AI in agriculture: what the evidence shows.

01The one-paragraph version.

AI is real and useful on farms. Camera-guided spot spraying has the strongest evidence, from use on millions of acres and from university field results. Much of the rest is promising and thinly tested. Surveys in 2026 found about one farmer in six using generative AI for farm tasks, but about one in seventeen naming it as a trusted source for decisions, and about half of a panel of larger U.S. producers said AI and data-driven tools gave them no meaningful benefit yet. For AI about a particular farm, the limit is usually the records: whether the operation has trustworthy, owned, well-labeled readings of its own conditions. A farm with several seasons of clean, labeled records has what farm-specific AI tools will need from it. A farm without them can use AI as a general reference, which is often unreliable on local specifics.

02Three questions, three pages.

• What the numbers show. How many farmers use AI, how few name it as a trusted source, how many see no benefit yet, who has the data layer, and the first rigorous trial. Every figure dated.
• Proven, promising, promised. Ten applications of AI in agriculture, graded by the strength of the evidence, with the data each one stands on. Re-graded every October.
• Reading the claims. Five patterns worth recognizing, a ladder of evidence from weakest to strongest, and ten questions to ask before you believe a number.

03The thread through all of it: data.

The same finding turns up on each of the three pages, in two forms. The applications with the strongest evidence sit on large data systems their makers built: millions of labeled images, satellite positioning. And AI that answers questions about one farm needs that farm's own records, USDA's figures show the tools that produce such records (yield and soil maps) on 68% of large crop farms and 13% of small ones.

04When platforms disappear.

One case is worth telling in full, because the company's own notice documents it. Microsoft retired its FarmBeats agriculture platform on September 30, 2023, and moved customers to a successor, Azure Data Manager for Agriculture. That successor was a preview, and it was retired too, on September 1, 2025. Microsoft's own notice told customers it no longer received security updates, that Microsoft kept no copies of their data, and that it would delete the data 30 days after retirement.

What the case shows:

• Large providers change course. Microsoft retired two agriculture platforms about two years apart.
• Exported records survived. Anyone who kept exported records could carry them to the next tool. Anyone whose only copy lived inside the platform had 30 days.
• A preview, pilot, or beta is a product its maker has not committed to yet. Fine for an experiment; risky as the only home of several seasons of farm records.

The rule that follows: before any AI or data platform touches your records, know how you get your data out, in a usable form, without asking anyone's permission, and then do it on a schedule. The open agriculture world has its own cautionary tales; we keep them on the open-source shelf. The principle is in the king belongs to the farmer.

05The OAT position in seven lines.

These are the choices this site makes. They are ours; other careful growers choose differently.

1. Data is king, and the king belongs to the farmer. See Data Is King.
2. Measure first. No data, no farm-specific AI.
3. Open formats, local first. So any AI, now or later, can read your records, and no vendor can take them away.
4. Rules act; AI advises. Irrigation, heat, and alarms stay on rules you can inspect. AI informs, and people decide. See AI in Home Assistant.
5. Equip the trusted person. AI should make your agronomist, your staff, and you better, not replace judgment.
6. Appropriate technology. Collect what answers a real question at the accuracy that question needs. See Appropriate Technology.
7. Weigh the evidence first. Grade claims honestly, including ours.

06How this fits the rest of the site.

To use AI well (chatbots, prompts, coding assistants, agents, local models), start with Lesson 8, Understanding AI. For why the data matters most, read Data Is King. To keep AI safely away from the switches, read AI in Home Assistant. This section is the evidence: what is true, what is claimed, and what that means for a working grower. We use AI to research and write this site, and we hold ourselves to the same standard (how this was written).