AI in Agriculture

Executive Summary

• Adoption is becoming more operational. AI is increasingly embedded in systems that move livestock, monitor pasture, control dairy-yard equipment, analyse fruit or support farm decisions.
• Livestock remains the strongest commercial segment. Halter has added an operational AI agent to support the reliability of its connected-farming platform, while Aimer continues expanding AI-assisted pasture measurement and decision support.
• Generative AI is still an early-adopter technology on farms. DairyNZ-commissioned research found farmers using tools such as ChatGPT, Claude, Gemini and Copilot for administration, knowledge access, communication and scenario analysis, but adoption remains small.
• The pipeline is broadening. New funding is supporting AI-enabled sheep-meat quality measurement, three-dimensional crop scanning and orchard yield forecasting. These projects remain development work, not proof of scaled deployment.
• Horticulture is gaining larger-scale digital infrastructure. A five-year kiwifruit programme will combine on-orchard innovation, advanced decision-support tools and applied science, although the public material does not establish that all components are AI-based.
• Reliability and service have become adoption requirements. AgriTechNZ reports that growers have stopped funding unproven technology and are prioritising dependable products, service and a credible path to value.
• The evidence base is improving but remains incomplete. New national business-survey work should eventually provide better adoption data, but there is still no comprehensive, independently verified measure of AI use across New Zealand farms, orchards and packhouses.

The sector is therefore moving beyond experimentation, but commercial pockets of maturity still sit inside a largely uneven adoption landscape.

What Has Changed Since the Last Update

Halter has moved agentic AI into live operating infrastructure

The most significant new development is not a new farm robot or autonomous herd-management product. It is the deployment of an AI agent inside the infrastructure supporting a commercial livestock platform.

On 22 September 2026, AWS reported that Halter had deployed “Clank”, an agent built with Amazon Bedrock and Anthropic Claude. The system analyses production alerts, investigates likely causes and prepares code changes for engineers to review. Halter said Clank was operating across more than 16,000 monitoring alerts and had saved more than 215 hours of engineering time in its first months, while automating more than 90 weekly tasks. These are AWS- and Halter-reported figures, not independently audited results. AWS case study aboutamazon.com.au

This matters because it shows AI adoption at a different layer of the agricultural technology stack. The agent is not making autonomous animal-welfare decisions or replacing farmers. It is helping maintain the reliability of the software, connectivity and alerting systems on which farmers depend. In connected agriculture, that reliability work is part of the product.

Virtual fencing is receiving broader farmer-facing validation

On 29 September, Beef + Lamb New Zealand reported that more than 100 Northland farmers attended a field day to see Halter virtual fencing in operation and discuss pasture measurement, multiple-mob management and data-supported decision-making. Attendance is evidence of sector interest and extension activity, not of adoption at that scale. Beef + Lamb New Zealand beeflambnz.com

A separate September case from Marlborough illustrates the distinction between installation and verified outcomes. Bonaveree Farm has placed 875 cattle on Halter’s Beef Pro system, but the farmer said it was too early to discuss results confidently. The system is operating; its longer-term production, environmental and financial effects remain to be demonstrated. Marlborough App marlboroughapp.co.nz

New grants have expanded the AI development pipeline

The Agricultural and Marketing Research and Development Trust’s August grant round, reported on 17 September, funded nine projects with approximately $500,000 in total support. Two projects have direct AI relevance:

• neXtgen Agri received $109,608 toward an AI-enabled ultrasound and computer-vision system to measure intramuscular fat in live sheep.
• HoloCrop received $54,000 toward a vehicle-agnostic three-dimensional crop-scanning system for yield forecasting, harvest planning and supply-chain efficiency.

Both are development projects. The funding demonstrates sector demand and a pathway toward commercial tools, but not deployment across farms or proven economic returns. AGMARDT funded projects agmardt.org.nz

AGMARDT’s accompanying analysis is also significant. It reported more applications involving apps, drones and AI for precision agrichemical use and yield prediction than in previous rounds, while noting that many promising ideas were being developed in isolation. The organisation identified a growing need for more strategic collaboration and clearer routes to adoption. AGMARDT: What we heard agmardt.org.nz

The horticulture pipeline is becoming more structured

The Government and kiwifruit industry announced a five-year, $47.87 million programme in June, including $19.14 million of Government investment. It will combine on-orchard innovation, advanced decision-support tools and applied science across New Zealand’s kiwifruit-growing regions. Its published objectives include improving yield and fruit quality while using water and nutrients more efficiently. (Beehive announcement; Zespri programme statement) beehive.govt.nz

This is an important adoption-enabling investment, but it should not be counted as evidence that AI systems are already running across kiwifruit orchards. The public descriptions refer to decision support and applied science without specifying which components use machine learning or generative AI.

The sector’s adoption test has become more demanding

AgriTechNZ’s September account of its Tauranga event described a clear change in buyer behaviour: growers had stopped funding unproven technology, while reliability and service were determining what was purchased. The organisation also argued that many apparent technology problems were actually matching problems between producers with real operational needs and companies able to solve them. AgriTechNZ agritechnz.org.nz

This reinforces the previous update’s warning about “agritech fatigue”. AI vendors are now being judged less on whether a model can produce an impressive demonstration and more on whether the complete system works through a season, integrates with existing routines and has a support model that farmers can trust.

Current State of AI Adoption

Livestock and pastoral systems

Livestock remains New Zealand’s most mature AI application area because the use cases connect directly to recurring operational decisions:

• Where cattle graze and when mobs move.
• Which animals require attention.
• How pasture is measured and allocated.
• How animals are weighed, drafted or monitored.
• How dairy-yard equipment responds to animal flow.
• How genetic and meat-quality decisions are supported.

Halter is the clearest example of a commercial physical-AI platform. Its system combines collars, positioning, animal-behaviour data, pasture information, communications infrastructure and farm-management software. In April, the company said it served more than 2,000 farmers and ranchers across New Zealand, Australia and the United States and had sold more than one million collars globally. Those figures are company-reported. Its direct-to-satellite service is intended to extend virtual fencing to remote beef operations that previously lacked suitable connectivity. Halter halterhq.com

The new Clank deployment shows a second form of maturity: AI is being used to maintain the platform itself. This is strategically important because agricultural systems operate in conditions where failure can affect animals, labour, pasture utilisation and farm confidence. Halter’s decision to keep human engineers in the approval loop for generated code is consistent with the risk profile of a system connected to live farm operations. The reported savings remain vendor-reported.

Aimer is operating at a different point in the stack. Its AIMER platform combines smartphone pasture measurement, satellite data and farm-specific modelling. Aimer says the platform is used on more than 650 farms and records more than 10,000 pasture measurements each week. The company’s Government-supported programme is expanding the existing system and developing “Ask AIMER”, a conversational interface intended to turn measurements into tailored recommendations. These scale figures are self-reported, and the AI-agent component remains under development. (Aimer Farming; MPI project page) aimer-farming.com

Livestock AI is also beginning to move into meat-quality and dairy-workflow applications. neXtScan’s planned ultrasound system targets a bottleneck in intramuscular-fat measurement, while Livestock Visibility Solutions’ Flow system uses cameras and machine learning to control dairy backing gates. Flow has been trialled on eight commercial farms, with the company reporting reductions in milking time and lameness. These results are reported through industry media and by the developer; they should not be treated as independent sector averages. Farmers Weekly International fwi.co.uk

Generative AI and farm administration

Generative AI remains less mature than embedded livestock and pasture systems.

A DairyNZ-commissioned Perrin Ag study, published in August, found that early adopters were using ChatGPT, Claude, Gemini and Copilot to:

• Draft emails, policies and employment documents.
• Summarise technical information.
• Analyse spreadsheets and farm data.
• Prepare standard operating procedures.
• Translate or simplify information.
• Build farm-specific chatbots.
• Explore business and farm-management scenarios.

The study found that current use was still small and concentrated among early adopters. DairyNZ’s guidance continues to position GenAI as an assistant rather than a replacement for farmer judgement, particularly because outputs can be inaccurate or overly confident. (Perrin Ag report summary; DairyNZ guidance) perrinag.net.nz

The likely near-term value is therefore administrative and interpretive rather than autonomous. Voice interfaces, multilingual support and farm-specific retrieval systems may be particularly useful for distributed teams and seasonal workers, provided the underlying source material is trusted and the system clearly distinguishes authoritative information from generated suggestions.

Horticulture, packhouses and crop monitoring

Horticulture has a broad range of AI-related activity, but the maturity of individual projects varies considerably.

Hectre’s existing computer-vision products operate in fruit-quality workflows, including sizing, colour and external defect assessment. Its Government-backed hyperspectral project is intended to detect internal apple characteristics such as maturity, starch content, firmness and defects without cutting fruit open. The most recently published project milestones still describe proof-of-concept development and commercial validation targets rather than an operating system at scale. Primary Sector Growth Fund announcement

HoloCrop’s grant-funded three-dimensional scanning project and other yield-forecasting initiatives represent a different stage: they are attempts to make crop intelligence more accessible through common farm vehicles and practical data collection. The commercial question will be whether growers receive forecasts early and accurately enough to improve labour, harvest, packing and market decisions.

There is also evidence that the New Zealand market will not automatically adopt every overseas AI product. Bitwise Agronomy, an Australian-based company founded by a New Zealand entrepreneur, piloted its AI crop-forecasting technology in New Zealand but did not secure grower uptake through the pilot. The company concluded that New Zealand growers had a stronger need for quality assessment than yield forecasting. This is a useful example of market fit constraining adoption, even where the underlying technology is available. Farm Innovation Network farminnovationnetwork.co.nz

Precision agriculture, drones and irrigation

Precision agriculture remains an important foundation for AI, but not every digital or automated system is itself AI-enabled. New Zealand farmers already use tools such as guidance systems, variable-rate application, sensors and satellite imagery, while uptake varies by farm type, region and expected return.

The strongest emerging direction is toward systems that combine multiple data sources and produce an actionable recommendation. Potential applications include variable-rate fertiliser, targeted agrichemical use, crop and pasture forecasting, soil-moisture management and environmental monitoring.

The Government’s proposed agricultural drone reforms remain an enabling measure rather than evidence of present AI adoption. Lower-risk agricultural activities are expected to move toward a notification-based pathway from mid-2027, subject to detailed design and implementation. The reforms could lower the cost of precision spraying and spreading, but their effect will depend on operator capability, environmental safeguards, airspace controls and liability arrangements.

Public-good and biosecurity applications

AI adoption also extends beyond private farm businesses into systems that protect production and market access.

Biosecurity New Zealand has been testing generative AI to assist with import health-standard preparation and AI-enabled camera systems for surveillance of yellow-legged hornets. These remain pilots or testing activities, with human experts retaining decision authority. There is no public evidence in the latest material reviewed that either has become a fully automated, sector-wide operational service.

This distinction remains important. Public-good AI may create substantial value through faster surveillance, better regulatory workflows and earlier detection, but the benefits are difficult to attribute to individual farms and depend on institutional trust, scientific validation and auditability.

Governance, Policy and Regulation

National policy remains principles-based

New Zealand continues to use a relatively light-touch, technology-neutral approach to AI. The Government’s AI strategy and responsible-business guidance emphasise OECD principles, including privacy, fairness, safety, security, transparency and accountability. The framework relies mainly on existing law rather than a standalone AI Act. MBIE AI strategy and guidance mbie.govt.nz

For agricultural businesses, relevant obligations remain distributed across:

• The Privacy Act and data-security duties.
• Consumer-protection and product-claim rules.
• Health and safety requirements.
• Animal-welfare obligations.
• Environmental and agrichemical controls.
• Aviation and drone rules.
• Employment and workplace-monitoring requirements.
• Intellectual-property and commercial-confidentiality rules.

This approach gives businesses flexibility, but it also leaves vendors and producers responsible for assembling their own governance frameworks around high-consequence systems.

Reliability, human oversight and liability

As AI moves from analysis toward physical intervention, governance requirements become more practical and operational. A virtual fence, autonomous vehicle, AI-controlled gate or precision-application drone needs more than model accuracy. It requires:

• A clear human-override process.
• Defined fallback modes.
• Monitoring for system degradation.
• Maintenance and support arrangements.
• Records of decisions and interventions.
• Clear allocation of liability when the system fails.

The Halter Clank example is instructive because the AI agent prepares changes while engineers retain review and approval authority. Similar boundaries are likely to be necessary for farm-facing decision systems, especially where animals, chemical application or worker safety are involved.

Agricultural data and Māori data sovereignty

Farm data may include production volumes, genetics, land capability, emissions, animal behaviour, staff information and commercial performance. Control of that data is becoming more important as AI systems derive value from combining datasets across farms, companies and supply chains.

The issue is not limited to commercial ownership. Data relating to whenua, taonga species, iwi and Māori agribusiness may also raise questions of tikanga, consent, governance and benefit sharing. Public research proposals for New Zealand’s AI platform have explicitly referred to Māori participation and Māori data sovereignty, but practical arrangements for agricultural data remain unsettled.

National AI research infrastructure

The Government has committed up to $70 million over seven years for an AI research platform through the New Zealand Institute for Advanced Technology. Publicly described concepts include BioAI, physical AI for outdoor environments and AI platforms focused on agriculture, forestry, robotics and complex real-world systems. MBIE’s latest public material also identifies applied AI tools that use New Zealand’s domain datasets and are accessible to smaller operators as a national priority. (MBIE AI Research Platform; Science Investment Plan) mbie.govt.nz

The latest public position remains transitional. The material reviewed confirms the funding intention and competing concepts, but does not establish a fully operating, agriculture-facing national AI platform. For agriculture, the value of such infrastructure would depend on shared datasets, access for small and medium-sized operators, real-world test sites, technical support and clear rules for commercialisation.

Better measurement is being introduced

In April, the Government announced that the new Survey of Business Operations would include questions about business AI use. The survey is intended to cover approximately 20,000 New Zealand businesses with six or more employees across sectors. It should eventually improve national measurement of adoption and barriers, although farms and small agricultural businesses may not be fully captured by the same statistical lens. Beehive announcement beehive.govt.nz

Until those results are available, industry adoption claims will continue to rely heavily on vendor figures, targeted studies and case reports.

Case Studies

Halter: AI as connected livestock infrastructure

Technology: Solar-powered collars, virtual fencing, positioning, animal-behaviour monitoring, pasture intelligence, satellite connectivity and agentic AI for platform operations.

Operating status: Commercially deployed across New Zealand and overseas. Halter reports more than 2,000 customers and more than one million collars sold globally. The Clank agent is operating inside Halter’s engineering workflows. Customer, collar and efficiency figures are company- or AWS-reported. (Halter satellite release; AWS case study) halterhq.com

Strategic significance: Halter demonstrates that agricultural AI can become part of operating infrastructure rather than a standalone advisory tool.

Key risk: When software influences animal movement and farm routines, uptime, communications, hardware failure and human fallback procedures are central to the safety case.

Aimer Farming: from pasture measurement to recommendations

Technology: Smartphone computer vision, satellite data, pasture measurement, farm modelling and conversational decision support.

Operating status: Aimer reports use across more than 650 farms. The Government-supported expansion and “Ask AIMER” capability remain in development and scale-up. (Aimer Farming; MPI project information) aimer-farming.com

Strategic significance: Aimer illustrates the value of low-friction tools that fit pasture-based farming routines and convert measurement into a practical next action.

Key risk: Recommendations are only as reliable as the pasture data, farm assumptions and contextual information behind them.

Flow: computer vision in the dairy yard

Technology: Cameras, machine learning and farm-specific models that regulate backing-gate movement according to cow density and yard conditions.

Operating status: Trialled on eight commercial New Zealand dairy farms. Reported improvements include shorter milking times and fewer lame cows, but the results are developer-reported and not yet an independent sector benchmark. Farmers Weekly International fwi.co.uk

Strategic significance: Flow shows how relatively narrow AI applications can target labour consistency, animal flow and welfare without attempting to automate the entire farm.

Key risk: Performance must remain stable across different herd sizes, parlour designs, lighting conditions, staff practices and yard configurations.

Hectre, neXtScan and HoloCrop: an expanding pipeline

Technology: Hyperspectral fruit analysis, AI-enabled ultrasound, computer vision and three-dimensional crop scanning.

Operating status: Hectre’s external fruit-quality tools are commercial; its internal-quality project is in research and prototype development. neXtScan and HoloCrop are grant-funded development projects. (Hectre project announcement; AGMARDT grants) agmardt.org.nz

Strategic significance: These projects target information gaps that have direct effects on breeding, harvest planning, packing, storage and export quality.

Key risk: The systems must demonstrate accuracy at commercial throughput and generate enough value to justify integration into existing production and packhouse operations.

DairyNZ and Perrin Ag: practical GenAI adoption

Technology: General-purpose large language models, custom farm chatbots and AI-assisted document, knowledge and data workflows.

Operating status: Early-adopter use, not broad sector deployment. DairyNZ has established guidance and a community-of-interest approach rather than endorsing a single product. (DairyNZ guidance; Perrin Ag study) perrinag.net.nz

Strategic significance: The most immediate value of GenAI may be reducing administrative load and making trusted farm knowledge easier to retrieve.

Key risk: Hallucinations, privacy breaches, poor source control and overconfidence could reduce trust if farmers treat generated responses as professional or regulatory advice.

Trends

1. AI is moving from dashboards into operating systems

The most important shift is from AI that reports information to AI that participates in a workflow. Halter’s agent supports platform reliability; Aimer is developing conversational recommendations; Flow controls a physical dairy-yard process; Hectre’s project aims to influence packhouse decisions.

This increases potential value, but also increases the need for monitoring, human oversight and clear accountability.

2. Reliability is becoming a competitive advantage

AgriTechNZ’s latest commentary suggests that farmers are no longer willing to absorb the risk of poorly supported pilots. Reliability, warranty, service response, integration and seasonal performance are becoming as important as model accuracy.

This is likely to favour vendors that can provide a complete operating system—hardware, connectivity, software, support and training—rather than a technically impressive model with no implementation capability.

3. Funding demand is outpacing coordinated adoption

AGMARDT reported strong demand for AI, drone and precision-agriculture projects, but also found many ideas addressing isolated parts of larger problems. The result is a familiar pattern: a growing pipeline of prototypes without enough shared infrastructure, commercial testing or coordinated pathways to adoption.

The next stage will require more industry-led problem definition, common data standards and larger demonstrations that can produce credible evidence for multiple growers.

4. New Zealand’s physical environment remains both an advantage and a constraint

Farms, orchards and forests provide valuable real-world conditions for testing AI in rugged, variable environments. They also expose systems to weak connectivity, changing weather, dust, terrain, different cultivars, animal behaviour and seasonal labour.

This favours edge processing, offline capability, resilient communications and models that can operate with incomplete data. It also creates an export opportunity for systems designed around difficult outdoor environments.

5. Data integration is more important than another isolated model

DairyNZ’s research identifies fragmented data as a barrier to wider GenAI adoption. The same issue applies across the sector: animal, pasture, weather, soil, machinery, emissions, processing and market data often sit in separate systems.

The commercial opportunity is increasingly likely to lie in connecting these datasets and producing decisions that farmers can understand and act on, rather than adding another independent dashboard.

6. Rural professionals are becoming interpreters and facilitators

A June 2026 Rural Leaders report found that AI is shifting rural professionals away from simple knowledge transfer and toward data interpretation, critical evaluation and decision facilitation. More than half of respondents did not feel well supported through the change. Rural Leaders report ruralleaders.co.nz

This suggests that adoption will depend partly on advisers, consultants, veterinarians, rural retailers and industry bodies being able to test AI outputs against farm context. Human expertise is not disappearing; its value is moving toward interpretation, judgement and trust.

7. Export potential continues to shape investment

New Zealand’s domestic market is too small to support many agritech businesses on its own. Halter, Hectre, Robotics Plus and other companies are therefore building products that can operate in overseas production systems.

The domestic farm sector remains an important test environment, but export success will depend on adapting systems to different regulations, labour models, connectivity conditions, farm sizes and production systems.

Outlook

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

1. More AI inside established platforms
   Companies with existing farm data, customer relationships and operating infrastructure are likely to add conversational interfaces, predictive alerts and workflow automation before fully autonomous decision-making becomes common.

2. Commercial validation of targeted physical systems
   Projects involving dairy-yard control, livestock measurement, orchard scanning, weed detection and packhouse inspection will face clearer tests of accuracy, reliability and payback.

3. Expansion of remote livestock management
   Satellite connectivity should extend virtual fencing and livestock-monitoring products to more remote beef and hill-country operations, although the economics of hardware, service and communications will determine the pace.

4. More cautious use of GenAI on farms
   Farm administration, translation, technical search, training and business planning are likely to grow faster than systems that make unsupervised animal, chemical or environmental decisions.

5. A stronger adoption and evidence infrastructure
   The national business survey, AI research investment and new industry collaboration models could improve measurement and reduce duplicated experimentation. Their practical value will depend on whether small and medium-sized operators can access the resulting tools and datasets.

6. Regulatory pressure around physical AI
   Drone reform expected from mid-2027 may expand precision application, but safety, chemical drift, environmental effects, privacy and liability will remain central implementation issues.

The main uncertainty is distribution. AI could substantially improve productivity in well-capitalised, data-rich operations while leaving smaller farms, remote producers and less-connected regions behind. The sector will need practical extension, finance, interoperability and independent evaluation if adoption is to broaden beyond early adopters.

Overall Assessment

As at 1 October 2026, AI in New Zealand agriculture is best described as commercially credible, increasingly operational and still far from sector-wide.

The strongest evidence is in livestock and pastoral systems. Halter is operating a connected physical-AI platform at meaningful commercial scale and has begun using agentic AI to support the reliability of that platform. Aimer is turning pasture measurement into farm-specific decision support, while targeted dairy and livestock applications are entering commercial trials.

Generative AI on farms is advancing more cautiously. Its early value lies in administration, knowledge access, communication and scenario analysis. DairyNZ’s research indicates genuine interest, but also low adoption, limited trust and a need for better data foundations.

Horticulture shows a widening pipeline of promising projects, from hyperspectral apple analysis to three-dimensional crop scanning and AI-assisted livestock quality measurement. Most remain in development or validation. Government and industry investment is creating stronger conditions for experimentation, but funding announcements should not be mistaken for operating deployment.

The central adoption question has changed. It is no longer whether AI can produce useful predictions in a demonstration. It is whether the complete system can work reliably in New Zealand conditions, integrate with existing farm practice, protect data, support human judgement and produce a return that farmers can see.

New Zealand’s opportunity is to build trusted, exportable AI around its distinctive agricultural environments: pasture-based livestock, high-value horticulture, remote farms, biosecurity-sensitive trade and data-rich food supply chains. The technologies most likely to endure will be those that combine sound agricultural knowledge, resilient infrastructure, clear accountability and independently credible results.