Automation isn't one thing. There's a world of difference between a system applying a decision a human already made, and a system making the decision itself.

Robots and AI are going to take on a meaningfully larger share of workforce activity in the near future. I think most Australian and New Zealand employers are still governing their workforce on a planning cycle built for a much slower pace of change.

Robots in the workplace aren't a new idea to me. I've watched manufacturing run on them for decades. What's changed is the speed everything else around them is moving at. But the question I actually want leaders asking isn't "should we adopt this." It's a sharper question than that: for any given task, is the machine applying a decision a human already made, or is it making a fresh decision every time you ask it?

The distinction that actually matters

To put the point Stacey makes about AI's shifting interpretation in more concrete terms, it's worth splitting "automation" into two genuinely different things, because they carry very different risk.

Human-decided, machine-applied

Most of the automation worth trusting works like this: a human works out the logic once, codes it or configures it, and the system applies that same logic consistently, every time, at a scale no person could match. Someone has already done the interpretive work, decided what a rule means and how it applies, and the system's job is to apply that decision the same way on the thousandth case as it did on the first. That's not a risk to be managed. That's exactly what good automation should be doing.

AI-improvised, decided fresh each time

The other kind of automation is different in kind, not just degree. This is asking an AI model to generate a judgement on the spot, with no fixed human-authored rule behind it. As Stacey points out, ask an AI the same interpretive question on different occasions and the answer isn't guaranteed to stay the same. For most tasks, that variability doesn't matter much. For anything touching pay, compliance, or a person's role, it matters enormously, because what you need is a decision that's settled and defensible, not one that shifts depending on when you asked.

Why this split changes how you should think about risk

Once you see these as two different things, "AI risk" stops being one conversation. A rules-based system applying a human's already-made decision is low risk almost by design, provided the rules stay current. A model improvising a fresh judgement each time is a genuinely different risk category, and it's the one that needs a human owner standing over it before it acts, not after.

Why governance can't keep pace anymore

The old planning model assumed things would hold still

Most of us still run workforce governance on a "set time aside, think, plan, execute" model. That model assumes the underlying landscape holds still long enough for the plan to still be relevant once it's executed. That assumption doesn't hold anymore, not with AI and automation moving at the pace they are. A plan built around this quarter's tools can be built around last quarter's tools by the time it's approved.

What this means for Australian award-covered workforces

For Australian employers running shift-based, award-covered workforces, this isn't abstract. Rostering patterns and workforce structures under the Fair Work Act are already hard enough to govern well when the underlying technology is stable. Layer in robotics and AI that can reshape a role, a shift pattern, or a team's structure within months rather than years, and a governance cycle that only gets revisited annually is effectively governing a workforce that no longer exists by the time the review happens.

What robots and AI are good at, and what has to stay human

Where automation earns its place

This isn't wholesale replacement of people, and it's not meant to be. Where I see automation and AI genuinely earning their place is in repetitive, pattern-based tasks: drafting routine communications, spotting anomalies, running the same check at scale without fatigue or inconsistency, and applying rules a human has already worked out and kept current.

Where the human has to stay accountable

What has to stay in human hands is the interpretive call itself: deciding what a clause means, how it applies to a genuinely novel situation, and taking responsibility for that judgement. Automation can apply a human's interpretation consistently. It shouldn't be the one making the interpretation, and it shouldn't be trusted to make a fresh one every time it's asked.

Why the split actually helps, rather than threatens, leaders

I see the upside here as deliberate, not incidental. Handing the repetitive, rules-applying share of the work to automation is what creates the space for the work only people can do: strategic thinking, planning, deciding what the business and its workforce should look like next. Automation taking on that volume isn't a threat to that kind of thinking. It's what makes more of it possible, and it's exactly the kind of deep thinking I think we owe it to ourselves to protect time for as things keep accelerating.

Four things I'd actually do about it

1. Map every AI and automation touchpoint in your workforce systems, and classify each one. For each, ask: is this applying a decision a human already made, or is it generating a decision fresh each time? That single question sorts your real risk from your low risk far better than a generic "AI policy" ever will.
2. For anything in the "human-decided, machine-applied" category, focus your governance on keeping the underlying rules current. The risk here isn't the automation, it's the rules going stale while the automation keeps confidently applying them anyway. Set a review cadence tied to actual changes (a new award variation, a new EA, a restructured role), not an annual calendar date.
3. For anything in the "AI-improvised" category, put a named human owner in front of the decision, not behind it. That means sign-off before the output is acted on, especially anywhere it touches pay, compliance, or a person's role, not a retrospective audit after the fact.
4. Shorten your planning cycle to match the pace of the technology, not the pace of your old governance calendar. An annual or biannual workforce plan assumes a stability that no longer exists once AI and robotics are in the mix. I'd be revisiting the two points above every few months, not every few years.

The practical takeaway

Robots and AI reshaping the workforce isn't a distant scenario as far as I'm concerned, it's already underway. But the organisations getting this right aren't the ones being broadly cautious about "AI." They're the ones that have actually drawn the line between automation that applies a human decision and automation that's making decisions of its own, and built their governance around that line specifically.

Whether the change comes from a person absorbing new duties or a machine taking over part of a role that used to be entirely human, ask yourself this: for every point where automation touches your workforce, do you actually know which side of that line it sits on, and who's accountable for keeping it there?

Stacey Kavanagh is the founder of KJM Consulting, an Australian workforce technology advisory firm, and a contributor to Global Payroll Magazine. She sat down with Dave Kenyon on OAHI's Evolving Workforce podcast to talk about where robots and AI are actually taking the workforce, and why most governance can't keep pace.

OAHI's Workforce Management platform is built around that same principle — rules a human has defined, applied consistently at scale, and kept up to date as your workforce and compliance obligations change. It helps organisations manage time & attendance, payroll compliance, and award interpretation with greater accuracy and less manual overhead. If you want to see how, book a demo.

‍