AI will cost people jobs.
That is no longer a statement leaders should soften into “AI will free people for higher-value work”. Some of that will happen. So will redundancy.
The management failure is treating those outcomes as if they are mutually exclusive. AI can augment thousands of people and still remove enough work to shrink teams, collapse layers and eliminate roles. The question for leaders is not whether disruption arrives. It is how much of that disruption becomes blunt job loss, and how much becomes role redesign, redeployment and better work.
Poor change management pushes organisations towards the first outcome.
Last month, I argued that “can AI replace this role?” is usually the wrong first management question. That argument still stands. Role evolution remains the real management question: what should disappear, what should change, and which human judgement becomes more valuable as the machine gets better.
But refusing to name displacement is how the change programme fails. If people can see that AI is taking work out of their role while leadership continues to talk only about “empowerment”, trust disappears before the redesign has even begun.
Exposure is not redundancy. It is also not reassurance.
The International Labour Organisation estimates that one in four workers globally is in an occupation with some exposure to generative AI, rising to 34% in high-income countries. Its 2025 update says transformation is more likely than wholesale redundancy because most occupations still contain tasks requiring human input.
Hold both halves of that conclusion. Exposure is not a redundancy list. But “transformation” does not mean headcount stays unchanged. If software can perform a meaningful share of the activities inside a role, organisations will eventually ask whether they need the same number of people performing that work in the same way.
The ILO has been explicit that employment outcomes are shaped by implementation choices. Its Director-General has said the balance between job loss and complementarity depends on how AI is integrated, management decisions and social dialogue. That is the point leaders should focus on.
AI capability creates the pressure. Management choices shape the workforce outcome.
The technology is already in the workplace
This is not a problem leaders can postpone until the technology settles down. MBIE’s 2026 research says around half of New Zealand businesses are already using AI tools.
In Australia, the government’s AI Adoption Tracker reported in June 2025 that 41% of small and medium-sized enterprises were adopting AI, up five percentage points in a single quarter.
The use is already in the building. The issue is whether organisations manage the change deliberately or allow the technology, employees and cost pressure to determine the outcome by default.
The fear is a reasonable question
I spend a lot of time helping teams learn AI tools. Some are enthusiastic. Some barely use them. Some are frustrated because they are using the wrong tool for the task or expecting a level of accuracy the technology cannot yet deliver.
And some ask, quietly, whether getting better at AI is how they help the company discover it needs fewer of them.
That is not irrational resistance.
If leadership tells employees that AI will make the organisation 20% more efficient, a rational employee hears a headcount question. What happens to the 20%? Does it become more customer work, shorter hours, new services, higher output, redesigned roles – or fewer people?
Answer as much of that as you genuinely can. Say what you know, what you are testing, who will make decisions and how role changes will be handled. Do not promise that every job is safe. You cannot know that, and people will not believe you.
This is why I increasingly see AI Change Management as part of AI strategy, not an HR activity that begins after the technology has been deployed.
Prosci’s own benchmarking of more than 10,800 respondents found that initiatives rated as having excellent change management met or exceeded objectives 88% of the time, compared with 13% for those rated poor. That is vendor benchmarking and correlation, not proof of causation. But the warning is relevant: technology does not create value merely because it is available. People still have to adopt it, change how they work and sustain the new behaviour.
Train the role, not the product
The solution is not to hold AI back. It is to train the role, redesign the work and give people a credible path into what comes next.
Finance, marketing, customer service, engineering and an executive team do not need the same AI programme. Each needs to understand which tasks AI performs well, which still require human judgement, what information can be used, how output is checked and when the technology should not be used at all.
Then involve people in identifying what should disappear. The report nobody wants to write. The rekeying between systems. The meeting summary. The first draft. The hour of administration that should take ten minutes.
That is not a veto on restructuring. Difficult workforce decisions will still be made. But there is a material difference between participating in redesign and having redesign done to you.
The workforce is already splitting
The other risk is that organisations create two workforces: people who are learning to work with AI and people watching the change from the sidelines. PwC’s 2026 Global Workforce Hopes and Fears Survey surveyed nearly 50,000 workers across 48 countries and regions. It found access to the learning and development resources workers felt they needed had fallen from 59% to 51%, while the share using generative AI daily had risen from 14% to 22%.
That is the wrong direction. The technology is accelerating while access to development is falling.
PwC also found daily AI users were more confident about their job security and their ability to learn new skills than infrequent users. That is association, not proof that AI use causes confidence. But the split matters: the employee using the tool and the employee fearing it from the sidelines are having very different experiences of the same disruption.
New Zealand’s labour market is showing the same pace of change. PwC New Zealand’s 2026 AI Jobs Barometer found AI-related job advertisements rose from about 3,900 in 2024 to 9,600 in 2025, while high-exposure occupations had added an average of 78 new skills per role since 2019.
The answer to job anxiety is not reassurance. It is capability, mobility and an honest view of where work is heading.
Good change management does five things differently
1. Tell the truth early. Explain what AI can already change, where uncertainty remains and how workforce decisions will be made.
2. Map work before headcount. Break roles into tasks and workflows before jumping to a redundancy target. Identify what should automate, what should augment and what still requires human accountability.
3. Build pathways before removing work. If a task disappears, decide whether the person can move into higher-value work, a redesigned role or a different team before redundancy becomes the default answer.
4. Give people time to build real ability. Training is not attendance. Employees need approved tools, practical exercises, coaching and repeated use in the work they actually do.
5. Govern the transition, not just the technology. Decide which decisions remain human, who is accountable for AI-assisted work, what happens when the system is wrong, and how workforce impacts will be monitored alongside productivity.
Good change management will not save every job
That needs to be said clearly.
Some roles will disappear because AI makes the underlying work cheaper, faster or unnecessary. Some organisations will deliberately use AI to reduce headcount. No change methodology can make that reality disappear.
Not every job loss is technologically predetermined.
Poor change management accelerates displacement because it removes work before it creates pathways. It leaves people unprepared, narrows the conversation to cost, widens the skills gap and makes redundancy the easiest lever to pull.
Good change management does the opposite. It gives leaders time to redesign work, gives employees a fair opportunity to build capability, and creates more options before the organisation decides that a person is surplus to requirement.
The objective is not to protect every task. It is to make sure that when AI changes the economics of work, the organisation changes the people system around it just as deliberately.
Some AI job losses are coming. The scale of avoidable displacement is not fixed.
Leaders should be judged not on whether they stopped AI, but on whether they gave their people a credible chance to move with it.