Workforce reskilling and leadership — question bank
38 standalone questions on the people half — the novice paradox, the jagged frontier, the three layers of reskilling, shadow AI, and what changes for a manager.
Audience: senior business leaders. No technical background assumed.
Source: the reskilling and leadership handout — the novice-gain paradox, the jagged frontier, the three layers of reskilling, shadow AI, the AI-literacy obligation and what changes for a manager.
How to use this: every question stands alone. Pick an option, then read the answer. The nine sections run in the order the argument builds: the paradox, the evidence, what reskilling means, the scale, what your workforce is already telling you, the legal floor, the manager's job, programme design and measurement.
A · Step 1 — The paradox
Q1A study of 5,179 customer support agents given a generative AI assistant found productivity rose 14% on average. How was that gain distributed?
Answer: (2)The mechanism the researchers propose: the AI had absorbed the practices of the best agents and was handing them to everyone else. It also improved customer sentiment and staff retention.
Q2Since 2022, workers aged 22–25 in the occupations most exposed to AI have seen employment fall about 13% — roughly 20% in entry-level software roles — while experienced workers in the same fields held steady. Put alongside the previous finding, what is the paradox?
Answer: (2)Both findings are true simultaneously, and the room should be left to resolve it rather than told. It is the single best opening in this topic precisely because the resolution is not obvious.
Q3What resolves the paradox?
Answer: (2)The value of a junior goes up while the number of juniors goes down. Those two effects are not in tension; one causes the other.
Q4What is the question that should genuinely worry every leader once the paradox is resolved?
Answer: (2)You cannot have master craftsmen without apprentices. Every trade that stopped taking apprentices discovered the consequence twenty years later, by which time it was uncorrectable. An organisation that quietly stops hiring at entry level because "AI covers it" is eating its seed corn — and the cost lands on a successor, which is exactly why the decision gets made.
B · Step 2 — What the evidence actually says
Q5If AI mostly lifts your weakest performers, what happens to your performance distribution — and why does that matter?
Answer: (2)The gap between your best and your average narrows. Nobody's HR system is ready for this, and existing calibration processes built for a wider spread will start producing strange results.
Q6In an experiment with 758 management consultants, what happened on tasks inside the AI's capability?
Answer: (2)Genuinely excellent results — which is why the second half of the finding is so important, and why quoting only the first half produces predictable mistakes.
Q7On a task that sat outside the AI's capability, what happened?
Answer: (2)Not "no better" — worse. The tool actively degraded performance, and the people using it could not tell. That last clause is what makes it a leadership problem rather than a training problem.
Q8The frozen lake. What is the skill that actually matters?
Answer: (2)Some ice will hold a truck; some will not hold a child; from above it all looks identical. The dangerous zone is not where the AI obviously fails — it is where it fails while sounding exactly as confident as when it succeeds.
Q9A board is told AI will make the organisation twice as productive. What do the controlled studies actually support?
Answer: (2)Quoting these figures is a useful way to bring an over-excited board back to earth without being the person who is against AI — you are citing the strongest evidence in favour, at its actual size.
C · Step 3 — Reskilling into what, exactly
Q10"Reskill for AI" breaks into three layers. What are they?
Answer: (2)Layer 1 (everyone, hours): what the tools do, where the approved ones are, what you must never paste in. Layer 2 (anyone whose output matters, months and never stops): knowing when it is wrong. Layer 3 (managers and process owners): rebuilding a workflow rather than bolting AI onto the existing one.
Q11Which layer is "the whole game" and the one everybody skips?
Answer: (2)Layer 1 is a webinar. It is cheap, it looks like action, and it is what most organisations mean when they say they have "trained 8,000 people." It is also almost worthless on its own.
Q12What is counter-intuitive about the expertise required for Layer 2?
Answer: (2)Writing a mediocre contract clause requires competence. Spotting the one clause in a fluent, professional-looking, AI-drafted contract that would cost you ₹4 crore requires mastery. This inverts the usual assumption — and it is exactly why hollowing out the junior pipeline is dangerous.
Q13Where does your Layer 2 curriculum come from?
Answer: (2)Push for specifics — a number that must reconcile, a regulation that changed last year, a customer promise that cannot be made. That list cannot be bought from a training vendor because it is specific to you.
D · Step 4 — The scale of the task
Q14If the global workforce were 100 people, 59 would need training by 2030. How does that 59 break down?
Answer: (2)Stop on the 11. That is roughly one in nine of your people, and it is not a technology statistic — it is a decision organisations make, mostly by not making it.
Q15Which group is described as the harder management problem, and why?
Answer: (2)Upskilling in role has an obvious owner: the line manager. Redeployment crosses organisational boundaries, and a person who needs to end up somewhere else has no natural sponsor in either place.
Q16The share of workers' skills expected to be transformed by 2030 was reported as 39% — down from 44% two years earlier. What should you conclude?
Answer: (2)A caution worth voicing out loud, because quoting a forecast as a plan is how workforce strategies get built on sand. Note alongside it that 63% of employers name skill gaps as the biggest barrier and 85% plan to prioritise upskilling — intent is not the constraint.
E · Step 5 — Shadow AI is a signal, not a crime
Q17Roughly what share of workers use unapproved AI tools at work?
Answer: (3)And around 66% of office professionals used AI believing it was against company policy, with about half saying they would rather use it secretly than risk being told no.
Q18Which group are the heaviest users of shadow AI?
Answer: (3)Read that aloud and pause. The heaviest users of unapproved AI are executives — quite possibly some of the people in the room. It reframes shadow AI from a discipline problem into something the leadership is doing too.
Q19What is the correct reframe of shadow AI?
Answer: (2)You could not buy market research that good. And the actual failure is upstream: the approved path is worse than the unapproved one — too slow to get access, too restricted to be useful, or simply nonexistent. People do not route around a good system.
Q20Which is not one of the three things that work better than a ban?
Answer: (4)Blocking drives usage onto personal devices, where you see nothing at all. "Don't use AI" is ignored. "Never paste customer data, employee records, unreleased financials or anything under NDA into a tool that isn't on this list" is followed — because it is precise, short, and people can see the sense in it.
Q21What does the honest answer to "if one of your people found an AI tool that saved them a day a week, would they tell you?" reveal?
Answer: (2)And around 71% of shadow-AI users put sensitive data into those tools — customer details, employee records, internal documents. You cannot govern what you cannot see, and you will not see it while disclosure is punished.
F · Step 6 — The legal floor
Q22Article 4 of the EU AI Act, in force since February 2025, requires what?
Answer: (2)It prescribes no training format and no certification — the AI Office explicitly declined to impose one — and there is no obligation to test employees. Supervision and enforcement begin August 2026.
Q23There are no direct fines attached to Article 4. How should a board be advised?
Answer: (2)Treat it as the floor, not the target. The absence of a fine makes it tempting to ignore, and the liability exposure has nothing to do with the fine.
Q24Article 4 says the required level of literacy must vary by role. Why is that worth noting?
Answer: (2)The obligation must match the person's technical knowledge, experience and the context in which they use the system — which is role-based training, described in legal language.
Q25What is the one record-keeping recommendation, despite there being no measurement obligation?
Answer: (2)The practical advice from counsel. You are not required to prove competence; you should be able to show what you provided, to whom, and when.
G · Step 7 — What changes for a manager
Q26You can no longer tell whether a memo took four hours or four minutes, and asking feels like an accusation. What is the response that works?
Answer: (2)The question "did a machine write this?" measures the wrong thing and teaches people to conceal. The standard that survives is about the output, not the process.
Q27Which manager behaviour makes disclosure safe?
Answer: (2)If people fear that admitting AI use makes them look replaceable, or lazy, or both, they will hide it — and you lose both the learning and the control. The two questions produce measurably different organisations.
Q28Which is the most uncomfortable of the five changes, and should be said plainly?
Answer: (3)Not to a customer, a regulator, or a court. Verification is now part of the job description of anyone who signs anything off — and that has to be stated once, explicitly, and then applied consistently.
Q29Who has to redesign the workflow around a new tool?
Answer: (3)This is the 70% of the 10-20-70 rule landing on a specific person. Giving a team a licence and expecting the process to reorganise itself is the single most common failure in this module.
H · Step 8 — Designing a programme that works
Q30Why must training be role-based rather than company-wide?
Answer: (2)The thin-ice knowledge that constitutes Layer 2 is domain-specific by definition. Generic training can only ever deliver Layer 1.
Q31Who should teach Layer 2?
Answer: (3)It is not in a vendor's catalogue because it does not exist outside your organisation. Related: managers should be trained one cycle ahead — a manager who cannot do it cannot coach it, and will quietly undermine it.
Q32What is described as the highest-return item in programme design, and almost nobody does it?
Answer: (2)A library of near-misses builds judgement faster than any curriculum, and it costs nothing but the willingness to admit mistakes happened — which is a leadership decision, not a training one.
Q33"Learn it alongside your day job." What does that phrase actually mean in practice?
Answer: (2)Training on tools people cannot then use, on data they do not have, decays within a fortnight. Time protection is the difference between a programme and an announcement.
I · Step 9 — Measure it without fooling yourself
Q34You have trained 6,000 employees with 94% completion. Twelve weeks later, weekly active use of approved tools is 11%. What went wrong?
Answer: (3)Completion measured attendance, which is why it looked like success. And the common wrong conclusion — "our people are resistant" — is worth naming: they are not resistant; they were never given anything usable.
Q35What is the single best question for assessing whether Layer 2 has landed, askable of any individual at any seniority?
Answer: (3)A specific, confident answer means real skill. A blank look means they are trusting output they cannot evaluate — and that is your risk register, not your training register.
Q36Your graduate intake is 40 people. The COO proposes cutting it to 15 because AI now covers much of what juniors did in year one, and the saving is real and immediate. What is the defensible answer?
Answer: (3)The tasks that used to teach juniors are the ones you just automated. The common mistake is taking the saving because it appears this year and the cost appears in someone else's tenure — which is the apprenticeship problem, arriving as a budget decision.
Q37Your best team has quietly built its own workflow around an unapproved tool. Output is up about 30%. Legal has just found out. What do you do?
Answer: (2)They have done your discovery work for free and identified a 30% gain. Punishing them is legally tidy and organisationally catastrophic — you will never hear about the next one, and there will be a next one.
Q38An analyst submits a flawless-looking assessment; one figure was invented by the model and never checked. "The AI produced it, I assumed it was right." What is the correct response?
Answer: (3)Making it purely an individual disciplinary matter guarantees the next person hides it instead of reporting it — which converts a visible, fixable problem into an invisible, systemic one.