Industries / Food, farming and forestry / Agriculture and farming
Weeds, pests and disease
Keeping weeds, insects and plant diseases from destroying crops, with resistant varieties, chemicals, natural enemies, machines and careful counting.
7
Milestones achieved, from 1845 to 2026.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1845
Blight starves Ireland
Late blight, caused by the water mould Phytophthora infestans, destroyed Irish potato crops from 1845 to 1849, and about one million people died of hunger or disease.
LessonIreland grew mostly one potato type, the Lumper, so one disease could take most of the food supply.
Credit
a failure; a 2013 genetic study of old herbarium samples named the famine strain HERB-1 and traced it to the Americas
1959
Spray only when needed
A 1959 paper in the journal Hilgardia set out "integrated control": combine natural enemies with selective insecticides, and spray only when pest counts pass an "economic threshold", the level at which damage would cost more than treatment. It became the basis of integrated pest management (IPM).
Credit
Vernon Stern, Ray Smith, Robert van den Bosch and Kenneth Hagen (University of California)
1972
US cancels DDT
DDT, the first modern synthetic insecticide, came into use in the 1940s, and Rachel Carson's Silent Spring (1962) raised wide public concern about pesticide misuse. In 1972 the US Environmental Protection Agency (EPA) cancelled DDT because of its effects on the environment.
LessonA chemical that lasts long enough to keep working in the field can also build up in wildlife.
Credit
Paul Müller of J.R. Geigy (Nobel Prize, 1948, for finding DDT's power against insects); Rachel Carson; US EPA
1998
Weeds outlast glyphosate
Monsanto chemist John Franz developed the weedkiller glyphosate in 1970 and Monsanto sold it as Roundup from 1974; in 1998 Australian scientists reported that, after 15 years of successful use, it failed to control a population of rigid ryegrass. The International Herbicide-Resistant Weed Database now lists 550 resistance cases in 276 weed species, in 76 countries, against 21 of the 31 known ways herbicides work.
LessonRelying on one weedkiller year after year selects for weeds that survive it.
Credit
John Franz (Monsanto) found glyphosate; Stephen Powles and colleagues reported the resistance
2008
Bollworms beat Bt cotton
Bt cotton makes a protein from the soil bacterium Bacillus thuringiensis that kills certain caterpillars; India approved it in 2002 and by 2009 it covered 87% of India's cotton. By 2008 pink bollworm in Gujarat's Amreli district had evolved resistance to its single protein, Cry1Ac, which the 2011 paper reporting it calls the first evidence of field-evolved resistance in this pest.
LessonA 2016 study blames too few "refuges", strips of non-Bt cotton that keep insects without resistance in the population.
Credit
S. Dhurua and G.T. Gujar (Indian Agricultural Research Institute)
2025
Cameras aim the spray
John Deere's See & Spray uses cameras on the sprayer boom to spray only where it sees weeds. A three-year University of Arkansas trial in soybeans, begun in 2022, measured 43–59% less herbicide than spraying the whole field.
LessonAt a low sensitivity setting, Palmer amaranth (a major weed) increased by 280% a year, so the savings come with a risk of missed weeds.
Credit
John Deere with Blue River Technology; measured by Jason Norsworthy, University of Arkansas
2026
China's crop-drone fleet passes 300,000
China's Ministry of Agriculture and Rural Affairs reports more than 300,000 agricultural drones, the most of any country, working over more than 30 million hectares a year.
Credit
China's farmers and drone service teams; figures from the Ministry of Agriculture and Rural Affairs, reported by China News Service
Watch for
Half the spray, fleet-wide
Deere says customers used See & Spray on more than 5 million acres in 2025 and cut "non-residual" herbicide (weedkiller that does not keep working in the soil) by nearly 50% on average, saving nearly 31 million gallons of spray mix.
Achieved whenan independent audit of fleet data, or university trials in several states, reports a similar average; Arkansas measured 43–59% in one state's soybeans only.
Claimed by
John Deere, November 2025
New fields in minutes
Carbon Robotics says its "Large Plant Model" was trained on 150 million labelled plants and lets an operator set up its laser weeder on a new field or crop within minutes by choosing two or three images on an iPad.
Achieved whenan independent trial on a crop the system has not seen reports weed kill and crop damage after this quick setup.
Claimed by
Carbon Robotics, February 2026
Open
01
Disease apps against lab results
A public set of field photos of sick crops, each with a diagnosis confirmed by a plant clinic or laboratory, and a scored head-to-head of free phone apps and open AI models on those photos.
The briefCredit · unclaimed02
Robot weeds nursery pots
A robot arm at a potting bench that finds weed seedlings in container-grown plants and pulls them out, roots included, without disturbing the crop plant.
The briefCredit · unclaimed
Open 01 · the brief
Disease apps against lab results
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of field photos of sick crops, each with a diagnosis confirmed by a plant clinic or laboratory, and a scored head-to-head of free phone apps and open AI models on those photos.
Why it is still open
A 2025 review in Plant Methods found that crop-disease models reach 95–99% accuracy in the lab but 70–85% in the field, and lists working across regions as a research need. The largest recent comparison, a September 2026 preprint on about 1.16 million farmer photos from Ethiopia, India, Kenya and Nigeria, does not say in its abstract that its diagnoses were confirmed by a laboratory.
Built with what exists
The free phone apps Plantix and PlantVillage Nuru; the open PlantDoc image set for comparison; open vision-language models run locally with ggml (Notes: "A first stable number for local AI"); a university plant diagnostic lab or plant clinic to confirm each case.
First step
Ask one university plant diagnostic lab to let you photograph incoming samples this month, with a fixed photo method, before the lab gives its diagnosis.
Done looks like
300 photos across three crops from at least two regions, each with a confirmed diagnosis, and each app's accuracy published with confidence intervals and its "no answer" rate, with the photos released under an open licence.
Open 02 · the brief
Robot weeds nursery pots
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A robot arm at a potting bench that finds weed seedlings in container-grown plants and pulls them out, roots included, without disturbing the crop plant.
Why it is still open
Alabama Extension says hand weeding is the only way to remove most weeds after they sprout, that even the best-run container nurseries still need it, and that it costs $500–$4,000 per acre in labour. Searches found weeding robots for fields, pastures, tree-nursery beds and gardens, but none that pulls weeds from pots.
Built with what exists
An SO-101 arm with LeRobot (Notes: "No download"); a control model trained from demonstrations with NVIDIA GR00T 2 (Notes: "Five demonstrations per task") or Runway Praxis-1 (Notes: "Open robot model weights"); an overhead camera; potted nursery plants with naturally sprouted weeds.
First step
Pot 20 small plants in ordinary field soil so weeds sprout, and record 50 operator-guided pulls.
Done looks like
20 filmed pots, with a published tally of weeds pulled with roots, weeds snapped off, crop plants disturbed, and seconds per pot compared with a person.