BRENNEN

Industries / Food, farming and forestry / Agriculture and farming

AgTech

AgTech is the business of selling technology to farms: seed with added or edited genes, sensors and satellite images, software for records and advice, and machines that steer, spray or weed with little or no driver.

1977

Priva, a Dutch company founded in 1959 to import greenhouse heaters, introduced its first climate computer for horticulture; a Rotterdam School of Management case study calls it "the first computer in horticulture that controlled climate, energy and water". ).

priva.comrepub.eur.nl

33

Milestones achieved across 6 domains.

8

To watch for: announced, not yet confirmed outside.

12

Open deliverables. Nobody has shipped them.

In one minute

Where it started

Seed came first: from 1926 US companies sold hybrid corn seed that farmers bought new each year instead of saving their own. Electronics followed: in 1977 Priva sold Dutch greenhouse growers a computer that ran heating, energy and water, and from 1987 Japanese farmers sprayed rice with pilotless helicopters. In the 1990s soil maps, yield monitors and satellite positioning let machines vary seed and fertilizer across a field, and genetically engineered seed went on sale in the US in 1996.

Where it stands

In 2023, 70% of large US crop farms and 52% of midsize ones used guidance that steers their machines, by the count of the US Department of Agriculture (USDA), but small farms used such tools least, and only 13% of irrigating farms used a soil moisture sensor to decide when to water. The UN Food and Agriculture Organization (FAO) judged in 2022 that the business case for digital automation on farms "is not yet mature", above all in low- and middle-income countries, where a smartphone is the tool most farmers can reach. Investors put $16.2 billion into agriculture and food start-ups in 2025, $9 billion of it into farms and food production, by AgFunder's count, and well-funded efforts have also failed: Calyxt wound down its edited soybean oil, Small Robot Company was liquidated, and Corteva broke up Granular, farm software DuPont had bought for a reported $300 million.

What is open

Most open jobs are public tests of claims that sellers already make: for cheap soil probes, satellite crop alerts, AI greenhouse control, advertised seed yields, and whether a farmer's records can leave one software platform for another. Others are records nobody keeps in one place, such as dicamba drift complaints by state, the limits printed on spray labels, and which countries regulate each gene-edited crop. Robot arms that learn from a few dozen demonstrations and AI agents that read long documents and browse the web put most of them within reach of a small team.

Domains

AchievedWatch forOpen

Advances to build with

2 research notes

Open in agtech

3 of 12 open deliverables

Open · Precise planting and spraying

A national dicamba complaint table

One public table of off-target dicamba complaints by state and year, 2017 to 2027, for every state where dicamba-tolerant soybeans or cotton are grown. Each number links to the state's own report or its reply to a public-records request, with a note on how that state counts a complaint.

Built with what exists

State agriculture department websites and public-records requests; a web-browsing AI agent to find each state's complaint pages and draft the requests (see the Note "The jobs agents repeat"; the Note "Deleting 80% to start over" covers browser agents); a spreadsheet or SQLite file; USDA NASS planted acres by state for context.

Credit · unclaimed

Open · Sensors and imagery

Score satellite alerts against scouting

A one-season test of the crop-stress alerts that satellite field-monitoring apps send to farmers. For every alert, a scout visits the flagged spot within three days and records whether there is a real problem and what caused it; scouts also walk every field on a fixed schedule, to catch problems the alerts missed.

Built with what exists

Free Sentinel-2 images through Copernicus Browser or Google Earth Engine; the free tiers of commercial field-monitoring apps; a phone that tags photos with their location; an open vision-language model, run locally, to sort scouting photos by cause (see the Note "A first stable number for local AI").

Credit · unclaimed

Open · Farm robots

Pull weeds by demonstration

A low-cost robot arm that learns, from a few dozen remote-controlled demonstrations, to pull weeds by the root from beds or trays of vegetable seedlings without touching the crop, with every attempt filmed and counted.

Built with what exists

The SO-101 open-design robot arm; LeRobot to record demonstrations and train (see the Note "No download"); an open robot model such as NVIDIA GR00T 2 (see the Note "Five demonstrations per task") or Runway Praxis-1 (see the Note "Open robot model weights"); trays of lettuce or onion seedlings with weeds sown among them.

Credit · unclaimed

Where it sits