The judgement gap: the risk your AI business case cannot see

The judgement gap: the risk your AI business case cannot see

AI can accelerate learning or hollow it out. CEOs and boards must govern what people know, see and experience before automating the work that once built future leaders.

By The CEO Institute, in collaboration with Dr Ben Hamer, CEO & Founder ThinkerTank

Most AI business cases can tell a board how many hours will be saved, how quickly a process will run and how much labour cost may be removed. Far fewer can explain what the organisation will stop learning.

That omission matters because entry-level work has always served two purposes. It produced an output, and it developed the person producing it. Junior employees built first drafts, reconciled accounts, researched precedents and watched more experienced colleagues correct their work. The tasks were often repetitive, but the repetition-built pattern recognition. The corrections built judgement.

AI is now capable of performing much of that work. Preserving outdated tasks for the sake of tradition would be a poor response. The strategic question is whether organisations are removing low-value work or liquidating the experiences that create future capability.

A productivity gain can still create a capability loss

Early evidence suggests entry-level disruption is emerging. Stanford Digital Economy Lab researchers found a 16 per cent relative decline in employment among 22 to 25-year-olds in the most AI-exposed US occupations, after controlling for firm-level shocks. The decline was concentrated where AI was more likely to automate work rather than augment people. The researchers describe this as early evidence, not a universal forecast, but the distinction matters.

AI does not inevitably weaken the talent pipeline. It can strengthen it.

A study published in the Quarterly Journal of Economics examined more than 5,000 customer-support staff and found that an AI assistant increased productivity by 15% on average. Less experienced workers gained the most, and newer workers moved down the learning curve more quickly. In that setting AI made some of the practices of stronger performers available to novices in real time.

But it’s important to note that the AI assistant worked because it had been built on the chat records of the company's best staff, and it could only bring new employees up to speed because experienced people were already there to learn from, which is the part that often gets missed. Because this kind of AI runs on human expertise and if a company stops growing experts, there is eventually no one left to copy.

These findings should change the boardroom conversation. The question is whether AI is being designed to automate people out of the learning system or augment them within it.

Dr Ben Hamer describes AI as a workforce transformation rather than a technology rollout. His definition of AI fluency goes beyond tool access and prompting. It is the judgement to know what to hand over, what to retain, when to trust an output and when to interrogate it. People must learn to “think with AI, rather than outsource their thinking to it”.

That judgement must come from somewhere.

The rise of managers who never learnt the work

The most serious risk is not simply that fewer graduates enter the organisation. It is that future managers may become responsible for governing work they never learnt to perform.

A finance leader may be expected to challenge an AI-generated model without having built enough models from first principles. A lawyer may be asked to identify a weak argument without having worked through the underlying research. They may be highly capable users of AI and hold the right titles, yet their authority could advance faster than their judgement.

The organisation may have names on its talent map, but fewer people with the depth to challenge an automated answer when it is polished, plausible and wrong.

Australian workers already recognise the importance of this human layer. Microsoft’s 2026 Australian Work Trend Index found that critical thinking and quality control were the two capabilities workers considered most important as AI takes on more work. Only 28 per cent said their organisation was clearly aligned on AI strategy and policy.

Design the learning back in

Dr Hamer’s challenge is direct. When AI removes the work through which people previously learnt, organisations must “design the learning back in”.

A practical way to do that is to reverse-engineer development through three lenses: education, exposure and experience.

1. Education
2. Exposure
3. Experience

Education is what a future leader must know.

It includes technical knowledge, commercial principles, regulatory obligations, AI fluency and the ability to evaluate an output. AI can improve this layer by making expertise more accessible and accelerating the acquisition of knowledge.

The risk is confusing access to an answer with understanding. An employee who can generate a technically correct response may still lack the foundations needed to recognise when the same system produces a convincing mistake.

Exposure is what a future leader must see.

It includes customers under pressure, senior decision-making, conflicting stakeholder interests, failed projects and difficult trade-offs. Exposure gives context to knowledge. It reveals how the organisation behaves when the answer is not clean.

A summary of a difficult negotiation is not the same as sitting through one. Reading the outcome of a failed investment does not recreate the debate, uncertainty and competing incentives that shaped the original decision.

Experience is what a future leader must own.

It comes from making decisions, defending a position, carrying accountability, learning from failure and living with consequences. This is the layer most at risk when AI moves a junior employee directly from requesting an answer to reviewing one.

Education can be delivered. Exposure can be arranged. Experience must be earned.

For every role materially changed by AI, leaders should identify what people previously learnt from the work, then decide how each element will be replaced or accelerated. Some tasks can disappear. Others may need to remain as controlled practice.

Junior employees may need earlier client exposure, structured rotations, simulations, post-decision reviews and responsibility for smaller but genuine commercial decisions. They should be required to explain an AI-generated recommendation, test its assumptions and identify where it might fail, rather than simply presenting the output.

This is deliberate capability formation, not artificial work.

There is an upside here as well, not just a risk. A company that keeps building judgement can push AI further and faster, because it still has people who can tell when the answer is wrong. What limits how hard you can go is not the technology. It is whether anyone left in the room can push back on what the machine says.

A board-level capability impact assessment

Every material AI investment or workforce redesign should include a capability impact assessment alongside the financial case.

It should answer five questions:

• Which tasks will be automated, and which will be augmented?
• What education, exposure and experience did those tasks previously provide?
• Which future critical roles depend on that development?
• How will the pathway be redesigned before the work disappears?
• What evidence will show that people are ready for greater accountability?

This cannot sit with the CIO alone. Technology leaders do not independently own job design, leadership development, succession or culture. Nor is it an HR initiative detached from strategy. It is a joint question of productivity, workforce planning, risk and capital allocation.

The Australian Institute of Company Directors’ 2026 AI governance guidance places oversight of AI’s human impact on employees and other stakeholders within the board’s role. That should include the long-term effect of automation on capability and succession, not only the immediate effect on jobs.

Boards should also examine incentives. If an executive is rewarded for reducing headcount this year while the capability deficit emerges in five years, the organisation has priced the saving and ignored the liability.

There is a simple test. If the AI proposal reaches the board before anyone has assessed what it does to capability, it has arrived in the wrong order.

Look for the human work

Dr Hamer urges organisations to stop asking, “Can AI replace this person?” and ask instead, “What is the highest-value work our people can do, and how do we free them up to do it?”

That is the right starting point, but it carries a second obligation. Once people are freed from yesterday’s work, the organisation must ensure they are building the judgement required for tomorrow.

AI can compress the time needed to become productive. It cannot be assumed to compress the time needed to become wise.

Before approving the next AI business case, CEOs and boards should ask: what will our people no longer learn when this work disappears?

If the investment case can quantify the labour it will remove but cannot explain how the organisation will develop the judgement it still requires, it is incomplete.

The board is not approving an AI strategy. It is approving an unpriced succession risk.

About Dr Ben Hamer - co-author

Future insight Practical strategy. Confidence in uncertainty. Led by Dr Ben Hamer, ThinkerTank is Australia’s leading foresight agency. They spark curiosity about the future and help organisations make sense of what’s coming next. They track global trends, run original research, and explain it in clear language so leaders and their people can act with clarity and confidence today.

Visit:  ThinkerTank  for more information