefore diverge.
This becomes particularly relevant when performance reviews, promotion, bonuses or job security are involved. If AI allows somebody to produce more work, respond more quickly and take on additional responsibilities, there may be a strong incentive to keep some of that know-how private. Organisations may therefore be experiencing more than an AI skills gap. They may also be creating private pockets of productivity that management cannot see, colleagues cannot easily replicate and the organisation does not fully understand.
AI capability does not stand still
There is another problem. AI competence is not something that can be learned once and then regarded as complete.
Consider Microsoft Excel. VLOOKUP was widely used for many years. XLOOKUP eventually provided a better way of solving many of the same problems, but the change was relatively contained. Someone who understood spreadsheets did not suddenly become obsolete because they had missed a few months of product development.
AI is different because the capabilities are changing rapidly. A person who became proficient at using a chatbot eighteen months ago may now be far behind somebody who is using multimodal tools, deep research, connected applications, automated workflows and AI agents. The gap is therefore not simply between users and non-users. There may be a widening gap between basic users, competent users and people who are continuously redesigning their work around AI.
Keeping up with AI developments can feel almost like a part-time job in itself, and that has implications for how organisations think about training. A one-off workshop on generative AI may have value, but it cannot solve a problem in which the underlying technology is changing every few months.
And then there is agentic AI
This may be where the productivity gap becomes much larger. Most people still think about AI at the level of individual tasks: draft this email, summarise this paper, improve this paragraph, analyse this spreadsheet.
Agentic AI moves the discussion towards sequences of work. An AI system may be able to monitor information, prepare updates, organise tasks, draft responses, maintain project records, coordinate calendars and trigger routine actions with relatively little intervention. That is a very different model of work.
The employee who occasionally asks AI to improve an email and the employee who has delegated large parts of a recurring workflow to AI may both say that they “use AI”, but they are not doing anything remotely comparable. The productivity implications could be significant.
Do managers know what productivity looks like anymore?
This brings us to what I think is the most difficult issue. Many organisations still assess performance using assumptions that were formed before generative AI existed. We look at output, responsiveness, workload, deadlines and quality. We compare one employee with another and make judgements about who appears highly productive and who appears to be struggling.
But if AI use is largely invisible, those comparisons become increasingly unreliable. A manager may praise one employee for producing twice as much as a colleague without realising that the two are effectively working with very different technological capabilities.
That does not mean the more AI-enabled employee deserves less credit. Learning how to use new tools effectively is itself a capability, and we would not criticise somebody for being better at Excel, statistical software or database tools. AI is unusual, however, because the difference in capability can be much larger, much less visible and much harder for a manager to evaluate. In some cases, the manager may know less about the technology than the employee being assessed.
The real divide may be invisible
The discussion about AI and employment often focuses on whether jobs will disappear. That is clearly important, but another change may already be happening inside jobs that continue to exist. People with the same job title may no longer have anything like the same productive capacity.
Some will use AI occasionally. Some will use it extensively. A smaller number will continually redesign the way they work around new AI capabilities. The differences between those groups may grow as the technology develops, but the danger is not simply that some people will become more productive than others. That has always been true. The danger is that organisations may not understand why.
If AI capability remains largely invisible, managers may confuse technological advantage with individual performance, employees may have incentives to keep useful workflows to themselves, and colleagues may be compared on the basis of very different working methods. We may therefore be entering a period in which one of the most important productivity differences in the workplace is also one of the hardest to see.
If two people appear to be doing the same job, but one has quietly multiplied their productive capacity through AI, do we really know which of them is performing better? Or do we first need to understand how the work is actually being done?
If you liked this edition
I publish two free newsletters each week.
Governance, Leadership & Research
A weekly brief for higher education leaders and researchers examining governance, leadership and research in practice.
https://buff.ly/GhsZqVl
Publishing with Integrity
Examining scholarly publishing through an ethical lens, challenging assumptions and supporting academic career development.
https://buff.ly/WiG1yFb
If you subscribe, you will be alerted whenever I publish a new edition.
About the author
Professor Graham Kendall is Vice-Chancellor of GlobalNxt University, Malaysia, and an Emeritus Professor at the University of Nottingham. He has held senior leadership roles across higher education, including Vice-Chancellor, Provost, Pro-Vice-Chancellor and Vice-Provost. His interests include university governance, leadership, research, artificial intelligence and scholarly publishing.