AgMap Research
Hands and heads
We assume AI gets into a farm business two ways: machines that do physical work, and software that does knowledge work. This page uses ABARES farm survey data to measure how much of each industry’s costs the two could actually replace, then checks that against how far along the technology really is. Paid labour runs from 9% of cash costs on grain and beef farms to 31% on vegetable farms; paid advice is 1% to 2% everywhere.
The two are different in kind: a robot can remove a wage, while AI mostly changes what a dollar of advice buys, or its quality. Currently mechanised industries (broadacre, cotton, sugar) are better targets for automation, but as a consequence, the labour saving is minimal in comparison to a less mechanised industry (many hort crops).
I think it’s also important to spell out that I’ve tried to simplify this analysis. As a consequence some significant AI-driven gains have been ignored (for now). The primary question of this report is “what will happen if X is automated?” We’re deliberately skipping the more nuanced question: “what margin improvement could be had with perfect decision making?” The answer to that question will have larger impacts on already mechanised industries (broadacre) with high materials costs. In a way we can estimate the answers to this latter question by comparing the variation between yields (and profits) within a single system. Someone often makes the right call at the right time and gets a bumper crop, dodges the frost or rain. Farming is often a lottery, and the dependence and exposure to climatic risk is part of the game, and also part of the reason I respect its practitioners.
Dr Thomas Lines · AgMap · August 2026
Every figure here comes from a published source, most of them ABARES farm surveys, and every chart has a table view. Labour cost data crosses two definitions and fifteen years of surveys, so each number carries its year and its basis on its face. If you find a mistake, tell me: info@agmap.org
Hired labour, whole sector
$8.7bn
14.7% of sector cash costs
9.1% of the $95bn gross value of farm production, 2024-25
Labour and advice, sheep farms
19.7%
of cash costs, highest in the current surveys
vegetable farms were 31% when last surveyed, in 2015-16
Labour on horticulture farms
35%
of operating expenditure
ABARES, 2020-21, including contract labour
Accounting and advice, grain farms
$20.6K
per farm per year, 1.2% of cash costs
more than any other surveyed industry
What we mean by hands and heads
Hands: machines doing physical work
- Autonomous sprayers and tractors. SwarmFarm machines already work paddocks in Queensland and WA.
- Drones mustering cattle instead of helicopters and bikes.
- Robotic milking sheds, running on about 1.5% of dairies.
- Fruit and vegetable picking robots, still prototypes everywhere.
- Robotic shearing, still a university research rig.
Where these work, wages and contract payments genuinely leave the cost sheet.
Heads: software doing knowledge work
- Answering agronomy questions between adviser visits.
- Drafting budgets, BAS, grant applications and compliance paperwork.
- Working through award rates and workforce rules, the heaviest paperwork in horticulture.
- Summarising trial results and market reports for one district.
- Helping an adviser prepare, so a farm visit goes further.
The advice bill mostly stays. The likelier change is what a dollar of it buys: routine answers get cheap, and one adviser can cover more farms.
Industry by industry - opportunity for AI and robotic disruption
- Wages and contract work
- Accounting and advisory services
Swipe the chart sideways, or press Table for the numbers.
Automation is advanced in mechanised industries
- Wages + contract work, 2023-24
- Older survey (year in tooltip)
Swipe the chart sideways, or press Table for the numbers.
One caution before adding up the savings. Mechanisation is not the same thing as automation, but it is the closest precedent for taking labour out of a farm industry, and sugar ran that experiment first: cane harvest went to machines by the 1970s, growing cane is still a tight-margin business, and the money that once went to cutting gangs goes to machinery, contractors and fertiliser instead.
The costs of running the average grain farm
- Wages, contractors and shearing
- Accounting and advisory
- All other costs
Swipe the chart sideways, or press Table for the numbers.
Grain growers buy the most advice
Swipe the chart sideways, or press Table for the numbers.
The advice line is small because most knowledge work on a farm is done by the farmer, unpaid and often after dark. A grain grower commits $1.7m a year with a workforce of about three, and the choices that money rides on are ordinary ones: which paddock gets nitrogen, when the fungicide goes on, when to sell. Those choices matter more than any cost line. ABARES’s farm performance reporting shows a wide, persistent gap between the top quarter of broadacre farms and the rest on similar country, and software that helped an average farm make top-quarter calls would be worth far more than the advice bill, though the benefit would arrive as yield and price rather than a cost saving. It would not make advisers redundant either. The scarce thing is trusted judgement about a specific paddock, and tools that handle the routine parts make each hour of it go further. The public system built to spread that judgement, extension, has lost 60% of its funding in twenty years; that story is in our first report.
Where each industry stands
The same sources, read sector by sector. No dollar forecasts: we haven’t found a published one we would rely on.
| Industry | Physical work (robots) | Knowledge work (language models) |
|---|---|---|
| Grain growing (GRDC) | Autonomous spraying is commercial now: SwarmFarm machines have covered more than 4m hectares, and John Deere is aiming for fully autonomous US row cropping by 2030. Wages are 3.8% of costs, so the gains come through timeliness and chemical savings more than the wage bill. | Grain farms spend $612K a year on fertiliser and chemicals, market $2.1m of crop, and already buy more advice than any other industry. |
| Beef, sheep and wool (MLA / AWI) | Drone mustering works and is spreading: SkyKelpie has mustered over 500,000 head. Robotic shearing has been attempted since the 1980s and is still a research rig, while shearing costs sheep farms 10.6% of cash costs. Most other labour is the family’s. | Fewer paid decisions a year than cropping. Advisory spending today is small, $580 to $940 a farm, so new tools will need to prove themselves inside tight budgets. |
| Dairy (Dairy Australia) | Robotic milking is commercial but runs on about 1.5% of farms, and NSW DPIRD found it performing level with conventional sheds. Adoption tends to follow labour shortage and generational change. | Daily feed and herd decisions, on the best-recorded animals in agriculture. Purchased feed is a third of all dairy costs, so small improvements are worth real money. |
| Horticulture (Hort Innovation) | The biggest labour bill in agriculture and the least automation. No machine picks fruit or vegetables at commercial scale anywhere; packhouses and picker-assist platforms are automating first. | The heaviest compliance and workforce paperwork in farming, which is the kind of work language models already do well. |
| Wine grapes (Wine Australia) | Over 90% of the crop is machine harvested with a driver in the seat, so the automation step is removing the driver. Autonomous vineyard tractors are commercial in the United States, and picker-assist platforms are being sold into Australia. Hand work survives at the premium end because wineries want it. | In a difficult market the pressing decisions are planning ones: what to grow, what to sell and where. They are knowledge work. |
| Cotton and sugar (CRDC / SRA) | Harvest is already a machine with an operator, so automation here means the same step grain is taking: removing the operator. Cotton labour was 11% of costs in 2017-18 even counting family labour, so the dollars at stake are smaller than anywhere else. | Both industries already lean on professional agronomy and close benchmarking. |
Read this before quoting anything
- The first chart is published ABARES lines plus a framing choice; there is no model behind it. Counting every dollar of wages and contract work as replaceable is generous to robots. Counting only accounting and advisory as the knowledge-work exposure is deliberately narrow: better decisions also change what fertiliser, feed and grain marketing return, but that arrives as margin, not as a removable cost line, so it is not in the bars. And the ochre segment marks money exposed to change rather than money that disappears: the likelier outcome is advice getting cheaper and reaching further, not advisers going away.
- Survey vintages differ and are always labelled. Broadacre and dairy figures are 2023-24. The vegetable figures are 2015-16, the last time ABARES surveyed the industry, and predate the post-COVID wage rises; the fruit and wine figures in the second chart are hired labour only and older still. Family and operator labour sits outside cash costs entirely, and on beef farms it is nearly five times the hired wage line.
- Contract payments bundle a machine with its operator, so the labour totals overstate pure labour. That is also why grain and cotton look labour-light: the operator’s seat moved to a contractor’s cost sheet decades ago. And the totals only cover paid work: the operator’s own hours sit outside cash costs and are not counted, so on family-run farms the physical work a robot could take on is larger than the teal bars show.
- These are per-farm averages with relative standard errors of 8-15% on the labour lines (up to 30% on small advisory lines). The distribution behind an average is wide.
- Per-farm dollars are real 2024-25 dollars; the industry production values are nominal. The 40%-of-income labour figure for vegetable growers (Hort Innovation benchmarking, 2022-23) uses income as its denominator and cannot be compared with any share-of-costs number here.
- There is no defensible whole-sector dollar estimate of what farm automation will save. The most-quoted number, $20.3bn from the 2017 Precision to Decision report, covers all of digital agriculture at once. And two widely repeated claims trace only to content-farm websites: that 15% of Australian dairies milk robotically (the real figure is about 1.5%) and that robots are shearing commercially (they are shearing a 3D-printed sheep at UTS).
Sources
- ABARES Farm Data Portal (AAGIS/ADIS per-farm averages, 2023-24, real 2024-25 dollars; retrieved 25 Aug 2026)
- ABARES, Financial performance of broadacre and dairy farms, 2023-24 to 2024-25 (June 2026)
- ABARES, Labour use in Australian agriculture, 2021-22
- ABARES, Agricultural Commodities, March 2019 (horticulture hired-labour shares, 2015-16)
- ABARES, Agricultural Commodities, June quarter 2026 (sector costs, returns and GVP)
- ABARES, Demand for farm workers, farm surveys 2018 (cotton labour share incl. family labour)
- University of Adelaide / Wine Australia, Vitivisor economics report (2022; ABARES MDB vineyard cost shares)
- NSW DPIRD, Milking Edge: automatic milking systems in Australia, final report (March 2025)
- Western Growers, Global Harvest Automation Initiative progress (harvest share automated after four years: zero)
- GRDC GroundCover, "Clocking up a week’s work in just one day" (July 2024; SwarmFarm on-farm case)