The Essentials
Australian agriculture offers a rare and quantified demonstration of what automation actually produces in concrete terms: a 13% increase in activity in the second quarter of 2025 while the sector lost approximately 60,000 jobs over five years, falling from 360,000 to 300,000 employees, a decline of 17% (MYOB SME Performance Indicator Q2 2025, published in September 2025). The decoupling between productivity and rural employment is no longer a management consulting projection—it is measured, dated, sectoral. The political question that follows is open: who finances the retraining of displaced workers when productivity gains do not generate a proportional increase in tax revenue?
Few sectors allow us to date a turning point with precision. Australian agriculture is one of them. While the country debated the future effects of automation, the sector was already experiencing them.
The phenomenon is precise and documented: an identified sector, on an identified continent, with quarterly figures that trace the mechanism in real time. The sector produces more, employs fewer people, and the countryside empties while balance sheets improve. The Australian case has become, almost by accident, the observation laboratory that the global debate on AI and work was waiting for.
The Sector That Produces More While Employing Fewer
The figure strikes by its precision. In the second quarter of 2025, Australian agricultural activity increased by 13% according to the MYOB SME Performance Indicator Q2 2025, a quarterly index that aggregates activity data from several hundred thousand small and medium-sized Australian businesses. Over the same five-year period, the number of people employed in the sector falls from 360,000 to approximately 300,000. Two simultaneous trends, in opposite directions.
This type of decoupling between activity and employment is the central figure in the automation debate over the past decade. It was projected, modeled, feared. In Australian agriculture, it can be read in a quarterly table.
The technologies involved are well-known: tractors and combine harvesters with GPS guidance, crop surveillance drones, sensor-controlled irrigation systems, robotic mowing and sorting systems. Australia has large-scale operations, with average farm size far exceeding European or Asian standards, which makes automation economically viable before other geographies. Robotic shearing systems, still in the prototype stage in Australia, are primarily developed to compensate for an existing shortage of qualified shearers rather than to render superfluous seasonal workers currently employed. The return on investment for available automated equipment remains nonetheless an implacable logic for operators.
What distinguishes this case from usual projections is precisely its grounding in reality. Most estimates on AI and employment rely on matrices of potentially automatable skills, extrapolations applied to occupational classifications. Australia, meanwhile, offers a five-year progress report in a homogeneous sector. The occupation-by-occupation study that Australia conducts across its entire labor market confirms this national singularity: the country is one of the few to instrument the monitoring of automation’s impact with sectoral granularity.
Rural Areas Pay the Bill That Balance Sheets Don’t Record
Sixty thousand rural jobs disappeared in five years in a sector experiencing growth—this is a sentence that needs to be broken down. These jobs were not destroyed by a crisis. They were made redundant by a productivity gain. The distinction is important: it means that agricultural businesses did not suffer. They prospered.
The displaced rural workers did not follow the same path. Australian agriculture employs disproportionately unskilled workers, often seasonal, sometimes migrant, geographically concentrated in regions far from major employment centers. A combine harvester driver replaced by an autonomous system in Queensland is not an hour’s drive from Brisbane to retrain in the service sector. Geographic mobility has a cost that aggregate employment statistics do not record.
This asymmetry is at the heart of what economist Dani Rodrik calls the problem of “good jobs,” the jobs that structure a life, a territory, a community, beyond their mere marginal productivity. In his work on industrial policy and employment, Rodrik argues that markets left to themselves tend to concentrate the gains of technological progress in a few hands and a few geographies, while the costs of adjustment are widely borne by displaced workers and abandoned regions. Australian agriculture offers a documented illustration of this.
Small rural Australian towns absorb these costs in visible ways: shuttered shops, public services under strain, young people fleeing to coastal metropolises. It is a process Australia has known for decades—the mechanization of the 1970s-1990s had already set the movement in motion—but digital automation accelerates it on a much shorter cycle.
The Fiscal Question That No One Is Yet Asking
The productivity balance is positive. The redistribution balance is an open question. And it may be the most important question the Australian case raises.
When a mechanized farm produces 13% more value with 17% less labor, it generates extra margin. This margin is taxed according to the prevailing tax regime, corporate income tax in Australia at 30% for large corporations, but often lower for family farms and sole proprietorships. The replaced workers, meanwhile, no longer contribute to the tax base. They potentially become its beneficiaries: allowances, retraining, mobility.
The arithmetic result of this mechanism is tax revenue that does not increase proportionally to the sector’s productivity gains, while the public financing needs linked to career transitions increase. It is the basic mechanics of a sector that is structurally losing jobs while becoming richer.
Australia has not solved this equation. It is not alone. The question of fiscal mechanisms capable of capturing the productivity gains of a sector that is shedding jobs to finance the retraining of displaced workers, before the loss becomes irreversible, remains largely without institutional answer in most developed economies. The OECD has documented for several years the inadequacy of continuing education schemes in Anglophone countries, where public spending for professional retraining remains significantly below what transition needs would require.
Carl Benedikt Frey, whose work on innovation and employment laid the empirical foundations for contemporary debate, has shown that industrial regions hit by automation in the 1980s-1990s never recovered their pre-mechanization employment levels. The adjustment did not occur, or occurred over an entire generation, at a social cost that no one had budgeted for. Australian agriculture may be at the beginning of the same cycle, in a rural sector where the alternative employment option is even more limited than in former industrial cities.
The Contributions and Limits of the Liberal Reading
Facing this diagnosis, the liberal reading has its arguments, and they are not without value. Agricultural productivity gains are partly transmitted to consumers in the form of lower food prices, freeing purchasing power for other consumption and, ultimately, other jobs. Schumpeterian creative destruction, which Philippe Aghion formalized in his work on growth, predicts that jobs lost in one sector are reborn elsewhere, in new activities that the surplus wealth makes possible to finance.
The argument holds in the long term and at the macro scale. It stumbles on geography and time. The agricultural worker in Queensland is not the direct beneficiary of the boom in digital services in Sydney. The horizon of creative destruction is measured in decades; the horizon of a professional life is shorter. And rural Australian regions do not have the diversified economic fabric that would allow local retraining—they are, structurally, mono-sectoral economies.
The tension between the two readings becomes productive rather than sterile. Rodrik and Frey do not dispute the aggregate gains of automation. They dispute the assumption of automatic distribution of these gains, and they have the data to back it up. The sharing of capital gains poses with comparable acuity in other configurations: when capital productivity progresses faster than redistribution mechanisms, the gap widens without correcting itself spontaneously.
The lucid version of liberal optimism acknowledges this point. It does not conclude in favor of inaction—blocking automation would impoverish the sector and raise food prices—but it admits that the transition requires active intervention: financing for training, supported mobility, tax adapted to capturing productivity gains. The market alone does not finance its own transition.
Why This Model Will Be Difficult to Sustain Elsewhere
Australian agriculture is a textbook case for another reason: it is a relatively sparsely populated political sector. Sixty thousand rural jobs lost over five years in a country of 27 million inhabitants is politically manageable, even if socially painful for the affected communities.
The same logic applied to much more populous sectors—transport, logistics, retail, personal services—will pose a sustainability question of a completely different order. A sector representing 0.2% of the workforce can restructure without triggering a major electoral reaction. A sector representing 5 or 10% of it changes the equation.
This is why the Australian agricultural case deserves disproportionate attention relative to its size. It is an observable precursor of a mechanism that will reach much broader perimeters. MYOB data over five years allow us to date the turning point—2020-2025—and observe that productivity progressed continuously while employment declined steadily. The turning point was not a punctual shock but a gradual slide, without visible alarm, precisely because the sector was doing well.
This silence is perhaps the most important characteristic to note. Employment crises linked to automation do not manifest as collapses; they look like productivity improvements. They are recorded positively in aggregate indicators, while human costs accumulate in regions and age groups that these indicators do not capture. The question of how to finance business in territories abandoned by these transitions is already being posed, and it has no stabilized answer.
The Decisions Australia Has Not Yet Made
Australia has a rare advantage: it can observe the phenomenon before it spreads. The Albanese government has displayed ambitions on industrial policy and training, but concrete retraining programs targeting rural workers displaced by automation remain embryonic. The National Skills Agreement signed in 2023 between the federal government and the states increases vocational training financing, but without specifically targeting sectors undergoing rapid technological restructuring.
Two paths are actively debated in the country. The first is an extension of the tax contribution from heavily automated companies, a form of levy on productivity gains linked to capital-labor substitution, distinct from corporate income tax. The second is strengthening portable training rights mechanisms, so that displaced workers have an envelope they can use without geographic conditions.
Neither is in place. Both face standard opposition: agricultural businesses argue that automation’s profitability already finances the local economy through equipment purchases and property taxes; state governments resist any reform touching their own tax base. The debate is open.
What is certain is that the window for acting upstream, before the loss becomes irreversible and rural regions have lost their critical demographic mass, is limited. The experience of European industrial regions deindustrialized in the 1980s suggests that once rural social fabric is weakened, retraining policies lose much of their effectiveness: there are not enough people left for collective schemes to function.
The true utility of the Australian case is there: it offers an observation period that most economies will not have. The question is whether this period will be used to build transition mechanisms, or simply to record their absence.
Sources
- MYOB SME Performance Indicator Q2 2025, published in September 2025 (MYOB)
- OECD, Employment Outlook, data on continuing training spending in member countries (OECD Employment Outlook, 2023-2024 editions)
- National Skills Agreement Australia 2023, Department of Employment and Workplace Relations (Australian Government)
- Dani Rodrik, Good Jobs, Bad Jobs, No Jobs and work on industrial policy for employment (Harvard Kennedy School)
- Carl Benedikt Frey, The Technology Trap: Capital, Labor, and Power in the Age of Automation (Princeton University Press, 2019)
- Philippe Aghion, work on Schumpeterian growth and creative destruction (Collège de France / Harvard)
- ABARES - Snapshot of Australian Agriculture 2026
- Australian Taxation Office - Company Tax Rates
- Dani Rodrik - ‘Fixing Capitalism’s Good Jobs Problem’
- AWI / Australian Farmers - Robotic Shearing
- OECD Employment Outlook 2023 & 2024