The debate about artificial intelligence and jobs is usually framed as a binary question.

Will AI replace workers, or will it simply make them more productive?

I think that is increasingly the wrong question for Australia.

We do not yet know how many jobs AI will ultimately eliminate. The evidence today certainly does not support claims that mass white-collar unemployment is already underway. But AI is improving quickly, businesses are adopting it rapidly, and there is enough evidence of meaningful productivity gains and labour substitution to take the possibility of a significant employment slowdown seriously.

Australia therefore faces a policy question that has received far less attention:

If AI substantially reduces the number of people required to perform knowledge work, where will the remaining human work be located?

That distinction matters enormously.

Imagine an Australian business today requires 1,000 people to deliver a particular function.

Perhaps 600 are in Australia and 400 are offshore.

AI eventually allows the company to produce the same output with only 400 people.

Most discussion about AI employment asks what happens to the other 600 jobs.

Australia should also be asking: where will those remaining 400 jobs be?

Will 300 be highly productive Australian workers supported by AI?

Or will 300 be offshore workers using exactly the same AI systems, leaving only a relatively small management and governance layer here?

Both outcomes could deliver impressive productivity improvements to the company.

They would produce very different outcomes for Australia.

We should not pretend we know exactly what AI will do to employment

There is an understandable temptation to jump ahead of the evidence.

Some forecasts imply AI will eliminate enormous numbers of jobs. At the other extreme is the comforting argument that technology has always created new work and will simply do so again.

Neither position is particularly satisfying.

Australia is not currently experiencing an AI-induced employment collapse. Unemployment remains relatively low and employment continues to grow over the longer term.

Jobs and Skills Australia also currently assesses generative AI as more likely to augment most jobs than eliminate them entirely.

But there are early signs worth watching.

Research by the Department of Employment and Workplace Relations has found employment growth has been somewhat weaker in occupations with greater exposure to AI automation. The effect is small and DEWR appropriately cautions that this does not prove AI caused it.

International evidence is similarly mixed.

Studies of customer-service and professional workers have demonstrated substantial productivity improvements when workers use generative AI. Other research into digitally traded work has found falling demand and earnings in areas particularly exposed to generative AI, including some writing and translation work.

Both can be true.

AI can make individual workers dramatically more productive while eventually reducing the number of workers a company needs.

A company facing rapidly growing demand may use a 20 per cent productivity gain to produce 20 per cent more.

A mature bank, insurer, government department or professional-services business with relatively stable demand may eventually decide it simply requires fewer people.

And that adjustment may not begin with mass redundancies.

The first signals could instead be quieter:

• contractors are not renewed;
• vacancies disappear;
• graduate programmes shrink;
• people who leave are not replaced;
• management layers consolidate;
• and only later does total employment clearly decline.

That is why I think a substantial white-collar employment slowdown over the next three years should be treated as a credible stress scenario, not a prediction.

Policy should reflect that uncertainty.

AI changes the economics of offshoring

Australia has lived with significant offshoring of knowledge work for decades.

The economic argument is familiar. Companies lower their costs, Australian consumers and businesses benefit from cheaper services, and workers displaced from one activity eventually move into more productive employment elsewhere.

AI complicates this equation.

One possibility is extremely attractive for Australia.

AI could substantially reduce the importance of labour-cost differentials.

If one AI-enabled Australian software engineer can perform work previously requiring two or three people, paying that engineer an Australian salary may become competitive with employing a larger offshore team.

Companies may also place greater value on proximity to customers, domain knowledge, regulation, intellectual property, security and faster collaboration.

AI could therefore become an extraordinary onshoring technology.

But there is no guarantee that this happens.

An offshore engineer, accountant or customer-service worker receives access to the same AI capabilities.

AI can also reduce some of the traditional disadvantages of distributed work. Translation gets easier. Documentation gets better. Code can be explained automatically. Handover notes can be generated. Quality checks can be automated. Communication barriers decline.

Work that once required substantial local knowledge may actually become more globally contestable.

And some offshore work will simply disappear through automation without returning to Australia at all.

This is why Australia should not leave the location of post-AI work entirely to chance.

The goal should be capability, not preserving every job

None of this means Australia should attempt to freeze the labour market in its current form.

That would be a serious mistake.

If AI allows 35 Australians to perform work previously requiring 100 people, government policy should not insist that a company continue employing 100 people simply to preserve historical headcount.

Productivity growth is ultimately essential to Australia's prosperity.

But consider the difference between two possible outcomes:

100 offshore workers + AI

versus

35 Australian workers + AI.

The second option may employ fewer humans overall, but from Australia's perspective those 35 people still matter.

They accumulate industry knowledge.

They train future workers.

They pay Australian taxes.

They develop intellectual property.

They understand the underlying systems.

Some eventually leave and establish new companies.

They retain our ability to operate critical business functions.

And they form part of the skills base from which the next generation of Australian companies can emerge.

Government therefore shouldn't try to preserve every pre-AI task.

It should try to maximise Australia's share of the high-value human capability that remains after AI changes the amount of labour required.

That is a much more defensible objective than simply protecting jobs from technology.

Use AI to bring work back to Australia

This leads to what I think should be one of Australia's most immediate policy experiments: AI-enabled onshoring.

Government should establish a time-limited programme encouraging businesses to determine whether functions currently delivered offshore could instead be delivered by smaller, highly productive Australian teams supported by AI.

Potential areas include software development and testing, cybersecurity operations, data engineering, finance operations, compliance, customer service and selected professional-services activities.

The important measure shouldn't be whether the same number of jobs comes back.

It should be whether additional productive capability exists in Australia as a result.

If a 100-person offshore function can genuinely become a 35-person Australian AI-enabled operation, those 35 jobs should count as a successful onshoring outcome.

Support should be temporary, measured against verified additional Australian capability and preferably paid after results have been demonstrated.

Companies should not receive subsidies for jobs they would have created anyway, and support should be clawed back if a supposedly onshored function is quickly moved offshore again.

Subsidise demonstrated incremental Australian capability before taxing globally sourced capability.

Government procurement is an obvious place to start

The Commonwealth already has significant purchasing power.

Rather than introducing simplistic rules requiring government to buy from Australian-owned companies, procurement should measure the economic capability actually retained in Australia.

For major technology, consulting and business-process contracts, bidders could disclose:

• Australian and offshore delivery employment;
• where architecture, security and product decisions are made;
• graduate and apprenticeship commitments;
• knowledge-transfer arrangements;
• Australian subcontracting;
• intellectual property created or retained locally;
• and the capabilities that will remain in Australia when the contract ends.

This avoids confusing company ownership with economic contribution.

A foreign-owned company employing hundreds of Australian engineers, developing local skills and transferring knowledge may generate substantially more Australian capability than an Australian-incorporated intermediary whose workforce is predominantly offshore.

This is not an argument against global delivery. Australian organisations will continue to benefit from access to international skills, global delivery centres and specialist capabilities that cannot always be economically maintained in every market.

The policy objective should instead be to ensure global delivery does not progressively hollow out the capabilities Australia needs to retain. Architecture, operational ownership, security, domain expertise, graduate development and the ability to govern and evolve critical systems all have economic value beyond the hourly cost of the people performing individual tasks.

The opportunity created by AI may therefore be a different balance: smaller Australian teams operating at much higher productivity, complemented by global capability where it genuinely adds scale or specialist expertise.

Score what remains economically in Australia, not simply the nationality of the supplier.

That approach is also much less likely to descend into crude protectionism.

Skilled migration also needs to become more responsive

Migration is another area where I think the debate risks becoming unnecessarily ideological.

I would not cut skilled migration today simply because AI may reduce employment several years from now.

Australia still has genuine skills shortages, and highly capable migrants contribute disproportionately to innovation, entrepreneurship and economic growth.

Australia should absolutely remain capable of attracting exceptional AI researchers, engineers, entrepreneurs and other scarce talent.

But migration policy should not permanently assume that Australia's problem will always be insufficient professional labour.

Suppose that by 2028 software vacancies have fallen substantially.

Graduate developers are struggling to find work.

Australian developers are being laid off.

Employment is falling and wage growth is weak.

If Australia simultaneously continues importing large numbers of ordinary software workers because software development appeared on a shortage list created under completely different labour-market conditions, the policy would no longer make sense.

That does not imply shutting the door.

The person capable of founding Australia's next globally significant AI company should not be excluded because the general software-development market is weak.

The solution is therefore not simply less migration.

It is more responsive migration.

Government should build mechanisms now that link skilled migration by occupation to actual vacancy levels, domestic unemployment, graduate outcomes, wage growth and demonstrated employer demand.

If shortages continue, migration continues.

If an occupation moves into persistent surplus, migration into that occupation progressively tightens.

The important change is composition before aggregate volume.

We also need to protect the skills ladder

There is another AI employment risk that may prove more important than headline redundancies.

Graduate jobs.

Professional careers generally have a learning structure built into them.

Junior accountants prepare work reviewed by senior accountants.

Junior lawyers conduct research and draft documents reviewed by senior lawyers.

Graduate consultants perform analysis before eventually leading engagements.

Junior developers write and test code while learning architecture and business systems.

Unfortunately, many of those junior tasks are precisely the activities AI is becoming good at.

A firm might therefore retain most of today's senior employees while dramatically cutting the number of graduates it hires.

In the short term, that looks like productivity.

Five or ten years later it can become a capability problem.

Where do the experienced accountants, lawyers, engineers and technologists of 2035 come from if organisations stop developing them in 2027?

Australia needs to start measuring graduate and entry-level employment in highly AI-exposed professions much more carefully.

And where those pathways begin breaking down, government and industry should consider incentives for AI-era apprenticeships and graduate programmes.

The answer isn't to make graduates perform obsolete tasks simply because previous generations did them.

It is to redesign professional development so young workers learn domain expertise while also learning to supervise, interrogate, validate and improve AI systems.

Don't introduce a broad offshoring tax

Some have suggest there is a strong argument for directly discouraging offshoring through taxation.

The more I look at it, the less convinced I am that a broad approach would work.

What exactly counts as offshoring?

An Australian engineering firm purchasing specialist modelling from Singapore?

A small software company employing customer support in the Philippines?

A bank operating a technology centre in India?

An Australian AI company buying cloud inference from the United States?

All involve purchasing foreign services.

They are not economically equivalent.

A general offshore-services tax could end up penalising Australian companies for purchasing inputs they need to compete internationally, while businesses would inevitably restructure contracts into SaaS, managed services and other categories to avoid arbitrary boundaries.

That doesn't mean offshoring should remain untouchable under every future scenario.

If Australia entered a severe labour-market downturn where companies were simultaneously making large numbers of Australians redundant and moving the same functions offshore, narrowly targeted adjustment measures could become reasonable.

But that lever belongs much further down the policy hierarchy.

Measure → procure strategically → incentivise onshoring → adjust occupation-specific migration → condition public support → introduce targeted offshoring disincentives if necessary.

Broad protection should be the last resort, not the first response.

Build the machinery before we need it

The biggest mistake Australia could make is waiting until unemployment has already risen sharply before deciding how to respond.

By then policy will inevitably be rushed.

Instead, Australia should develop a labour-market contingency framework for AI now.

It should distinguish between:

• automation, where human work disappears;
• augmentation, where workers remain but become more productive;
• outsourcing, where work leaves a particular employer but remains in Australia;
• offshoring, where the human work itself moves overseas.

Those outcomes are frequently bundled together even though they require very different policy responses.

Government should monitor professional vacancies, graduate recruitment, replacement hiring, wages, AI adoption, migration by occupation and the domestic/offshore composition of major employers.

It should then establish transparent thresholds.

In a green scenario, where productivity rises and labour demand remains strong, Australia should maximise AI adoption, maintain skilled migration and encourage modest onshoring.

In an amber scenario, where vacancies and graduate hiring fall substantially despite reasonable headline employment, Australia should increase onshoring incentives and tighten migration selectively in affected occupations.

In a red scenario, where multiple knowledge industries are shedding workers and unemployment and underemployment rise materially, much stronger capability-retention, migration and offshoring measures become justifiable.

We don't need to know today which scenario will occur.

We do need to know what we intend to do in each one.

This is not an argument against AI

Australia should want companies to automate.

We should want workers to become dramatically more productive.

We should want businesses to use the world's best AI systems.

We should continue trading internationally.

And we should continue attracting exceptional people from around the world.

The objective is not to protect Australians from artificial intelligence.

It is to ensure Australians capture a reasonable share of the prosperity artificial intelligence creates.

That requires recognising something that conventional discussion about AI and employment often misses.

If AI eventually allows an Australian company to do with 400 people what once required 1,000, there is a major difference between 300 of those remaining people being Australians and only 50 being Australians.

Both companies may have identical productivity.

The national outcomes are not identical.

For the past several decades, Australia's economic policy has understandably been built around labour shortages, globalisation and the assumption that workers displaced from one form of employment can ultimately move to another.

AI may not invalidate those assumptions.

But it could.

And a three-year period in which AI capabilities and adoption are changing extremely quickly is not the time to discover that our labour, migration and industry policies cannot adapt equally quickly.

It is too early for Australia to become broadly protectionist.

It is not too early for Australia to stop being indifferent to where post-AI work is performed.

The central question for Australian policymakers therefore shouldn't simply be:

Will AI take our jobs?

It should be:

After AI determines how many human workers are still required, what are we doing to make sure a meaningful share of those workers, capabilities, businesses and opportunities are here?