The Augmented Mindset

The augmented mindset is the only one of the seven that is about the tools. It is also the one most likely to be read as enthusiasm for them, which it is not.

It is the cultivated ability to work with a machine in a way that extends what you can do without handing over the part that decides. The distinction that matters is not how much AI somebody uses. It is whether they can still tell when it is wrong.

What the evidence supports#

The central finding is a meta-analysis of 106 experiments comparing humans alone, AI alone, and the two together. On average, human and AI pairs performed worse than the better of the two on its own.

The average hides the useful part. In creative and generative tasks, adding AI tended to improve the result. In analytic decision tasks, human oversight often failed to correct the machine's errors. Where the human was naturally better, the pair beat either alone. Where the AI was better, adding a human reduced performance. The variable is not effort. It is whether the person knows which of the two to trust on this particular question.

A study of 5,172 customer support agents found productivity up 15 per cent on average, with a 30 per cent improvement among novice and low-skilled staff and almost none among the most experienced. Those workers followed about 38 per cent of the system's recommendations. They were not deferring. They were choosing.

The cost of not choosing is measured too. When an AI suggested incorrect mammogram assessments, radiologists often deferred and accuracy dropped. In primary care, AI decision support changed prescribing in roughly one in five cases, and in about one in twenty the machine's bad advice turned a correct decision into an incorrect one.

What the evidence does not support#

That the mindset can be trained. Nothing here tested an intervention and measured a disposition afterwards.

It also does not support the reassuring reading of the 38 per cent. Those agents rejected most of what they were offered, and no study has shown that a person who starts out deferring can be moved to that position. The support agents who benefited most were the least experienced, which is the finding that cuts against the story everybody prefers: the tool helped the people with the least to check it with.

What AI changes#

The judgement being asked for is second-order. Not is this answer right, but is this the kind of question this system is good at. That is the jagged frontier, and it cannot be read off the output, because the output looks the same either way.

The failure mode is automation bias, and the reason it is hard to catch in yourself is the METR result: developers were 19 per cent slower with the tool and believed they were 20 per cent faster. The thing that failed first was not the work. It was the ability to judge the work.

What absence looks like#

A person who is fast, confident and correct while the system is correct. Nothing distinguishes them from a person who would have caught it, until something needs catching.

Practising the Augmented Mindset#

The page above describes a disposition. This section is the set of practices, and all of it is drawn from SuperSkills (Kogan Page, 2026), chapter seven, with one later addition.

Map the terrain before the project, not during it. Three lines, written down: the patterns we trust the system on first, the calls we keep, and the decisions where an override always runs. That is the three terrains, and writing it in advance is the whole of it, because afterwards the answer bends towards whatever is convenient.

Keep four receipts. Hand, Head, Hours, Heart: what stayed human and who will answer for it, what was learned from a miss and what changed, where the saved time actually went, and who bears the consequence including the people who never enter the system as data. Four short lines, kept as the work happens rather than reconstructed for an audit.

Ask six questions about the piece of work in hand. Can I check it. What happens if it is wrong and I do not notice. Is this a repetition I need. Would doing it myself teach me anything. Has the model seen this before. Who answers for it. That is the use-or-keep test, the task-level companion to the terrain map.

Give one person sixty seconds. The pause minute at the close of any meeting: who kept the decision human, what did we learn and change, where did the saved time go. Called by anybody in the room rather than by the chair, because the cost of asking is what usually stops the question.

Write the rules before you open the tool. Before any system that thinks alongside you, write down what will never be delegated. Not the summary and not the first draft. The apology when something breaks. The consent when the stakes are real. The boundary that holds under pressure. The final call when the cost falls on another life. Tools change next quarter; those lines do not.

Keep some work unaided. Not for the output. A person who never produces anything of their own has nothing for a senior colleague to correct and nothing to compare their own reasoning against, which is the mechanism described at cognitive apprenticeship.

What none of this establishes#

None of these practices has been tested against outcomes. No study has compared people or teams who keep the receipts against those who do not, and the adjacent evidence on structured review supports something longer and more facilitated than a sixty-second version. They are offered as a discipline with reasoning behind it, not as a measured intervention.

Assessing the Augmented Mindset#

Assessment here is evidence rather than self-report, for the reason the failure modes section already gives: skill substitution means a person can use the machine's performance as a proxy for their own capability and not notice the substitution. A self-rating asks the person with the least reliable view.

What can be observed instead, organised by the ten subskills that sit under this SuperSkill:

• Human-AI collaboration, automation oversight. Can they produce a Hand line for recent decisions, naming who kept the call? A person exercising oversight can. A person approving cannot, and the gap appears without anybody having to make an accusation.
• Critical thinking with tech outputs, AI-assisted decision making. Do their Head entries exist, and do they keep appearing? A record that stops finding misses usually means somebody stopped looking rather than that the system stopped being wrong.
• Technological literacy, digital tools mastery, tool selection and integration. Malleability across projects, which is the test the section above gives: somebody who carries their way of working into an unfamiliar problem with a different system has the mindset, and somebody excellent inside one setup who starts from nothing when it changes has a configuration.
• Prompt crafting. Whether they can say what they changed in the machine's output and why. The answer distinguishes editing from accepting. It is what the shared prompt review puts on the table.
• Data interpretation and application. On statistical terrain, whether they interpret the signal or compete with it. Competing with the model on its own ground is the failure that gets mistaken for judgement.
• Responsible tech use. Whether a Heart line exists at all for decisions that reach beyond the room, and whether the people named in it were ever asked.

What this assessment cannot do#

It is not scored, not banded and not validated against anything. Records can be performed: a Hand line can name somebody who rubber-stamped, and a Head entry can describe a miss that cost nothing. It makes judgement legible where it exists rather than creating it, and it will not by itself catch an organisation determined to look compliant. The wider question of what assessing a disposition can establish is treated at how you assess capability rather than output.

The defining capability#

There is a reason this one comes last of the seven. Curiosity drives exploration of what AI can do. Change readiness allows adaptation to evolving tools. Big picture thinking places AI within larger systems. Empathy keeps the human element central. Global adaptability navigates how AI is deployed and received. Principled innovation provides the ethical grounding. But all of these require a final integration: the capacity to actually work with AI, day after day, in the specific tasks that constitute professional life. The professional who develops it does not merely survive technological change; they extend their reach, accelerate their learning, and tackle problems that would have been intractable alone. They are not competing with AI. They are compounding with it.

Key research and primary sources

• Vaccaro, M., Almaatouq, A. and Malone, T. (2024). When combinations of humans and AI are useful. Nature Human Behaviour, 8.
• Brynjolfsson, E., Li, D. and Raymond, L. R. (2023). Generative AI at Work. NBER Working Paper 31161; Quarterly Journal of Economics, 140(2), 2025.

Every figure on this page has been checked against the source that reports it. Where a number could not be confirmed, it was removed rather than left standing, and what was removed is recorded in the build register.

About this research#

Written by Rahim Hirji, author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company.

How this research works · Reviewed quarterly · Found an error? Tell me and it is corrected on the page.

Listen. Heard on AI for Business Leaders, The Fourth Dimension: Building Human Capability into the AI Strategy with Rahim Hirji. Heard on Education Futures, SuperSkills: The 7 human skills AI can't replace (with Rahim Hirji). Heard on The Entropy Podcast, SuperSkills for the AI Age with Rahim Hirji. Heard on Inside Learning, Super Skills: The 7 Human Skills for the Age of AI with Rahim Hirji. Heard on Nothing Ventured, Are We Raising a Generation That Can't Think? | Rahim Hirji. Heard on Solutionary Voices, How to Resist Algorithmic Drift.

• Curiosity
• Change
  readiness
• Principled
  innovation
• Global
  adaptability
• Empathic
  communication
• Big
  picture thinking
• Augmented
  mindset