Bushel Finds Farm AI Living in the Office

Fourteen percent of more than 1,400 U.S. and Canadian farmers told Bushel they use AI on the farm. Most of that use is writing, analysis, and other desk work, not yield models. ChatGPT led the tools they named, then Gemini and Copilot.

Farmers are using AI, in other words, the way a grain office already uses email. The agronomy pitch that usually travels with “farm of the future” still sits at the bottom of the list.

Fourteen Percent Is an Office Statistic

Bushel, an independent software firm in Fargo, N.D., released the 2026 State of the Farm report on April 2 after surveying more than 1,400 farmers in the United States and Canada. It was the first year the annual study, which dates to 2017, asked about artificial intelligence. Julia Eberhart, Bushel’s director of marketing, said 75 percent had not tried it and 11 percent were not sure, which left the 14 percent who said yes.

The company’s own summary put early adopters in business analysis, writing and planning. Among larger farms that already use AI, 50 percent said they use it for business or financial analysis. Only 25 percent of AI users named yield prediction or agronomy. Eberhart told a May 26 radio briefing that decision-making on input planning was still 36 percent of that 14 percent group, and that writing and business work ran higher.

THE BUSHEL AI MIX

Farmers under 50 made up 38.4 percent of the 2026 pool, up from 28.8 percent a year earlier, the youngest mix Bushel has published. That shift did not drag agronomy to the top of the AI list. It produced a younger office.

ChatGPT Beat Yield Models to the Farm

The platforms in the survey were not farm-branded yield engines. Eberhart said farmers named ChatGPT first, then Gemini and Copilot. That order matches how the work actually gets done: a browser tab, a document, a spreadsheet, a prompt about cash flow or a contract, then a human who still signs the check.

Interestingly, it’s a lot more on like editing documents, helping with business and financial analysis. We have yield prediction in agronomy is like the last selectable available answer.

Julia Eberhart, director of marketing, Bushel, on the Ag Information Network

Rachael Sharp, who farms with her father Don C. Sharp III near Allendale, S.C., has been putting that desk stack to work. She uses ChatGPT to size pesticide orders so the shed is not left with two pallets worth $20,000 at year-end, and she uses it to file mandatory water-use reports for South Carolina that used to take two to three days. “We don’t order extra chemicals anymore,” she said. Jeremy Groeteke, global head of IT and digital strategy at Syngenta, has watched the same pattern among growers: large language models such as Gemini and ChatGPT are moving fast, while agentic systems that act on their own have not hit the farm gate.

Eberhart’s other read on the 11 percent who were “not sure” is that AI is already sitting inside software they open every week. They may be using it without a name for it. That still is not a yield model making the call.

The Cost of Being Wrong in a Field

The gap is not that farms are analog. It is that the current tools want fast feedback, tidy data, and mistakes that are cheap to undo. A crop year gives you one shot. Neighboring fields need different answers. Data that looks abundant on a card in the cab is often unusable in a prompt. So the work that actually moved first is the work where a wrong answer costs an hour, not a season.

WHY YIELD MODELS WAIT

• Reversible errors: A weak paragraph is cheap to fix. A bad tank mix can wipe a crop.
• One cycle a year: Chatbots learn from rapid retries. A corn plan does not.
• Messy field data: Yield files, as-applied maps, and soil tests rarely arrive in one clean table a public model can trust.

Precision hardware already cleared that bar because the gain shows up this season: a guidance pass you can see, a section shutoff that does not overlap. The lag is software that claims to decide. Until a model can show its work on a label rate, a hybrid, or a nitrogen split, it stays in the office, which is exactly where Bushel found it.

What Purdue Heard in June

Three months after Bushel published, Purdue University’s Center for Commercial Agriculture asked a blunter question. The June Purdue University-CME Group Ag Economy Barometer, fielded among 400 farmers from June 15 to 19, 2026, asked what they saw as the main benefit of AI or data-driven tools. Michael Langemeier and Joana Colussi reported that 52 percent did not see a meaningful benefit.

WHAT PURDUE’S 400 FARMERS PICKED

On a second question, 63 percent said recommendations from data-driven tools would sometimes be hard to follow, and 22 percent said they would often be hard to follow. High input costs were the top worry, named by 47 percent, and 42 percent said those costs were limiting any improvement in their financial position. The barometer itself slipped to 113 in June from 119 in May.

May’s survey, taken May 11 to 15 among another 400 farmers, had already asked whether AI tools would help with labor and equipment. Fifty-nine percent said no, 37 percent said a little, and 4 percent said a lot. Langemeier’s later read was that most respondents still had not figured out how the tools would help their operation. That is the same desk-versus-field split Bushel measured, asked from the other side of the ledger.

Phones Already Run the Farm Desk

The pipes for a chatbot were in place before anyone labeled the work as AI. USDA’s National Agricultural Statistics Service, in its August 2025 farm technology report, found that 85 percent of farms had internet access, 82 percent had a smartphone, and 68 percent had a desktop or laptop. Fifty percent used the internet to buy agricultural inputs, up from 32 percent in 2023, and 29 percent used it to market agricultural activities, up from 23 percent. Fifty-five percent connected through broadband and 74 percent through a cellular data plan.

THE PIPES WERE ALREADY THERE

• Internet: 85% of U.S. farms had access in 2025.
• Phones: 82% of farms had a smartphone; 68% had a computer.
• Buying: 50% purchased inputs online, up from 32% in 2023.
• Selling: 29% marketed agricultural activity online, up from 23% in 2023.

Bushel’s own digital-commerce numbers sit on top of that base. Digital tools for grain marketing rose from 21 percent in 2024 to more than 31 percent in 2026, and 56 percent of farmers said they use an app or software for grain marketing. Among farmers under 50, 54 percent said they were somewhat or very likely to use an app or website to submit a firm offer or sell grain, even though many still cannot. The same under-50 group is paid by paper check 82.8 percent of the time and prefers a check only 54.9 percent of the time, a 27.9-point gap, the widest in the report.

ChatGPT did not have to invent a farm network. It sat down on a desk that already had a phone, a login, and a pile of PDFs.

A Trial Cannot Cost $30,000

The money around that desk is tight, which is a poor climate for a new agronomy subscription and a fine climate for a free chatbot. Bushel found equipment financing up to 39.1 percent in 2026 from 28.0 percent in 2025, operating loans up to 38.9 percent from 29.6 percent, and real estate loans up to 31.2 percent from 21.6 percent. Eberhart has said farmers will try new tools, but the trial cannot cost $30,000 just to see if the thing works.

That price test sorts the market. A ChatGPT tab clears it. A dealer package that wants new sensors, a data plan, and a per-acre fee often does not, especially when Purdue’s June sample is already naming high input costs as the thing holding the operation back. Bushel now powers more than 3,500 grain and ag retail facilities, more than 50 percent of U.S. and Canadian grain origination, and more than 100,000 farmers use its software. Even inside that network, AI showed up as writing and finance, not as an in-field advisor.

Trust sits behind the price. A later write-up of the same Bushel study said growers who skip AI most often cited confusion about what it would help with, a lack of time to look, no clear benefit, and mistrust of the tools and the data. Katya Kheistver, chief product officer and chief operating officer at the precision-ag firm OneSoil, said farmers need to see how a tool reached a conclusion so they can catch a mistake before it gets expensive, and that many systems still run like a black box.

The Field Bill Came After ChatGPT

Washington is still writing the field version of this story. Sen. Ted Budd, with Sens. Adam Schiff, Jim Banks, Catherine Cortez Masto, Mike Rounds, and Lisa Blunt Rochester, introduced the FARM AI Act of 2026 on May 21. The bill’s full title is the Fostering Agricultural Research and Modernization through Artificial Intelligence Act. It would add AI as a priority under USDA’s Agriculture and Food Research Initiative, expand work through the Agriculture Advanced Research and Development Authority, put more weight on Extension, and create an AI in Agriculture Advisor to work with the National Institute of Standards and Technology on farm AI standards. Budd said that if access barriers are not fixed, American producers will fall behind.

Reps. Zach Nunn of Iowa and Don Davis of North Carolina filed the House companion on July 14 and said the aim was to put artificial intelligence to work on farms by modernizing USDA research and workforce programs. Both bills were still in committee after introduction. They fund the agronomy layer Bushel found in last place.

FROM THE SURVEY TO A TRACTOR CHATBOT

1. April 2, 2026: Bushel releases the State of the Farm report and puts farm AI in the office.
2. May 21, 2026: The Senate FARM AI Act is introduced to push USDA research and Extension toward field AI.
3. June 15, 2026: Purdue begins a 400-farmer survey that later finds 52% see no meaningful benefit from AI or data-driven tools.
4. July 14, 2026: The House companion lands in the Agriculture Committee.
5. Early September 2026: John Deere introduces JD, an AI assistant built into Operations Center.

JD is the first large farm system to meet growers inside software they already open, rather than sending them to a public chatbot. It arrives after the desk has already moved. The 14 percent who told Bushel they use AI were not waiting for a Senate finding or a cab assistant. They opened ChatGPT, edited the document, and left yield prediction on the last line.