Most businesses I speak to are moving quickly, sometimes very quickly, on AI, and there is understandable enthusiasm around productivity, efficiency, automation and competitive advantage because there is enormous potential here.
What I think we are paying less attention to is the idea that AI is more than a technology shift, it changes work.
It changes what people do, how quickly they do it, what decisions remain theirs, what they are accountable for, how their performance is monitored, how they learn and, increasingly, who or what they interact with during the working day.
Once you look at AI through that lens, it stops being purely an IT conversation and becomes a psychosocial risk conversation as well.
Safe Work Australia identifies job demands, low job control, poor support, lack of role clarity, poor organisational change management, inadequate recognition, poor organisational justice, isolation and conflict among the psychosocial hazards organisations need to identify and manage.[1] If an AI implementation changes any of those conditions, then it is changing the organisation’s psychosocial risk profile, whether anybody has called it that or not.
In the evidence mapping I have been doing, eight themes appear repeatedly: displacement anxiety, work intensification, accountability gaps, skill atrophy, workslop, surveillance and algorithmic management, disconnection, and rapid unmanaged change.
Start with job security, because people are hearing almost daily that AI will replace roles, reduce headcount and reshape entire professions. The World Economic Forum estimates that labour-market disruption will affect around 22% of current jobs by 2030, while 41% of surveyed employers expect to reduce workforce numbers where AI can automate tasks.[3]
Whether an individual’s own job is genuinely at risk may almost become secondary, because if they believe their value is disappearing, that belief changes the way they experience work. People can become more anxious about performance, more reluctant to speak up, more inclined to work longer or harder to prove their usefulness, and they may begin psychologically competing with a machine that never gets tired, never takes leave and produces a first draft in seconds.
That is a very different working environment.
Then there is workload, where I think we need to challenge one of the more convenient assumptions in the AI conversation, which is that removing tasks automatically reduces demand.
Some of the tasks AI removes are the lighter parts of the day, such as a straightforward email, a routine report or some basic administration, and although those things consume time, they also create variation in cognitive demand. Remove enough of them and the day can become increasingly concentrated around complex judgement, difficult conversations, problem-solving and decision-making.
The work becomes denser, which means fewer tasks can sometimes feel like more work.
The International Labour Organization is now explicitly warning that AI systems can contribute to work intensification, reduced autonomy, surveillance and other changes to psychosocial working conditions.[4] This is exactly why measuring only time saved tells us very little about what has happened to the work itself.
Control and accountability interest me even more.
AI is increasingly contributing to recommendations, analysis, decisions, drafts and workflows, yet when something goes wrong somebody still has to own the outcome, and that somebody is usually human.
We are therefore beginning to create situations in which people carry responsibility for outcomes they only partly controlled, while perhaps not fully understanding how the recommendation was produced in the first place.
High responsibility combined with low control is hardly a new psychosocial problem, AI has simply given it a new haircut.
The same question applies to capability. Most experienced leaders developed judgement by doing work imperfectly, being corrected, trying again, watching people who were better at it and gradually developing an instinct for what good looks like.
If AI starts doing too much of that early thinking, particularly for younger employees and emerging leaders, we need to ask not only what time they are saving but what they are no longer practising.
Research from Microsoft involving 319 knowledge workers found that higher confidence in generative AI was associated with less critical-thinking effort, while AI also shifted critical thinking towards verification and oversight rather than necessarily removing it altogether.[5] That distinction matters, because AI may extend capability when someone already has strong foundations while also creating what I think of as cognitive debt when it substitutes for the very practice through which those foundations would normally develop.
Then there is workslop, which I suspect most workplaces have encountered already.
A beautifully formatted AI-generated document arrives looking finished, but when somebody reads it there is very little thinking underneath, so the recipient has to interpret it, correct it, add the missing context or redo much of the work.
The sender saved time, but the recipient inherited the work.
Research published in Harvard Business Review, drawing on BetterUp Labs and Stanford research, found that 41% of workers surveyed had encountered this kind of low-value AI output, with almost two hours of rework associated with each occurrence, alongside reported damage to trust and collaboration.[6]
That is the point where an apparent productivity gain starts becoming a relationship problem.
There is also a quieter change occurring as people increasingly ask machines questions they once asked each other. Those small interactions did more than transfer information, they built relationships, gave managers opportunities to notice when somebody was struggling, created informal coaching and helped people learn how colleagues think.
When the work continues but the human exchange surrounding it starts disappearing, I call that shadow collaboration.
None of this means organisations should slow AI adoption simply because change creates risk, but it does mean WHS, HR and people leaders need to be much closer to the implementation conversation.
NSW has already moved explicitly in this direction through the Work Health and Safety Amendment (Digital Work Systems) Act 2026, which defines digital work systems to include algorithms, artificial intelligence, automation and online platforms, and extends WHS duties to risks arising from their use.[7]
So, before asking what the technology can do, I think every AI rollout needs to ask another question:
What does this technology change for the people doing the work?
Does it increase pace while reducing variation, does it alter control or accountability, does it remove opportunities to learn, change how people are monitored, reduce human interaction, create uncertainty about roles, or introduce significant change without meaningful consultation?
Those questions belong in the same room as the ROI conversation, because the success of AI will eventually depend on far more than how clever the technology becomes.
It will depend on whether the people working alongside it can still think clearly, exercise judgement, trust each other, understand what they are responsible for and remain well enough to do the work.
That is the part I think we need to get ahead of now.
____________________________________________________________________________________________
Guest Contributor:
Mark Butler, Performance Specialist
Mark Butler is a clinical psychotherapist and mental health strategist who works with executives and teams under sustained pressure. He is a former clinical director, an author on burnout, and a member of the World Federation for Mental Health’s Workplace Committee.
