AI in agriculture: what's new, what works and how to start
Latest in farm AI
Every development →ForTy map tells natural forest from plantations at 10-metre detail
Google DeepMind and partners released Forest Typology (ForTy) 2020, a global 10-metre map sorting land into six forest and land-use types, with an early yearly series for 2017–2025 showing where natural forest became plantation or farm. It is meant as a baseline for the EU Deforestation Regulation. source
Global Fishing Watch rebuilds its AI models for spotting fishing
Global Fishing Watch replaced the core AI models it uses to estimate apparent fishing activity and classify vessels, reprocessing data from 2012 onward. It says the update cuts cases where vessels near ports or in transit were wrongly marked as fishing. source
Google's agricultural understanding models open in public preview
Google opened its Agricultural Understanding models (field boundaries, crop type and field events) to developers in public preview; they now cover 11 countries. It reports that Gujarat's Krushi Pragati platform used them for over 900 million SMS advisories to over 600,000 farmers in a year. source
AI-supported genetic atlas of sorghum from ICRISAT's genebank
Google DeepMind and Google.org are working with ICRISAT on a sorghum genetic atlas that pairs genetic analysis with machine learning to find promising plant combinations for breeders, drawing on a genebank of over 44,000 seeds from over 90 countries. source
How many are using it
of US agriculture, forestry and fishing businesses used AI in the fortnight 21 Sep 2026 to 4 Oct 2026 (US Census Bureau, BTOS). All businesses: 24.6%. ±2.6 points.
More official figures
| Publisher | What it measures | Figure | Source |
|---|---|---|---|
| Australian Bureau of StatisticsAU · year ended 30 June 2025 | Characteristics of Australian Business 2024-25, data cube 6 (Use of ICT), Table 2: Selected ICTs used by the business, by industryAgriculture, Forestry and Fishing: 2.8% used artificial intelligence (AI) | 2.8% | source |
| Statistics CanadaCA · 12 months to Q2 2026 | Analysis in Brief: use of AI in producing goods or delivering services (Canadian Survey on Business Conditions, second quarter 2026)agriculture, forestry, fishing and hunting (4.5%) | 4.5% | source |
| Statistics CanadaCA · 12 months to Q2 2025 | Analysis in Brief: use of AI in producing goods or delivering services (Canadian Survey on Business Conditions, second quarter 2025)agriculture, forestry, fishing and hunting (1.8%) | 1.8% | source |
| Statistics CanadaCA · 12 months to Q2 2024 | Analysis in Brief: use of AI in producing goods or delivering services (Canadian Survey on Business Conditions, second quarter 2024)agriculture, forestry, fishing and hunting (0.7%) | 0.7% | source |
Start from your role
Look up any job →Farmers and farm managers
Notes into a report
It is worth trying only with the checks on this page. Part of the work in 1 of the 1 jobs we map for this role.
- Does AI save time on notes and routine records? not enough evidence yet (strength 1 of 5)
- Do AI assistants save time on everyday office work? not enough evidence yet (strength 0 of 5)
Draft the commentary on a finance report
It is worth trying only with the checks on this page. Part of the work in 1 of the 1 jobs we map for this role.
- Does AI make financial analysis better? not enough evidence yet (strength 1 of 5)
Farm and ranch workers
Notes into a report
It is worth trying only with the checks on this page. Part of the work in 3 of the 7 jobs we map for this role.
- Does AI save time on notes and routine records? not enough evidence yet (strength 1 of 5)
- Do AI assistants save time on everyday office work? not enough evidence yet (strength 0 of 5)
Forestry and fishing
Pressure-test a plan
It is worth trying only with the checks on this page. Part of the work in 2 of the 6 jobs we map for this role.
- Does AI business advice help small businesses? not enough evidence yet (strength 0 of 5)
Notes into a report
It is worth trying only with the checks on this page. Part of the work in 2 of the 6 jobs we map for this role.
- Does AI save time on notes and routine records? not enough evidence yet (strength 1 of 5)
- Do AI assistants save time on everyday office work? not enough evidence yet (strength 0 of 5)
Agricultural scientists and technicians
Make slides and visuals
It is worth trying only with the checks on this page. Part of the work in 4 of the 5 jobs we map for this role.
- Does AI help people make better slides? not enough evidence yet (strength 1 of 5)
- Does AI make creative and design work better? not enough evidence yet (strength 1 of 5)
Notes into a report
It is worth trying only with the checks on this page. Part of the work in 2 of the 5 jobs we map for this role.
- Does AI save time on notes and routine records? not enough evidence yet (strength 1 of 5)
- Do AI assistants save time on everyday office work? not enough evidence yet (strength 0 of 5)
Pressure-test a plan
It is worth trying only with the checks on this page. Part of the work in 1 of the 5 jobs we map for this role.
- Does AI business advice help small businesses? not enough evidence yet (strength 0 of 5)
Research a question
It is worth trying only with the checks on this page. Part of the work in 1 of the 5 jobs we map for this role.
- Does AI help people find information for their work? not enough evidence yet (strength 0 of 5)
- Does AI help with consulting and analysis work? may work for tasks that sit inside what the AI does well (strength 2 of 5)
- Does AI speed up systematic reviews without more errors? not enough evidence yet (strength 1 of 5)
What's new: key developments
Oct 2026 · 4 developments
- Google DeepMind; Google Research; World Resources Institute; IIASAResearchNew ForTy map tells natural forest from plantations at 10-metre detailGoogle DeepMind and partners released Forest Typology (ForTy) 2020, a global 10-metre map sorting land into six forest and land-use types, with an early yearly series for 2017–2025 showing where natural forest became plantation or farm. It is meant as a baseline for the EU Deforestation Regulation. source
- Global Fishing WatchModelNew Global Fishing Watch rebuilds its AI models for spotting fishingGlobal Fishing Watch replaced the core AI models it uses to estimate apparent fishing activity and classify vessels, reprocessing data from 2012 onward. It says the update cuts cases where vessels near ports or in transit were wrongly marked as fishing. source
- GoogleProductNew Google's agricultural understanding models open in public previewGoogle opened its Agricultural Understanding models (field boundaries, crop type and field events) to developers in public preview; they now cover 11 countries. It reports that Gujarat's Krushi Pragati platform used them for over 900 million SMS advisories to over 600,000 farmers in a year. source
- Google DeepMind; Google.org; ICRISATResearchNew AI-supported genetic atlas of sorghum from ICRISAT's genebankGoogle DeepMind and Google.org are working with ICRISAT on a sorghum genetic atlas that pairs genetic analysis with machine learning to find promising plant combinations for breeders, drawing on a genebank of over 44,000 seeds from over 90 countries. source
Aug 2026 · 1 development
- Department for Environment, Food & Rural Affairs (Defra) and Innovate UKRulesNew UK opens £20 million fund for farm robots and automationDefra opened a £20 million Farming Innovation Programme round for businesses working with researchers and farmers on robots and automated systems that plant, tend and harvest crops; livestock uses are eligible. source
Jul 2026 · 1 development
- European Parliament and Council (EU)RulesNew EU AI Act Omnibus resets high-risk dates and moves machinery to Annex I Section BThe Digital Omnibus on AI amended the AI Act: Annex I high-risk duties now apply from 2 August 2028 and Annex III duties from 2 December 2027. Both the agricultural and forestry vehicles regulation and, after the Omnibus, the Machinery Regulation sit in Section B of Annex I, for which only Article 6(1), Article 60a and Articles 102 to 112 apply; AI requirements reach tractors and farm machinery through those sector laws rather than directly. In force 27 Jul 2026. source
Jun 2026 · 1 development
- KrishokChat authors (Reza, Nimi, Shahid)BenchmarkNew KrishokChat: Bengali farm-advice benchmark that tracks chemical errorsKrishokChat has 85,979 Bengali items built from 284 government publications, plus 1,000 real farmer queries. Even with the right evidence supplied, the models tested still showed chemical hallucination rates of 4–7%. source
May 2026 · 1 development
- AgroTools authors (Ye et al.)BenchmarkNew AgroTools: benchmark for AI agents that use farm toolsAgroTools tests whether multimodal AI agents can use 14 agricultural tools correctly across 539 tasks with 1,097 images. Thirteen open and closed models were tested and none was reliable. source
Mar 2026 · 1 development
- Ministry of Agriculture, Forestry and Fisheries (MAFF), JapanRulesNew Japan revises its safety guideline for autonomous farm machineryMAFF issued a revised safety guideline for robot farm machinery. It sets risk-assessment duties and roles for makers, sellers, adopters and users, covers machines watched on site or remotely, and has sections for tractors, rice transplanters, combines, mowers, tea-field machines and small carts. source
Feb 2026 · 1 development
- Carbon RoboticsModelNew Large Plant Model lets LaserWeeders handle new fields in minutesCarbon Robotics launched its Large Plant Model, a plant-detection model trained on labelled plant images gathered by its LaserWeeder machines. With Plant Profiles, growers tune it to a field from two or three photos. source
Jan 2026 · 1 development
- Nanjing Agricultural University; Nanjing University of Science and TechnologyModelNew Sinong 1.0 open Chinese agriculture LLMs (8B and 32B)Nanjing Agricultural University released Sinong 1.0, open 8B and 32B agriculture language models trained on more than 4 billion tokens of domain data, including books, papers, policies and standards, across fields from crop breeding to veterinary medicine. source
Dec 2025 · 1 development
- ai71 (UAE), with the Gates Foundation, CGIAR, Embrapa, FAO, IFAD, the World Bank, Digital Green and othersModelNew AgriLLM: open agriculture LLMs, training data and evaluation setai71 released AgriLLM, agriculture fine-tunes of Llama 3.3 70B and Qwen3 30B-A3B, with a 146,000-pair training set and an 800-question evaluation set. The models are meant to sit inside retrieval (RAG) systems for farmers and extension agents, not to work alone. source
Nov 2025 · 1 development
- SyngentaProductNew Cropwise farm platform opened to outside developersSyngenta opened its AI-powered Cropwise farm-management platform to third-party developers, so outside tools can reach the farms it covers. Syngenta says Cropwise covers over 70 million hectares in more than 30 countries. source
Sep 2025 · 1 development
- Syngenta Vegetable Seeds; Heritable AgricultureProductNew AI to choose which vegetable varieties to offer growers by regionSyngenta Vegetable Seeds signed a collaboration with Heritable Agriculture, an Alphabet spinout, to use AI on historical geographic and crop-trial data to predict the best-performing commercial varieties for different regions. source
Aug 2025 · 1 development
- State Council of the People's Republic of ChinaRulesNew China's 'AI+' policy calls for AI in breeding, farm machinery and dronesThe State Council's opinions on the 'AI+' action (Guo Fa [2025] No. 11) include a section on the digital and intelligent upgrade of agriculture: AI-driven breeding, smart farm machinery, agricultural drones and robots, and AI for farm management and risk prevention. source
Jul 2025 · 4 developments
- Google DeepMindModelNew AlphaEarth Foundations satellite embeddings released in Earth EngineGoogle DeepMind released yearly embeddings from its AlphaEarth Foundations model as the Satellite Embedding dataset in Google Earth Engine. It says partners use it to map farm plots and agricultural change, including through cloud cover. source
- AgriEval authors (Yan et al.)BenchmarkNew AgriEval: large Chinese-language agricultural benchmark for LLMsAgriEval has 14,697 multiple-choice and 2,167 open-ended questions drawn from Chinese university agriculture exams and assignments, covering six categories and 29 subcategories. Across 51 models tested, most scored under 60% accuracy. source
- AgroBench authors (Shinoda, Inoue, Kataoka, Onishi, Ushiku)BenchmarkNew AgroBench: expert-annotated test of vision-language models on farm tasksAgroBench evaluates vision-language models across seven agricultural topics, with 203 crop and 682 disease categories labelled by expert agronomists. It found most open models perform close to random at weed identification. source
- Google (Google DeepMind and Google Partnerships Innovation)ProductNew AMED API gives field-level crop data for IndiaGoogle launched the Agricultural Monitoring and Event Detection (AMED) API, which reports crop type, season, field size and three years of field history for individual fields across India, refreshed about every two weeks. It builds on the earlier Agricultural Landscape Understanding (ALU) API. source
May 2025 · 1 development
- KissanAIModelNew Small open reasoning model for climate-smart farming advice in IndiaKissanAI released ThinkingDhenu1-CRSA-India-preview, a 4B model fine-tuned from Qwen3-4B on about 101,000 synthetic dialogues on climate-resilient and sustainable agriculture in India. It is labelled an experimental research preview. source
Apr 2025 · 1 development
- AgMMU authors (Gauba, Pi, Man, Pang, Adve, Wang)BenchmarkNew AgMMU benchmark built from real grower questions to extension expertsAgMMU tests vision-language models on 746 multiple-choice and 746 open-ended agricultural questions drawn from real exchanges between growers and US Cooperative Extension experts. The authors report that open models trail proprietary ones by a wide margin. source
Mar 2025 · 1 development
- Carbon RoboticsProductNew Retrofit tractor autonomy with remote human supervisionCarbon Autonomy is a retrofit kit that lets existing Deere tractors do ground preparation without a driver, with a remote Carbon operator able to take control in real time. source
Feb 2025 · 1 development
- John DeereProductNew Autonomy upgrade kit for tillage goes on sale for existing large tractorsDeere released its next-generation perception system as a precision upgrade kit that lets select 8R, 8RX, 9R and 9RX tractors do tillage without a driver in the cab. Select model year 2025 tractors are autonomy-ready from the factory. source
AI models built for agriculture, open and closed
| Model | Weights | What it is for | Availability and status | Published scores |
|---|---|---|---|---|
| Agricultural Understanding models (ALU and AMED APIs)Google (Google DeepMind, Google Partnerships Innovation, AnthroKrishi team) · Oct 2026 | Closed | Field-level landscape segmentation, crop monitoring and agricultural event detection for developers, governments and researchers building farmer services.Not stated · satellite imagery in; field boundaries, crop type, season and field events out | API in public preview (from Oct 2026); earlier research API access in India | None publishedsource |
| Large Plant Model (Carbon AI)Carbon Robotics · Feb 2026 | Closed | Detecting crops and weeds so LaserWeeder machines can laser-weed any field or crop, tuned per field with Plant Profiles.Not stated · images (camera data from LaserWeeder machines) | Built into Carbon Robotics LaserWeeder machines | None publishedsource |
| Sinong 1.0 (8B, 32B)Nanjing Agricultural University with Nanjing University of Science and Technology · Jan 2026 | Open weightsNot stated in the ModelScope metadata or README | General agricultural-domain language model for Chinese-language questions across animal science, veterinary medicine, agricultural economics, horticulture, plant protection and crop breeding.8B, 32B · text | open download (ModelScope) | None publishedsource |
| AgriLLM (Llama 3.3 70B fine-tune; Qwen3 30B-A3B fine-tune)ai71 (UAE) · Dec 2025 | Open weightsLlama 3.3 Community License (70B); Apache 2.0 (30B-A3B) | Agronomic question answering inside retrieval-augmented (RAG) systems for farmers, extension agents, researchers and policymakers.70B; 30B total with 3B active (mixture of experts) · text | open download (Hugging Face); training set agrillm-train-146k and evaluation set agrillm-qa-eval-800 also releasedMaker: not intended for medical or veterinary diagnosis, legally binding recommendations, or high-risk decisions without expert supervision. | AgriLLM QA eval, answer correctness (Llama-3.3-70B fine-tuned with RAG; GPT-4o with RAG scored 0.489 on the same table): 0.557AgriLLM QA eval, answer correctness (Qwen3-30B-A3B fine-tuned with RAG): 0.565source |
| ThinkingDhenu1-CRSA-India-previewKissanAI · May 2025 | Open weightsApache 2.0 (per model card) | Climate-resilient and sustainable agriculture advice for Indian conditions; decision-support micro-services and extension content.4B · text | open download (Hugging Face)Maker: out of scope for medical, legal or financial advice and for real-time critical decisions without human validation. | None publishedsource |
| E.L.Y. Crop ProtectionBayer (with Microsoft) · 13 Nov 2024 | Closed | Questions on crop protection product use, application and label compliance, adaptable to regional and crop needs.Small language model (size not stated) · text | Azure AI model catalog | None publishedsource |
| Dhenu2 India (1B, 3B, 8B)KissanAI · 24 Oct 2024 | Open weightsLlama 3.2 Community License (as stated on the 8B card) | Building advisory applications for Indian farmers on crop management, pest control and packages of practice.1B, 3B, 8B · text | open download (Hugging Face); hosted API at dhenu.ai | None publishedsource |
Who's building farm AI
Company trackers →AI-native agriculture companies
Halter
Sells solar-powered GPS collars and software that fence, move and monitor cattle virtually, with automatic heat and health alerts.
Cropin
Sells an AI platform that monitors crops and fields from satellite and field data for agribusinesses, food companies, lenders and governments.
Carbon Robotics
Builds the LaserWeeder, a tractor-pulled machine that uses cameras and AI to find weeds and kill them with lasers, plus tractor autonomy kits.
Solinftec
Runs an AI farm-operations platform (ALICE AI) and Solix, a solar-powered field robot that scouts crops and spot-sprays weeds.
CropX
Combines soil sensors, satellite and machinery data in software that advises farms on irrigation, nutrients and crop health.
Aerobotics
Uses AI on smartphone images to measure fruit size, colour and quality so fruit growers, packers and marketers can forecast yields.
Aquabyte
Puts AI camera systems in salmon pens to estimate fish weight, lice and welfare and to guide feeding.
- Over 800 systems in pens (14 Nov 2026) source
Plantix (PEAT GmbH)
Runs Plantix, a free phone app that diagnoses sick crops from a photo and suggests treatments, with a community of agri-experts.
- More than 150 million crop-related questions (14 Nov 2026) source
Dryad Networks
Makes solar-powered, AI-enabled gas sensors and a mesh network that detect forest fires within minutes of ignition, often before open flames appear.
Ecorobotix
Builds the ARA, a camera-guided sprayer that uses AI plant detection to spray single weeds or crop plants in 6 x 6 cm patches.
Taranis
Flies drones over fields and uses AI on leaf-level images to find weeds, insects, disease and nutrient problems for farm advisers and retailers.
Pano AI
Sells AI-powered wildfire detection that spots fires early and sends alerts, images and real-time data to utilities, fire professionals and landowners.
What the big AI companies are doing in agriculture
Who's using it
Which vendor, for which use
| Vendor | Crop and field monitoring | Precision farming | Livestock | Forestry and fisheries | Advice for farmers | Other | All |
|---|---|---|---|---|---|---|---|
| Aerobotics | |||||||
| Ecorobotix | |||||||
| Wadhwani AI | |||||||
| Aquabyte | |||||||
| OpenAI | |||||||
| Pano AI | |||||||
| Plivo | |||||||
| Taranis | |||||||
| Built in-house |
The organisations
| Organisation | Use and vendor | Stage and scale | What |
|---|---|---|---|
| STOV DruzhbaUA · 6 Oct 2026 | Precision farmingEcorobotix | In useTwo years later, ARA is used across his operation. | A 2,700-hectare arable farm in Cherkasy Oblast, Ukraine, tested Ecorobotix's ARA precision sprayer on 300 hectares of corn and soybeans in 2024 and now uses it across the farm.Reported by the organisation: 40-90% savings on herbicides and fertilizers and a 1-3% corn yield increase in the test; the farmer estimates an average 70% cut in plant protection products.source · announced by the vendor |
| Agriculture and Agri-Food CanadaCanada · 21 Aug 2026 | Crop and field monitoringin-house (AAFC Earth Observation Team) | In usein support of a national crop inventory | Canada's agriculture department maps crop types across the country every year from optical and radar satellite images, using a decision-tree classifier.source · announced by the government |
| ETA La Vallée des UssesFR · 14 Aug 2026 | Precision farmingEcorobotix | In usetreating 1,000 hectares with the ARA | An agricultural contractor in Haute-Savoie uses Ecorobotix's camera-guided ARA sprayer to spot-spray dock weeds (Rumex) in grassland for its farm clients.Reported by the organisation: Chemical use cut by up to 80-90%, and 10-30% more usable forage yield for clients (the contractor's own estimate).source · announced by the vendor |
| Digital GreenIndia · 10 Feb 2026 | Advice for farmersOpenAI | In useover 1 million users across India, Kenya, Ethiopia, Nigeria, and Brazil | Digital Green, a non-profit, runs FarmerChat, an AI assistant built on OpenAI models that gives smallholder farmers real-time, localised advice. It operates in India, Kenya, Ethiopia, Nigeria and Brazil.Reported by the organisation: A 2025 study by 60 Decibels found 70% of farmers applied FarmerChat's recommendations within 30 days.source · announced by the organisation |
| Apex Ag SolutionsUS · 28 Jan 2026 | Crop and field monitoringTaranis | Rolling out | An Arkansas farm-input retailer added Taranis drone scouting to its grower services. AI reads high-resolution drone images six times a season to find weeds, insects, disease and nutrient problems.source · announced by the vendor |
| Kisan Call Centre, Ministry of Agriculture and Farmers Welfare (India)India · Aug 2024 | Advice for farmersWadhwani AI (built on OpenAI GPT-4o-mini) | In useresolved over 279,000 queries since August 2024 | Krishi Saathi is a chatbot inside the Kisan Call Centre portal. It gives call-centre farm advisers weather updates from the India Meteorological Department, summarised by a large language model in 11 Indian languages.source · announced by the vendor |
| Ministry of Agriculture and Farmers Welfare (India)India · Nov 2023 | OtherWadhwani AI | In useproviding crucial insights to 16 divisions at the Ministry of Agriculture and Farmers’ Welfare | Krishi 24/7 scans agricultural news in 11 regional languages and flags significant agricultural events on a dashboard; Wadhwani AI says it serves 16 divisions of India's agriculture ministry.source · announced by the vendor |
| USDA Forest ServiceUS · 1 Mar 2021 | Forestry and fisheriesin-house (USDA Forest Service) | In use | The Forest Service's Landscape Change Monitoring System uses a model to map vegetation gain and loss, land cover and land use over time across large areas.source · announced by the government |
| USDA National Agricultural Statistics ServiceUS · 1 Jan 2008 | Crop and field monitoringin-house (USDA NASS, on Google Earth Engine) | In usethe entire Continental United States | USDA's statistics agency maps field crops across the continental US each year from satellite images with a machine-learning classifier. The map supports its crop acreage estimates.source · announced by the government |
| Amul (Gujarat Cooperative Milk Marketing Federation)India | LivestockPlivo (voice platform) | Rolling outVoice AI for 3.6 million dairy farmers | Amul launched Sarlaben, a voice assistant that dairy farmers can phone for advice on cattle health, feeding and subsidies in Gujarati. It needs no smartphone or app.source · announced by the vendor |
| Fundo El ParaísoPE | Crop and field monitoringAerobotics | In useempowering smarter decisions across 240 hectares | A family-owned citrus estate in Peru replaced manual fruit sampling with Aerobotics' AI fruit-sizing app, which measures fruit from phone images.Reported by the organisation: Over 300 fruit samples per plot instead of 12, at 10 seconds per plant (page headline: '25x More Samples, 90% Less Time').source · announced by the vendor |
| Southern Cross FarmsAustralia | Crop and field monitoringAerobotics | In usemanaging nearly 700 hectares of citrus in Australia | An Australian citrus grower uses Aerobotics' TrueFruit phone app to size fruit across all its blocks instead of hand-sampling with calipers.Reported by the organisation: Fruit size data across 100% of citrus blocks, up from under 10%, with the same time and labour.source · announced by the vendor |
| Washington State Department of Natural ResourcesUS | Forestry and fisheriesPano AI | In use | Washington's state lands agency uses Pano AI's camera-based wildfire detection. Its former Commissioner of Public Lands credits it, among other tools, with keeping most fires small.Reported by the organisation: 95% of fires kept below 10 acres over three years (credited partly to Pano AI).source · announced by the vendor |
| BlumarCL | Forestry and fisheriesAquabyte | In use | Salmon farmer Blumar uses Aquabyte's monitoring and data-analysis technology at its fish-farming sites. Aquabyte trained Blumar staff in the first level of a new operator certification launched in Chile.source · announced by the vendor |
AI in agriculture by country
AI by Country →Leading countries
| Country | Named deployments | Rules on this page |
|---|---|---|
| United States of America | 4 | 1 |
| India | 4 | 0 |
| France | 1 | 4 |
| Canada | 1 | 0 |
| Peru | 1 | 0 |
| Australia | 1 | 0 |
| Ukraine | 1 | 0 |
| Chile | 1 | 0 |
Official registers you can search
2024 Federal Agency AI Use Case Inventory (consolidated)
Every publicly reported AI use case at 41 US federal agencies, with bureau, purpose, outputs, stage of development, dates and in-house or contracted development. The Department of Agriculture has 89 rows (Forest Service, NASS, APHIS, NRCS, ARS, FSA, RMA and others). About 2,133 entries.
What works: early studies, no rating yet
All rated use cases →Studies of AI in agriculture
- Phone advisories for 28,796 coffee farmers, written and voiced with AI, were listened to slightly less (2 points) and recalled less than advisories written by agronomists.
Precision Development (2026) · grade C, randomised trial in live work · task: Farm advice with AI
1 more study found but not counted
- Aouad et al. (2026), grade C: Only the abstract could be read: SSRN blocks automated access, and no other copy was found, so attrition and how outcomes were measured could not be checked. It will be counted once the full paper can be read.
Not enough yet for a rating. These studies are a start, but our fixed rules need more, or stronger, studies of the same task before they rate it. Until then, the products and deployments above are what organisations say they are doing, not findings that it works.
Rated evidence for tasks done in every sector:
AI for drafting work documents may help, but the evidence is not yet strong enough for a firmer verdict.
3 studies counted →Research and analysisDoes AI help people analyse data more accurately?Mixed results: studies of AI for data analysis disagree on whether it helps.
7 studies counted →Creative and designDoes AI make creative and design work better?Not enough evidence yet to say whether AI makes creative and design work better.
4 studies counted →TranslationDoes AI help professional translators?Not enough evidence yet to say whether AI helps professional translators.
4 studies counted →HR and recruitingDo AI-written job posts and applications lead to more hires?Not enough evidence yet to say whether AI-written job posts and applications lead to more hires.
3 studies counted →Scientific researchDoes AI speed up systematic reviews without more errors?Not enough evidence yet to say whether AI speeds up systematic reviews without more errors.
2 studies counted →Rules to check first
Under the EU AI Act, AI that is a safety component of a tractor or forestry vehicle (Regulation 167/2013) or, since the 2026 Digital Omnibus, of machinery (Regulation 2023/1230) is high-risk under Article 6(1). Both laws sit in Section B of Annex I, so only Article 6(1), Article 60a and Articles 102 to 112 apply directly; the AI requirements reach makers through rules made under those sector laws. European Union →
Annex III of the AI Act has no point for farming, forestry or fishing as such. Point 2 (critical infrastructure) can apply where AI is a safety component in water supply, such as a water or irrigation utility; Annex III duties apply from 2 December 2027. European Union →
The EU Machinery Regulation (EU) 2023/1230 applies from 20 January 2027 and adds rules for machinery whose safety functions are AI-powered. Farm robots, smart sprayers and other machinery placed on the EU market from that date must meet it; machinery placed on the market before then follows the Machinery Directive 2006/42/EC. European Union →
In the US, spraying or spreading crop inputs from a drone is an agricultural aircraft operation under FAA rules (14 CFR Part 137) and needs a Part 137 certificate and the matching exemption, whatever software guides the drone. United States →
Japan's agriculture ministry sets a safety guideline for autonomous farm machinery (revised March 2026). Makers, sellers, adopters and users must carry out risk assessment, follow the conditions of use and keep the machine supervised, on site or remotely. Japan →
Before you start
Will AI help with your work? →- Start with one taskPick one task with a clear owner and a measure you already track, not a company-wide launch.
- Measure against a baselineRecord time, quality and errors before AI and after, for users and a comparison group.
- Keep a person checkingSomeone accountable reviews what the AI produces before it reaches a customer or a decision.
- Check the rules and dataConfirm what data may go into the tool, the vendor's terms, and the rules for your sector below.
- Train the people using itShort, task-specific training; free courses are in our learning directory.
Learn for free
All free courses →Questions people ask
How many farms and agricultural organisations use AI?
6.0% of US agriculture, forestry and fishing businesses said they used AI in the latest US Census Bureau survey, against 24.6% of all businesses. Surveys count organisations, not individual workers.
What do farms and agricultural organisations use AI for?
Among the 14 named deployments on this page, the most common uses are crop and field monitoring (5), forestry and fisheries (3), advice for farmers (2). That is what organisations have announced, not market share. See who →
Are there AI models built for agriculture?
Yes. This page lists 7, of which 4 have open weights anyone can download. Compare them →
Does AI work in agriculture?
We have found 1 study of AI in agriculture, not yet enough for our rules to give a rating. See what they found → Evidence from other sectors for tasks such as writing, customer service and coding is on What works.
Which rules apply?
Start with eu: farm machinery ai: Under the EU AI Act, AI that is a safety component of a tractor or forestry vehicle (Regulation 167/2013) or, since the 2026 Digital Omnibus, of machinery (Regulation 2023/1230) is high-risk under Article 6(1). Both laws sit in Section B of Annex I, so only Article 6(1), Article 60a and Articles 102 to 112 apply directly; the AI requirements reach makers through rules made under those sector laws. All the rules to check →
Follow AI in agriculture
How this page is made
This guide covers ISIC Rev. 5 section A (all 14 sector guides). Developments, models, deployments, rules and company figures come from each organisation’s own publication, a regulator or a statistics office, and each was checked against its source by a second, independent pass; items that failed were removed. Adoption figures are the US Census Bureau’s and Eurostat’s own estimates. Use cases are rated from published studies by fixed rules; a product listed here is not a finding that it works. Editorial owner: AI Adoption editors. Disclosure: AI Adoption uses AI tools, including Claude from Anthropic, in its research and editorial work, and some companies on this page make them. No company has any say in this page. Spot a mistake? Corrections.