AI in Agriculture in Aotearoa New Zealand: A Living Whitepaper

AI adoption in New Zealand agriculture is moving from pilots toward embedded, workflow-specific use, but scale remains uneven. Livestock and packhouse applications are furthest ahead; reliability, connectivity, data rights and independent evidence of return now define the next phase.

Executive Summary

• Adoption is practical but fragmented. The strongest examples attach AI to a defined action: move cattle, measure pasture, grade fruit, detect weeds or support irrigation.
• Livestock remains the most mature segment. Halter and Aimer are operating in commercial farm workflows, although many scale and performance figures remain company-reported.
• Horticulture is advancing through computer vision and robotics. Hectre’s new Government-backed project will extend fruit inspection from external characteristics to internal quality, but the funded work is still research and prototype development rather than scaled deployment.
• The policy environment is becoming more adoption-oriented. Drone reforms, government AI capability-building and primary-sector co-investment are reducing some barriers, while the proposed AI research platform remains publicly unresolved.
• Farmers are becoming more demanding customers. Recent industry commentary identifies “agritech fatigue”: repeated trials that fail, provide little feedback or do not meet expected standards for reliability, warranty and return on investment.
• Evidence remains thin at sector level. New Zealand does not yet have a comprehensive, independently verified measure of AI adoption across farms, orchards, vineyards and packhouses.

The overall picture is therefore one of commercial pockets of maturity inside a sector that is not yet broadly AI-enabled. (aimer-farming.com)

What Has Changed Since the Last Update

Hectre has secured a significant new commercialisation pathway

On 28 August 2026, the Government announced $1.84 million of Primary Sector Growth Fund support for a three-year, $4.6 million project led by Hectre. The project will develop hyperspectral imaging and AI-enabled analysis to identify internal apple characteristics such as maturity, starch content, firmness and defects without cutting the fruit open. (beehive.govt.nz)

The project has clearly defined milestones:

• A proof-of-concept algorithm targeting at least 80% detection accuracy by June 2027.
• A packhouse prototype validated under commercial conditions by June 2028.
• Real-time scanning at commercial throughput speeds.

This is important investment in New Zealand agricultural AI, but it should not be described as evidence that the new system is already operating at scale. It is a funded research and commercialisation programme. Hectre’s existing computer-vision tools are already collecting sizing, colour and quality data from fruit, while the internal-quality capability remains under development. (beehive.govt.nz)

Drone reform has moved from discussion to a defined implementation timetable

On 20 August, the Government announced reforms intended to make routine agricultural drone operations easier. Lower-risk activities such as spraying, spreading fertiliser, applying lime or distributing seed are expected to move from a certification process to a notification-based pathway. The reforms are scheduled to take effect in mid-2027 following detailed design and consultation. (beehive.govt.nz)

This is an enabling policy decision, not evidence of current widespread autonomous or AI-enabled drone deployment. Its significance is that it may reduce the regulatory and operating cost of precision application, especially in orchards, vineyards, market gardens and difficult terrain.

Industry concern has shifted from innovation to execution

At AgriTechNZ’s 25 August seminar in Tauranga, Robotics Plus co-founder Steve Saunders warned of growing “agritech fatigue”. His criticism was directed at technologies that are trialled on farms, fail to deliver, and do not communicate the results back to participants. Farmers increasingly expect new equipment to provide tractor-like reliability, meaningful warranty coverage and a payback period of roughly two years. (farmersweekly.co.nz)

The message is broader than robotics. AI products that produce technically impressive outputs but require unreliable connectivity, duplicate existing data entry or offer unclear financial benefits will face the same adoption resistance.

Adoption infrastructure is receiving more attention

The Helen Clark Foundation’s 24 August discussion paper argues that New Zealand’s agritech opportunity is being constrained by fragmented data ownership, rural connectivity gaps, limited financing for smaller operators and insufficient mechanisms for safe experimentation. It recommends an AgriTech Adoption Fund, regulatory sandboxes, open data infrastructure, farmer-in-residence programmes and stronger digital skills development. (helenclark.foundation)

These recommendations are not Government policy, but they reflect a growing consensus that invention alone will not produce sector-wide adoption. Extension, finance, workforce capability and trust are becoming central parts of the AI debate.

Public-sector AI capability has expanded

The Government Digital Delivery Agency’s 2026 survey recorded 545 AI use cases across 59 public-sector organisations, including 167 cases in operational or deployment stages. That was more than three times the number of operational cases reported in 2025. (digital.govt.nz)

The figures are not agricultural adoption statistics, and they should not be used to imply that farms are adopting AI at the same rate. They do show that the public-sector environment surrounding agriculture—particularly MPI, biosecurity, regulation and policy—is developing more formal AI capability.

Current State of AI Adoption

Livestock and dairy

The evidence still points to dairy and livestock as New Zealand agriculture’s most mature AI adoption segment. The leading systems combine sensors, machine learning, farm-management software and automated or semi-automated recommendations.

Current applications include:

• Virtual fencing and remote stock movement.
• Animal location and behaviour monitoring.
• Heat and reproductive management.
• Pasture measurement and utilisation.
• Feed allocation and grazing planning.
• Livestock weighing and condition assessment.
• Farm administration and staff support.

Halter is the clearest example of AI becoming part of farm infrastructure. Its collars and software influence where cattle graze, monitor animal behaviour and provide pasture-management information. In April, Halter added direct-to-satellite connectivity through One NZ Satellite powered by Starlink, targeting remote beef operations that previously lacked the communications infrastructure required for virtual fencing. (halterhq.com)

Halter’s latest New Zealand beef figures are substantial but self-reported. In August, the company said 500 New Zealand beef farms had joined its satellite solution and 185 farms had adopted its Beef Pro product. It also reported more than one million collars sold globally and nearly 400,000 satellite-enabled collars. These figures indicate meaningful commercial traction, but independent verification of deployment and outcomes remains limited. (fbtech.co.nz)

Aimer represents a different model. It uses smartphone video, satellite data, pasture measurements and farm-specific modelling to produce grazing and feed recommendations. MPI’s co-funded project is intended to expand the system across hundreds of dairy and beef farms and develop an AI agent capable of recommending actions related to productivity, emissions and profit. (aimer-farming.com)

The adoption distinction is important. Aimer reports that its existing platform is used on more than 650 farms, with more than 10,000 pasture measurements recorded each week. However, MPI’s project page recorded only $25,489 of Government contribution spent as at 30 June 2026. The established measurement and decision-support product is therefore further along than the new AI-agent layer. (aimer-farming.com)

Generative AI and farm administration

Generative AI adoption is earlier-stage than embedded livestock and pasture systems.

A DairyNZ-commissioned study found that some farmers are using ChatGPT, Copilot, Claude and Gemini for:

• Drafting emails, policies and employment documents.
• Summarising technical information.
• Analysing spreadsheets and farm data.
• Preparing standard operating procedures.
• Translating or simplifying instructions.
• Building farm-specific chatbots.
• Supporting scenario analysis and business planning.

The same research described overall farmer adoption as small and concentrated among innovators and early adopters. Farmers generally viewed GenAI as a support tool rather than a replacement for practical experience, professional advice or final decision-making. (perrinag.net.nz)

This is consistent with the likely near-term role of GenAI in agriculture: reducing administrative workload, making existing information easier to access and helping people interpret data. It is not yet evidence of autonomous farm management.

Horticulture, packhouses and physical automation

Horticulture has fewer highly visible New Zealand deployments than dairy, but the commercial value of individual systems can be high because small improvements in quality, timing or labour productivity affect export returns directly.

Hectre’s existing computer-vision systems assess fruit size, colour and quality. Its new hyperspectral project aims to add information about internal condition, potentially improving decisions about storage, packing and shipping. If the commercial prototype is validated, the system could shift AI use in packhouses from external grading toward predictive quality management. (beehive.govt.nz)

Robotics Plus provides another route into physical AI. Its Prospr platform is a modular autonomous vehicle designed for orchard and vineyard work. At the Tauranga seminar, the company reported a payback period of more than 1.8 seasons and savings of US$35 per acre per spray pass for some growers. These are vendor-reported figures, not independently established sector averages. (farmersweekly.co.nz)

New Zealand is also adopting overseas-developed AI hardware. Gisborne grower LeaderBrand reported installing a Carbon Robotics G2 LaserWeeder, which uses cameras, AI and lasers to identify and remove weeds in salad-leaf production. The company described the installation as the first of its kind at that scale in New Zealand, but published evidence of longer-term productivity, cost or environmental outcomes remains limited. (leaderbrand.co.nz)

The main constraint in horticultural robotics is not whether computer vision can recognise a fruit, plant or weed in controlled conditions. It is whether the system can operate reliably across different cultivars, weather, dust, canopy structures, terrain and production systems while meeting a commercially acceptable payback period.

Arable farming and irrigation

Arable farming appears relatively receptive to digital technology, but adoption is uneven across use cases.

The Foundation for Arable Research reports widespread use of technologies such as tractor auto-steer and variable-rate irrigation, while uptake of variable-rate seeding, nitrogen application and other precision-agriculture tools remains mixed. FAR identifies doubts about suitability for New Zealand conditions, lack of return-on-investment evidence and rural connectivity as continuing barriers. (far.org.nz)

FAR’s emphasis on low-bandwidth, intermittent or offline functionality is significant. AI systems designed for continuous high-speed connectivity are poorly matched to many New Zealand farms. In practice, edge processing, offline mobile tools and satellite connectivity may matter as much as model sophistication.

The University of Canterbury’s ANZ Soil Moisture Data Assimilation System remains a research project rather than a scaled commercial service. It combines ground sensors, satellite signals and modelling to generate more frequent field-scale soil-moisture estimates. Potential uses include irrigation timing, drought response and water-use efficiency, but evidence of widespread farm deployment is not yet available. (canterbury.ac.nz)

Biosecurity and public-good applications

AI adoption is also expanding beyond private farm businesses.

Biosecurity New Zealand is running a four-month pilot of a generative AI tool to assist with the preparation of import health standards. The stated purpose is to reduce document-heavy work and improve consistency while retaining expert control over substantive decisions. This is a live government pilot, but not yet evidence of a fully operational or autonomous regulatory process. (beehive.govt.nz)

Biosecurity New Zealand has also tested AI-enabled cameras from the University of Exeter’s Vespa AI team to monitor yellow-legged hornet activity and help identify areas of interest. This is an example of AI being used as part of a wider surveillance operation, with human teams still responsible for field response and eradication decisions. (mpi.govt.nz)

These applications matter because agriculture depends on shared systems for disease surveillance, import approvals, environmental monitoring and emergency response. Public-good AI may deliver benefits that individual farms cannot finance independently.

Governance, Policy and Regulation

A light-touch national AI policy

New Zealand continues to follow a principles-based, technology-neutral approach rather than introducing a standalone AI Act. MBIE’s policy position is that existing frameworks—including privacy, consumer protection and human rights law—should generally apply to AI, with targeted changes where necessary. (mbie.govt.nz)

For agricultural businesses, this means AI governance is currently distributed across existing obligations covering:

• Privacy and farm-data handling.
• Consumer and product claims.
• Health and safety.
• Animal welfare.
• Environmental compliance.
• Aviation and drone operations.
• Employment and workplace monitoring.
• Intellectual property and commercial confidentiality.

There is no agriculture-specific AI statute or comprehensive national standard covering farm algorithms, autonomous machinery or virtual fencing.

Drone regulation

The Government’s proposed drone reforms are likely to be one of the most directly relevant regulatory developments for AI-enabled agriculture. They aim to make routine agricultural operations cheaper and faster while retaining stronger controls for high-risk or complex activities. The changes are not expected to take effect until mid-2027. (beehive.govt.nz)

The central governance question will be whether simplified access is matched by adequate controls for:

• Chemical application.
• Airspace and worker safety.
• Drift and environmental effects.
• Data collection over neighbouring properties.
• Automated decision-making.
• Liability when a drone or AI recommendation causes harm.

Public-sector guidance and biosecurity oversight

The Public Service AI Framework emphasises inclusive and sustainable development, human-centred values, transparency and explainability, safety and security, and accountability. New training released on 18 August 2026 is intended to improve AI literacy and safe use across government agencies. (digital.govt.nz)

For agricultural agencies, the most relevant principle is that AI should support accountable human decision-making rather than obscure it. This is particularly important for biosecurity, where an inaccurate recommendation can affect trade access, animal and plant health, and public confidence.

Data ownership and Māori data sovereignty

Data governance remains one of the sector’s unresolved issues. Farm data can contain commercial information about production, genetics, inputs, land capability, emissions and business performance. It may also have cultural significance, particularly where data relates to whenua, taonga species, mātauranga Māori or iwi and Māori agribusiness.

The Helen Clark Foundation’s latest paper recommends open data infrastructure and stronger farmer control over value generated from agricultural data. This does not resolve the question of ownership, but it reflects the strategic risk of allowing valuable data to accumulate in disconnected, privately controlled systems. (helenclark.foundation)

National research infrastructure

The proposed national AI Research Platform remains publicly incomplete. MBIE’s current material continues to describe the BioAI and broader New Zealand AI Platform concepts, including applications in agriculture, forestry and other complex outdoor environments, but does not publicly confirm a final operating platform. MBIE’s August 2026 innovation update still lists establishment of the New Zealand Institute for Advanced Technology as work underway. (mbie.govt.nz)

This is a capability gap rather than evidence of cancellation. For agriculture, a national platform could help address fragmented datasets, limited access to compute, shortage of specialist talent and weak links between research and commercial deployment.

Case Studies

Halter: operating livestock infrastructure

Technology: Solar-powered collars, virtual fencing, GPS, machine learning, satellite connectivity and farm-management software.

Operating status: Commercially deployed across New Zealand and overseas. Halter reports more than 2,000 farmers and ranchers served globally and more than one million collars sold. New Zealand beef adoption figures are company-reported. (halterhq.com)

Strategic significance: Halter demonstrates that AI can become part of a farm’s daily operating infrastructure when it is linked directly to stock movement, labour and pasture use.

Key risk: Reliability is not a secondary product feature. Hardware, communications, software and fallback procedures all become part of farm risk management when a system influences animal movement.

Aimer Farming: from measurement to recommendations

Technology: Smartphone computer vision, satellite data, pasture measurement, paddock modelling and AI-assisted decision support.

Operating status: Aimer reports more than 650 farms using its platform. Its Government-supported AI-agent development programme remains in an expansion and development phase. (aimer-farming.com)

Strategic significance: Aimer is designed around New Zealand’s pasture-based systems and illustrates the value of low-friction, offline-capable tools that fit existing farm routines.

Key risk: AI recommendations will only be as useful as the farm data, assumptions and contextual information behind them. The farmer remains responsible for checking whether the recommendation fits current conditions.

Hectre: extending computer vision inside the fruit

Technology: Computer vision, fruit sizing and colour analysis, quality assessment, hyperspectral sensing and AI algorithms.

Operating status: Existing external-quality systems are operating commercially. The new internal-quality capability is a funded research and prototype programme, with commercial validation targeted for June 2028. (beehive.govt.nz)

Strategic significance: Hectre shows how New Zealand agricultural AI can target a specific supply-chain information gap and potentially create exportable intellectual property.

Key risk: The commercial test will be whether the technology improves storage, packing and market allocation decisions sufficiently to justify integration costs at packhouse throughput.

Robotics Plus: physical AI under commercial pressure

Technology: Autonomous orchard vehicles, modular implements, machine vision and automated spraying or orchard operations.

Operating status: Prospr is being commercialised for orchard and vineyard markets. Reported payback and savings figures are company-reported through industry media. (farmersweekly.co.nz)

Strategic significance: Robotics Plus illustrates New Zealand’s opportunity to export agricultural machinery and operating systems, rather than only adopt overseas products.

Key risk: Agricultural robotics must work across multiple markets and operating environments. A machine proven in New Zealand may require substantial redesign for heat, dust, terrain, crop structure and labour practices elsewhere.

Biosecurity New Zealand: AI in regulatory work

Technology: Generative AI for import health standard development and AI-enabled surveillance cameras for hornet monitoring.

Operating status: Pilot and testing stages, with human experts retaining decision authority. (beehive.govt.nz)

Strategic significance: This broadens agricultural AI beyond commercial production into the systems that protect market access and biological security.

Key risk: Accuracy, traceability and explainability are essential. AI-generated regulatory content must be auditable and checked against authoritative scientific and legal sources.

Trends

1. Embedded AI is ahead of general-purpose GenAI

The strongest commercial systems are connected to an existing workflow and a measurable decision. Generic chatbots are easier to access, but their use remains concentrated among early adopters and administrative tasks.

2. The sector is moving from prediction toward intervention

AI is increasingly connected to physical action:

• Collars influence cattle movement.
• Robots manipulate or spray around plants.
• Computer vision directs packhouse decisions.
• Drones are being positioned for targeted application.
• Sensors and models inform irrigation and soil management.

This increases the value of AI, but also raises the importance of safety cases, human override, maintenance and liability.

3. Reliability and return on investment are adoption infrastructure

Recent industry commentary indicates that farmers are no longer judging agritech mainly by novelty. They want:

• Clear economic benefits.
• Robust performance in local conditions.
• Usable warranties.
• Integration with existing platforms.
• Transparent trial results.
• Support after installation.

This is likely to favour established agribusinesses, co-operatives and vendors able to provide implementation support, rather than standalone applications that add another disconnected dashboard.

4. Connectivity remains a design constraint

The successful New Zealand systems increasingly accommodate rural conditions through offline mobile processing, local infrastructure or satellite communications. Connectivity is not simply an infrastructure issue; it shapes which AI business models are viable.

5. AI adoption is becoming a workforce and knowledge question

The Rural Leaders AI report argues that rural professionals are shifting from primarily transferring knowledge toward interpreting data and facilitating decisions. The Helen Clark Foundation similarly recommends capturing intergenerational farming knowledge, including mātauranga Māori, before it is lost. (ruralleaders.co.nz)

The likely outcome is not the removal of farm expertise. It is a change in how expertise is recorded, accessed and combined with machine-generated analysis.

6. New Zealand’s strategic opportunity is export-oriented

New Zealand’s domestic market is relatively small. Companies such as Halter, Hectre and Robotics Plus are therefore designing for international scale from the beginning. The domestic sector functions as a test environment, but commercial success depends on building systems that can be adapted to overseas production models.

7. Evidence is improving, but remains uneven

The sector now has more pilots, investment announcements and company-reported deployments. It has fewer independent, longitudinal studies showing:

• Whole-farm profitability effects.
• Effects across different farm sizes.
• Long-term animal-welfare outcomes.
• Environmental impacts.
• Failure rates and maintenance costs.
• Distribution of benefits between technology providers and producers.

Announcements should continue to be separated from operating deployments and from independently validated outcomes.

Outlook

Over the next two years, AI adoption in New Zealand agriculture is most likely to advance through five pathways.

1. Expansion of existing livestock and pasture platforms
   Halter, Aimer and related systems will add more recommendations, reproductive analytics, emissions functions and integrations with farm-management software.

2. Commercial validation in horticulture
   Hectre’s internal fruit-quality project, orchard robotics and AI-enabled weed control will provide clearer evidence of whether computer vision and automation can deliver acceptable payback in New Zealand conditions.

3. More regulated use of drones and autonomous machinery
   Drone reforms planned for mid-2027 may increase adoption, but the practical effect will depend on implementation detail, operator capability and environmental safeguards.

4. Growth of low-risk generative AI use
   Farm administration, translation, staff training, technical search and report preparation are likely to grow faster than fully autonomous production decisions.

5. Greater pressure for integrated data systems
   The next commercial advantage may come less from another isolated model and more from connecting animal, pasture, weather, soil, processing, emissions, traceability and market data.

The main uncertainty is whether adoption will broaden beyond well-capitalised farms and export-focused businesses. Without shared infrastructure, independent trials, advisory support and accessible financing, AI could improve leading operations while widening the technology gap across the sector.

Overall Assessment

As at 1 September 2026, AI in New Zealand agriculture is best described as commercially credible but not yet sector-wide.

The most mature systems are workflow-embedded: Halter’s livestock operating platform, Aimer’s pasture intelligence, Hectre’s fruit-quality tools and selected robotics and weed-control systems. These applications solve defined operational problems and can be evaluated against labour, input, productivity or quality outcomes.

The next stage is more demanding. Farmers and growers are asking whether systems work reliably in local conditions, whether they integrate with existing data, whether vendors stand behind them and whether benefits justify the cost. Recent industry criticism of “agritech fatigue” suggests that weak trials and exaggerated claims may now be a more serious barrier than lack of interest.

Government is strengthening the surrounding infrastructure through co-investment, public-sector capability-building, biosecurity pilots and planned drone reform. However, the sector still lacks a comprehensive adoption baseline, widespread independent validation and settled rules for agricultural data ownership and value sharing.

New Zealand’s strongest opportunity is not to deploy AI everywhere at once. It is to build trusted, exportable systems around the country’s distinctive production environments: pasture-based livestock, high-value horticulture, remote farms, biosecurity-sensitive trade and data-rich food supply chains. The technologies most likely to scale will be those that combine sound agronomy, practical farm knowledge, resilient infrastructure and demonstrable returns.