Industries / Food, farming and forestry / Agriculture and farming
Livestock
Raising animals for meat, milk, eggs and fibre: choosing which animals breed, then feeding, housing, milking and keeping them healthy up to the farm gate.
7
Milestones achieved, from c. 9000 BCE to 2011.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
In depth
The same ground is charted in more depth, as industries of their own.
The road
AchievedWatch forOpen
Achieved
c. 9000 BCE
Sheep, goats and cattle domesticated
People in the Near East began keeping and breeding wild sheep, goats and cattle instead of hunting them. Archaeozoologist Melinda Zeder's 2008 review dates sheep and goats to about 11,000–10,500 years ago, and cattle to about 11,000–10,000 years ago in the upper Euphrates valley.
Credit
early farmers of the Fertile Crescent; dating summarised by Melinda Zeder (PNAS, 2008)
1949
Frozen semen with glycerol
Christopher Polge, Audrey Smith and Alan Parkes found that glycerol keeps sperm alive through freezing, so a chosen bull's semen could be stored and shipped for artificial insemination (placing semen in a cow by hand instead of using a bull). By 1965 almost all cattle semen used in the US was frozen.
LessonThe protective effect of glycerol "was discovered quite by chance" (Lonergan, 2018).
Credit
Polge, Smith and Parkes (Nature, 1949); a Danish cooperative had already bred 1,070 cows by insemination in 1936
1966
Sterile flies end screwworm
The screwworm is a fly whose larvae feed in wounds on living animals. USDA scientists bred male flies by the million, sterilised them and released them, so wild females that mated with them laid no fertile eggs; this cleared the fly from the US by 1966, and a barrier of sterile flies was later held at Panama.
LessonA barrier has to be held: USDA confirmed the fly in a calf in Texas on 3 June 2026, and DTN, citing USDA, counted 71 US cases in Texas, New Mexico and Arizona by 30 October 2026.
Credit
Edward F. Knipling and Raymond C. Bushland, USDA
1986
Mad cow disease in Britain
BSE (bovine spongiform encephalopathy, a fatal brain disease of cattle) was first reported in UK cattle in November 1986 and traced to meat and bone meal, a feed made by rendering carcasses, fed back to cattle. By 2015 nearly 191,000 BSE cases had been found worldwide, and 176 people in the UK had had the related human disease, variant CJD, all of whom died.
LessonFeeding animal remains back to cattle spread the disease, and the UK's partial feed ban was not made absolute and fully enforced until 2001.
Credit
a failure; account by Dermot McNerney, NSW Department of Primary Industries
1992
Robots begin milking cows
The first commercial milking robots went onto Dutch dairy farms in 1992; with a robot the cow walks in to be milked when she chooses, so no one has to be there at fixed milking times. More than 10,000 farms used them by the end of 2010, and a 2023 review gives an estimated 50,000 units on 25,000 farms in 2019.
LessonThe farmer's job shifts from milking to managing; "Manual labour is partly replaced by management and control" (Kees de Koning, Wageningen).
Credit
Dutch makers, including Lely, whose Astronaut prototype milked its first cow in 1992
2009
DNA tests pick breeding bulls
In January 2009 the USDA made genomic evaluations official for dairy cattle, predicting a young bull's breeding value from 38,416 DNA markers instead of waiting years for his daughters' milk records. A 2016 study of US Holsteins found the age of the fathers of new breeding bulls fell from about 7 years to under 2.5 years, and yearly genetic gain in milk yield traits rose by about 50–100%.
Credit
USDA Animal Improvement Programs Laboratory (Paul VanRaden, George Wiggans and colleagues); measured by García-Ruiz and colleagues (PNAS, 2016)
2011
Rinderpest wiped out worldwide
Rinderpest, a viral plague of cattle, was declared eradicated by the World Organisation for Animal Health (WOAH) in May 2011 and by FAO in June 2011. WOAH calls it only the second infectious disease, after smallpox, to be wiped out worldwide.
Credit
FAO's Global Rinderpest Eradication Programme (from 1994) with WOAH; vaccine developed in 1960 by Walter Plowright and published with R. D. Ferris in 1962 (dates from Roeder, Mariner and Kock, 2013)
Watch for
Methane additive under review
Bovaer (3-NOP) is a feed additive that reduces the methane cows burp out; Denmark required conventional dairy farms to feed it from 2025, and after farmers reported sick cows the government let affected farms opt out for a time. The maker says "there remains absolutely no evidence that Bovaer is a cause of any problems", while an Aarhus University analysis of 73 herds found no clear overall effect on production, disease or deaths but serious problems in some herds.
Achieved whenan independent controlled trial or a regulator's review explains the problem herds and confirms both the methane cut and the cows' safety on ordinary farms; the best-known methane figure, a 32.7% cut in a 2023 analysis, comes from a study the maker funded.
Claimed by
DSM-Firmenich, January 2026
Gene-edited pigs resist PRRS
PRRS (porcine reproductive and respiratory syndrome) is a viral disease of pigs. Genus PIC removed part of the gene the virus uses to enter pig cells and says "Without that entry point, the PRRS virus cannot infect the pig"; the US FDA approved the edit in April 2025 and Canada in January 2026, but Genus will not sell the pigs until key export markets, particularly Mexico and Japan, also approve.
Achieved whenthe pigs are sold to commercial farms and independent herd records show fewer PRRS outbreaks.
Claimed by
Genus PIC, January 2026
Texas sterile-fly factory
USDA is building a plant at Moore Air Base, Texas, to breed sterile screwworm flies, with a projected capacity of 300 million flies a week, three times the roughly 100 million a week from the plant in Panama.
Achieved whenthe Texas plant runs at its stated output and USDA reports no new US screwworm cases for a full fly season.
Claimed by
USDA Animal and Plant Health Inspection Service, October 2026
Open
01
Open lameness video benchmark
A public, openly licensed set of videos of dairy cows walking out of the milking parlour, from several farms and camera positions, with each cow scored for lameness (limping) on a standard mobility scale by at least two trained scorers. Add a baseline model and a leaderboard so camera systems can be compared on the same test.
The briefCredit · unclaimed02
Calf cough counter
Microphones over calf pens and a model that counts coughs per pen per hour, with a public labelled audio set and an on-farm test of whether cough counts rise before farmers or vets diagnose bovine respiratory disease (pneumonia in young cattle).
The briefCredit · unclaimed03
Spoken herd records for smallholders
A phone app that works offline. A smallholder says, in Swahili, "Neema gave six litres this morning" or "Zawadi was served today", and the app writes a structured record (animal, event, quantity, date) and reads it back for confirmation.
The briefCredit · unclaimed
Open 01 · the brief
Open lameness video benchmark
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, openly licensed set of videos of dairy cows walking out of the milking parlour, from several farms and camera positions, with each cow scored for lameness (limping) on a standard mobility scale by at least two trained scorers. Add a baseline model and a leaderboard so camera systems can be compared on the same test.
Why it is still open
A 2025 paper says "Few lameness video datasets are openly released, which slows reproducibility and benchmarking." The sets that exist are small or private: one Dutch set of 5,164 clips is described by its authors as "a proprietary dataset", and the main published check of a commercial lameness camera (CattleEye) was funded by the company.
Built with what exists
Low-cost IP security cameras over the parlour exit; open detection and pose models such as Ultralytics YOLO or DeepLabCut; Label Studio for annotation; hosting on Zenodo or Hugging Face Datasets under a CC-BY licence.
First step
Get one dairy farm's permission to mount a camera over the parlour exit for two weeks, and arrange for a trained scorer to score the same cows on the same days.
Done looks like
At least 500 videos from at least five farms, each scored by at least two trained scorers, with scorer-to-scorer agreement reported, and a baseline model tested on a farm it never saw, its agreement with the scorers published next to the scorers' agreement with each other.
Open 02 · the brief
Calf cough counter
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Microphones over calf pens and a model that counts coughs per pen per hour, with a public labelled audio set and an on-farm test of whether cough counts rise before farmers or vets diagnose bovine respiratory disease (pneumonia in young cattle).
Why it is still open
The newest multi-farm study, from La Salle–Universitat Ramon Llull in 2026, detected cattle sounds including coughs with an F1-score of only 57.4% (a measure that combines how many events are caught with how many alerts are false), and its authors say a sellable sensor must run on cheap hardware without a constant internet link. The known commercial cough monitor, SoundTalks, is for pigs.
Built with what exists
An omnidirectional microphone with a field recorder, or USB microphones on a Raspberry Pi 5; open audio models such as YAMNet or PANNs; Audacity or Label Studio for labelling; a low-power AI chip design such as Google's Coral NPU to run the detector in the barn (Notes: "An open compiler").
First step
With permission, hang one recorder in a calf barn for a week and label one hour of audio to measure how long labelling takes.
Done looks like
A public set of at least 100 hours of calf-barn audio with at least 2,000 expert-labelled coughs from at least three farms; precision and recall of at least 80% each on a farm not used in training; and a three-month log comparing cough alerts with the calves the farmer actually treated.
Open 03 · the brief
Spoken herd records for smallholders
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A phone app that works offline. A smallholder says, in Swahili, "Neema gave six litres this morning" or "Zawadi was served today", and the app writes a structured record (animal, event, quantity, date) and reads it back for confirmation.
Why it is still open
In DigiCow, a leading Kenyan dairy app, voice is used to send lessons out, not to take records in: "The lessons are sent via voice notes, SMS or through the app itself", while records are kept with in-app tools. Earlier ILRI phone record systems relied on typed entry. Voice entry exists in English-language herd apps, but no Swahili spoken-record app with a published accuracy test was found.
Built with what exists
ElevenLabs Scribe v3 for speech-to-text (Notes: "Fewer word errors"); for offline use, OpenAI's open Whisper models running on the phone through ggml (Notes: "A first stable number for local AI"); a small Liquid AI LFM3 model, fine-tuned on one graphics chip, to turn the words into a record (Notes: "One GPU to fine-tune"); Mozilla Common Voice Swahili recordings for testing; any low-cost Android phone.
First step
Write 200 typical record sentences with three dairy farmers, record them in Swahili, and run them through Scribe v3 to get a first word error rate.
Done looks like
On a public test set of 1,000 spoken records from at least 50 farmers, the app gets the animal, event and quantity right in at least 95% of records while offline, with word error rate reported, and a four-week trial with 20 farmers compares spoken records with their paper notebooks.