Gartner's AI skills atrophy warning: what NZ employers should protect

By DataForge team19 November 20265 min read

Gartner says AI use is wearing down Australian workers' skills, and NZ firms should find and protect the skills they can't afford to lose.

What Gartner found

Gartner surveyed 589 employees in Australia. The findings, reported on 18 November, show 27% fear AI will make their skills obsolete, and 46.2% of Gen Z workers feel the same. Fewer than one in five (18.6%) are frustrated by how AI has changed their role. A larger group (29.8%) worry AI will make errors that harm their work. Among Gen Z respondents the frustration and error figures rise to 30.8% and 42.3%.

The support figures matter more for employers. Gartner says 46% of Australian employees are already expected to use AI in their jobs. Within that group, 90% had AI tools supplied by their employer, yet [just 60.9% felt they'd been given clear guidance(https://channellife.com.au/story/gartner-warns-ai-is-eroding-australian-workers-skills) on using them. Formal learning opportunities (54.6%) and manager support (56.9%) scored lower still. Most staff have the tools, and a large minority are working out the rules for themselves.

Jonathan Tabah, a director analyst at Gartner, said the main danger lies in skills workers already have and will slowly lose. He compared employees to pilots who must still be able to fly when the autopilot fails. Gartner's own warning puts it plainly: "Unmanaged AI-driven skills atrophy can reduce performance while increasing business risk." The firm's advice is for HR and business leaders to agree on which fading skills would cause the most operational, regulatory or performance damage if they disappeared. Staff should then keep practising those skills through manual runs, hands-on exercises and simulations. Gartner accepts that some skill decline is harmless and needs no action.

The published coverage doesn't give fieldwork dates or say how the 589 respondents were chosen, so read the percentages as indicative. They fit an earlier Gartner study. In June, Gartner's Global Talent Monitor survey of 574 Australian employees, run in the first quarter of 2026, found 38% were expected to use AI in their role. Only half reported clear guidance, training or support. The two samples differ, but both show the same gap: AI tools arrive faster than the support for using them.

Why skill loss becomes an operational risk

AI tools now handle a lot of routine work in mid-sized firms. They code supplier invoices, draft customer replies, summarise contracts, triage service desk tickets and suggest journal entries at month end. When the tool is right and available, nobody notices the human skill underneath fading. The loss only shows up when the tool goes down, gets a case wrong, or meets something it wasn't trained on. By then the person who used to do the task by hand may have done it twice in a year.

Accounts payable is a good example. A team that has approved AI-coded invoices for 18 months can still click "approve" quickly. The open question is whether anyone on that team can still spot a GST treatment that's wrong, or a supplier whose bank details changed without explanation. Service desks have a similar problem. Analysts who mostly accept AI triage suggestions may lose the diagnostic habits they need on a day when the suggestions are confidently wrong.

The Gen Z figures point to a longer-term problem. Junior staff who start their careers reviewing AI output have never done the underlying task by hand. Reviewing well depends on judgement, and people build judgement by doing the work. If a firm's graduate intake never prepares a reconciliation from scratch, the firm will struggle to find capable reviewers in five years.

What it means for NZ businesses

No equivalent Gartner figure for New Zealand has been published. Local data suggests the guidance gap here is at least as wide. Spark surveyed more than 400 NZ business leaders and found 69% use AI to some extent, mostly general-purpose chat tools, but only 33% have policies governing its use. The barriers respondents listed included employee skills and doubts about the reliability of AI outputs. Doubting AI output is reasonable, but staff can only act on that doubt if they still have the skills to check the work.

The same Spark survey found 44% of NZ businesses have a resilience plan, and only 16% of those regularly review or test it. Few continuity plans we've seen in New Zealand cover the scenario where an AI tool is unavailable or producing bad output. Smaller firms carry extra risk because one person often holds the manual knowledge for payroll, GST or a key production step. When AI takes over the routine part of that person's role, the firm is relying on a single person's memory of a process nobody has practised recently.

Most firms don't need a formal skills program to fix this. They need a short, deliberate decision about which tasks still need a human who can do them unaided, and a habit of keeping those people in practice. For most firms that conversation belongs inside their wider AI adoption and governance planning, alongside the usage policy most of them still lack.

Practical next steps

  • List every task where staff now rely on AI output. Mark the ones where an undetected error would cause regulatory, financial, safety or customer harm. In most NZ firms that list will include GST and payroll, privacy requests, credit decisions, and health and safety records.
  • Name a person who can do each marked task by hand. Record when they last did it. A blank or a date more than six months old is a gap to close.
  • Schedule manual runs. Run one month-end reconciliation, a sample of invoice coding, or a batch of service desk tickets without AI suggestions each quarter. Compare the results with what the AI produced.
  • Write guidance staff can use. Say which AI outputs need human checking, who signs them off, and what a reviewer should look for. Gartner's 60.9% figure suggests many workers are guessing at these rules.
  • Design junior roles so people do the work before they review it. Graduates should prepare a task manually a set number of times before they start approving AI versions of it.
  • Add "AI tool unavailable or wrong" to your continuity and incident plans. Then test the scenario. If AI is built into your workflows, review those process and automation designs so a manual fallback path exists and is documented.
  • Track what reviewers catch. Measure the errors found in AI output and how long manual completion takes. Both numbers tell you whether the skill is still there.

Gartner's figures come from Australia, but they describe the same tool rollouts NZ firms are running now. Pick the handful of skills your business can't afford to lose, and keep the people who hold them in regular practice.

Want to know what this means for your organisation? Our team in Auckland can show you your top three opportunities in a free 30 minute assessment.

Get my free assessment How we help
Get new articles by RSS