Upstream Ag Insights GenAi in Agribusiness Report: How are Industry Professionals Using Artificial Intelligence?

Index

1. Introduction

2. Key Takeaways

3. Are Agribusinesses Encouraging AI Usage?

4. How Agribusiness Professionals Are Actually Using AI

5. What are the biggest downsides or frustrations you experience with current AI tools?

6. What tools are most commonly used by Agribusiness Professionals?

7. What Tasks Do You Wish AI Could Handle But Currently Does Poorly?

8. What Is Your #1 Expectation from Agricultural Software Providers Regarding AI?

9. How Would You Rate the AI Training Provided by Your Organization?

10. Is Your Organization Currently Building Its Own Internal AI Tools?

11. How Trusting Would You Be to Let an AI Agent Execute a Transaction Without Your Final Approval?

12. Which Agribusiness Activity Will Always Require a Human, No Matter How Good AI Gets?

13. What fears do you have surrounding artificial intelligence in agriculture?

14. What opportunities do you see AI creating in agriculture?

15. What This All Means

16. About the Survey and the Survey Takers

Introduction

Every technological shift follows a similar arc. First, skepticism. Then, experimentation by a curious few. Then, a tipping point where non-adoption becomes the bigger risk.

We're seemingly somewhere in the middle of that arc with generative AI where we are beyond the curiosity phase, but not yet at the point where the industry has figured out what "good" looks like, or exactly where the value lies.

One we reach a tipping point, the groups deriving the biggest advantage will be the one that creatively integrate into their business.

Tom Goodwin made an observation in Digital Darwinism that applies directly to GenAI — the companies that struggle most with new technology aren't the ones that fail to adopt it. They're the ones that layer it on top of old processes and call it transformation. They might buy the tool, but then don’t evolve the workflow. Or, they buy licenses, but don't invest in the training or knowledge infrastructure.

That pattern of clunky, in cohesive usage shows up in agribusiness right now and is something to keep in mind.

Below you will find an overview of how agribusiness professionals are using AI, where it's falling short, what they wish it could do, and what it all means for the companies building tools, the organizations deploying them, and the individuals whose daily work is being adjusted.

Note: This survey is intended to establish a baseline for how agribusiness professionals are thinking about and using generative AI in their daily work. A few important caveats: This isn't a statistically significant study, nor is it meant to be the definitive word on GenAI adoption in agriculture. Rather, it's intended to move beyond anecdotal conversations and provide directional insight into where the industry stands with AI as of February 2025.

Key Takeaways

This survey ended up being of 198 agribusiness professionals with a heavy skew toward experienced, senior leaders at mid-to-large companies. For more on the survey takers, see the end of this report.

• 80% of respondents at companies that are 500+ people stated that usage of AI tools was encouraged.

  • If a respondent worked at a company where AI usage was encouraged they were 2x as likely to use AI tools multiple times per day.

  • Respondents that worked at a company where AI was encouraged often felt AI saved them even more time per week on average. They were most likely to say that they saved over 8 hours per week.

• 65% of GenAI users reported saving 3+ hours per week, with 30% saving 5+ hours. 17% said they feel that AI tools save them 8 hours or more per week. Marketing/Communications professionals reported the highest gains of any job-type cohort.

• Over 50% of respondents are using GenAi tools multiple times per day.

• 86% of users suggest the quality of their work improves with AI tool usage.

• 50% of respondents said they paid for AI tools out of their own pocket.

• 57% used only one GenAI tool, while 23% used two and 20% used 3 or more.

• Those who pay for AI out-of-pocket use coding tools at 10x the rate of non-payers, such as Claude Code, Replit or Loveable.

• Those that had access to company provided training were much more likely to use AI multiple times per day and save more time than their counterparts without AI training. 55% of those with advanced training save 5+ hours per week, compared to just 18% of those with no training. 80% of those with advanced training use AI multiple times daily, compared to just 35% of those with no training. 61% of those with advanced training report significantly better quality vs. 27% with no training.

• The biggest fear with AI is system accuracy. 68% cite hallucinations/inaccuracy as their top concern, far exceeding job displacement (18%).

• Agentic AI adoption remains cautious. Only 11% fully trust AI to execute transactions autonomously. 60% will only trust it for small, low-stakes tasks.

• 65% of companies with 1,000+ employees are building proprietary AI tools, compared to just 13% of solo/small businesses.

For any questions on the survey, please reach out to [email protected]

Are Agribusinesses Encouraging AI Usage?

Over two-thirds of respondents said that their company encouraged AI usage:

Notably, 80% of respondents at companies that are 500+ people stated that usage of AI tools was encouraged, exceeding those that are sub 500.

When it came to using GenAI tools, if a respondent worked at a company where AI usage was encouraged they were 2x as likely to use AI tools multiple times per day, and also take advantage of multiple different types of AI tools.

It’s interesting to consider that when we factor in time saving and efficiency:

Respondents that worked at a company where AI was encouraged often felt AI saved them even more time per week on average — they were most likely to say that they saved over 8 hours per week, essentially gaining a full work day worth of productivity.

Over 50% of respondents are using GenAi tools multiple times per day, with more usage correlated to higher time savings.

49% of respondents said that they have become reliant on GenAI tools and would be very disappointed if they lost access to the tools. I was surprised this was not higher — humans do not do well with having something valuable and then losing it.

What is more interesting is how GenAi is perceived to impact quality of work:

 86% of users suggest the quality of their work improves.

Take the efficiency gains and quality gains together and it suggests that if a company is not actively encouraging AI usage and providing the frameworks and expectations accordingly, they are at a risk of falling behind their competitors.

As an aside, about 50% of respondents said they paid for AI tools out of their own pocket. Those that spent out of pocket had the highest perceived time saved and quality improvements.

How Agribusiness Professionals Are Actually Using AI

The survey data shows that AI adoption in agribusiness isn't a monolith. There is a lot of variance in how people use these tools based on their role, company size, and years of experience.

40% of individuals that identified as Consultants cite technical research as their primary AI use case — the highest of any role surveyed.

Product Managers usage splits between drafting communications (38%) and coding/technical work (25%), which is interesting. They're using AI on both ends of their workflow — writing things like product specifications and stakeholder updates on one side, and increasingly building functional prototypes on the other. This dual use is consistent with what Marc Andreessen shared was occurring on Lenny’s Podcast, at least with the digital product managers.

Unsurprisingly, Sales and Marketing teams lean heavily into communications, with 36% focused on drafting emails, prospecting messages, and customer outreach.

There were some Farmers that completed the survey. Interestingly, they showed no dominant use case. Their AI usage spreads evenly across data analysis, ideation, and decision tools.

When it comes to what agribusiness professionals want AI to do the most, the survey overwhelmingly identified two main areas:

1. Administrative Tasks (CRM entry, paying invoices, following up on late payments etc.) with 50% citing it as what they wish AI did right now.

2. Market Insights (synthesize insights from market data materials available), with 31% suggesting they want AI to do this.

As one aside on the synthesis of market insights side of things — using the same models and the same data as competitors means a convergence to the mean if using AI for deriving insights (and therefore to create strategy), that means there is a need to acquire better data than competitors, or rely less heavily on AI to develop insights so that there is human creativity involved.

By Company Size

What stood out to me was variance of use by company size.

Companies with 500+ employees use AI for meeting synthesis at nearly four times the rate of smaller firms. I do wonder a lot about these meeting note takers as it pertains to lawsuit risks in the future and I have been involved in meetings where participants disagree vehemently with allowing virtual note takers into the meeting.

Small companies, meanwhile, are using AI to make up for where they might lack in resources. They lead in both research (22%) and ideation (19%) — leveraging AI for analysis and strategic thinking that larger competitors might get from bigger teams and more resources. For a five-person agtech startup or a small consulting firm, AI effectively expands the team without adding headcount, and is consistent with what Microsoft CEO Satya Nadella shared as a use case he sees with AI in creating a digital twin of colleagues by role.

By Years of Experience

An interesting finding is how AI use cases vary by years of experience.

Early-career professionals use AI more for writing. Nearly half (46%) of those with less than five years cite drafting as their primary use case.

Mid-career professionals (6-10 years) shift toward ideation, using AI to develop and pressure-test concepts they now have enough domain knowledge to generate.

Veterans (21+ years) primary use was in research, with nearly one-third citing it as the most prominent use case.

What are the biggest downsides or frustrations you experience with current AI tools?

Generic outputs were cited as a universal frustration. 73% of individuals using AI primarily for writing/communications had concerns about generic/shallow outputs. The more creative the task, the more AI's generic nature leads to frustration.

Hallucinations worry researchers and farmers most. 59% of research-focused individuals cited hallucinations as a top frustration. Farmers (57%) share this concern, likely because they will feel the consequences very directly.

47% of those with 21+ years experience cite erosion of critical thinking as a major concern which was the highest of any experience level. The heaviest users were concerned about critical thinking erosion, too, with 50% of daily users citing this concern.

Workflow disruption was most prominent in commercial roles (market, sales) with 24% concerned about workflow disruption, the highest of any function. They dislike the bouncing between CRMs, ERPs, email, and AI tools constantly, which reinforces the opportunity for agribusiness software cited in AI and Agribusiness Software: From Systems of Record to Systems of Action.

In fact, on the question about how would you prefer to access AI capabilities in the future, 62% cited wanting to have everything integrated into where they already work which is likely their company control point. At the end of the day, a technology is truly here when it fits into the background. If it is not invisible, it still can be improved, and today, most tools exist outside core workflows which shows plenty of runway for improvement.

Of note, the most frequent users notice hallucinations more. 47% of multiple-times-daily users cite hallucinations, almost 2x higher than infrequent users.

What tools are most commonly used by Agribusiness Professionals?

Generalist tools like ChatGPT are used by the majority of respondents whereas more specific tools are used at a lower rate:

_{Note: I did not try and get a specific understanding of what generalist LLM was used the most. For example, if I took the survey three months ago I would have said ChatGPT, if I took it one month ago I would have said Gemini and if I took it today I would say Claude. I do not think I am the exception as everyone tries to learn more about the capabilities and functionality across different offerings.}

57% used only one GenAI tool, while 23% used two and 20% used 3 or more.

Unsurprisingly, Microsoft Copilot is used by over 80% of individuals at company sizes that are 500 or more people illustrating the penetratiok of Microsoft Office and the effort to bundle CoPilot.

Overall, large companies are less experimental. They dominate Copilot usage but trail in every other specialized category — meeting tools, coding tools, and ag-specific AI. Smaller and mid-sized companies are much more experimental. This could be due to policy constraints.

Coding tools correlated highly with personal investment in AI tools. Those who pay for AI out-of-pocket use coding tools at 10x the rate of non-payers, such as Claude Code, Replit or Loveable. Personal payers are also 3x more likely to use image tools (eg: Midjourney).

What Tasks Do You Wish AI Could Handle But Currently Does Poorly?

The industry wants AI that takes actions. There are trends that illustrate this is coming, particularly for anyone that spends any time on X reading about agentic AI.

One of the top wishes was to remove administrative tasks, moving beyond just answering questions towards autonomously performing analysis and taking action on what is found.

Prediction is a universal aspiration with ~45% of leadership wanting things to move in that direction.

Junior employees, those with less than 5 years want market intelligence, illustrating that they are wanting to get up to speed on market dynamics more rapidly. Given the level of market data out there from the likes of AgData, Kynetec or Stratus Ag Research for example, it seems like this is a natural product extension for them that not only supports less experienced professionals, but also those with more experience.

Automated reporting was most in demand from commercial roles (sales, marketing) and leadership. 42% of leadership and 40% of commercial roles want automated reporting capabilities.

One survey taker shared a comment that they want customer digital twins with the ability to "interrogate" customers about hidden pain points. I think that’s a really interesting call out, and there are some tools out there today, like askrally.com that can give aggregate insights. There is an opportunity to create this using coding tools, too — for example, I recently cited AgTalk Pulse where an individual built a scraper to collect data from online farmer focused forums, aggregated the data and then built an interface to share real-time agricultural market sentiment. There is an ability to do that same thing, except use the specific data to to create persona’s from actual farmers, using real farmer online behavor, supporting better market and customer intelligence.

One other idea shared by a survey taker was the desire for an AI scheduling and meeting booking tool for setting up in-person conversations with farmers.

What Is Your #1 Expectation from Agricultural Software Providers Regarding AI?

Accuracy was the top demand.

There was some nuance by role. 60% of technical staff prioritized accuracy as #1, far above any other role in the survey. Those working for large companies (500+) also prioritized accuracy the most which isn’t surprising given the reputational risk, number of affected people etc. with a ny inaccuracy.

Small companies are the only segment interested in adaptable interfaces (8%) and show the highest interest in workflow integration (18%). My gut feel surrounding this is those working at smaller enterprises have been more exposed to what’s possible with AI and have the expectations that align with that incremental exposure.

Notably, only a small percentage were concerned about the price. It seems survey takers understand that if enough value can be unlocked, they are willing to pay for it. Only 4% cite cost as their #1 expectation.

While there were not many farmers that took the survey, what did stand out to me was that they disportionately want adaptable interfaces. 40% of farmers want adaptable UX/UI, which I would extrapolate to more voice capabilities.

How Would You Rate the AI Training Provided by Your Organization?

This was an eye opening segment for me.

Many organizations provide no AI training.

43% of respondents report non-existent training within their employer, and are left entirely on their own to figure it out. Another 36% received only a webinar or policy email.

Just 1 in 5 have access to dedicated training sessions, internal AI resources, or AI experts.

Large companies (500+) are 4x more likely to provide advanced training. Small companies (<500) are adopting AI but not investing as heavily in staff enablement.

One stat that surprised me was that junior employees suggested they get the least help, as 75% of those with <5 years experience have had no access to AI training, higher than any experience level.

The really interesting part is around the training relationship to time saved and improvement in output quality.

Those that had access to training were much more likely to use AI multiple times per day and save more time than their counterparts without AI training — 55% of those with advanced training save 5+ hours per week, compared to just 18% of those with no training. 80% of those with advanced training use AI multiple times daily, compared to just 35% of those with no training.

61% of those with advanced training report significantly better quality vs. 27% with no training.

Arguably, there is a “chicken and egg” scenario going on, but the clearest takeaway from the survey was that when companies invest in giving their team tools, they take advantage of it in a way that should deliver compounding returns to the business over time.

Is Your Organization Currently Building Its Own Internal AI Tools?

Building internal AI is primarily a large company (>500) initiative.

61% of large enterprises are building proprietary AI tools vs. just 16% of small companies.

Still, almost one-third of respondents suggested their organization was going to begin building AI tools internally in 2026.

42% of organizations building internal AI provide advanced training, compared to just 6% for those not providing it, which shows that building AI and enabling employees to use it go hand-in-hand.

Building organizations have higher trust in agentic AI, too. 72% of those at building organizations would trust AI for small tasks (vs. 54% at non-builders).

How Trusting Would You Be to Let an AI Agent Execute a Transaction Without Your Final Approval?

There wasn’t a lot of surprises here — I have highlighted the psychological dynamics of human trust in The Shortcomings and Opportunities of Large Language Models in Agronomy, so overall, this was as expected:

What will be interesting to see is what this looks like in 2-3 years. One thing I think I have poorly considered until this point is human adaptation. We are incrementally more exposed to AI and it’s capabilities on a daily basis, which theoretically should increase our openness to trusting it. For anyone that has taken a ride in a Waymo for example becoming much more open to autonomous driving, this same sort of phenomona could occur with task execution in the future. We can even look at the data from this survey: Heavy users are already the most trusting with 77% of those who fully trust AI using it multiple times daily.

Intensive use builds up confidence faster.

In that same vein, there are some tasks that agribusiness professionals never see AI taking over:

Which Agribusiness Activity Will Always Require a Human, No Matter How Good AI Gets?

68% believe handling frustrated farmers and claims will always require a human. Even as AI improves, empathy, judgment, and accountability in disputes should remain fundamentally human according to survey takers.

Notably, technical staff defend diagnostics requiring a human. 35% of technical staff believe complex field diagnostics will require a human — the highest of any role. They likely consider the nuance, edge cases, and contextual judgment that goes into identifying an in field issue.

What fears do you have surrounding artificial intelligence in agriculture?

Inaccuracy is the dominant fear with 68% of respondents citing inaccuracy/hallucination as a fear, more than double any other concern.

100% of farmers fear inaccuracy and given they have a lot to lose in a recommendation setting — whether selling grain, or using a specific agronomic recommendation, it shows that there is going to be a desire for human judgement to “buffer” against as many wrong answers as possible. Notably, humans make mistakes as it is, however, just like in autonomous vehicles, even though autonomous driving has been shown to be safer, there is still an uproar when an autonomous vehicle is in an accident.

Job displacement fear was surprisingly low, except among those with less than 5 years industry experience. Almost one-third of junior employees cited it as a fear. Those with 21+ years seem confident their expertise will remain valuable, which to me further reinforces the need to improve AI education and show young staff how they can use AI tools to improve their ability to learn.

Only 1% have no fears at all. Virtually everyone has concerns about AI in agriculture, so if something is concerning to you surrounding AI, you aren’t alone.

What opportunities do you see AI creating in agriculture?

On a more optimistic segment, there was an open ended question about the biggest opportunities. Here are some of the more than 100 responses to this question (primarily copy and pasted, only edits were for spelling errors or any comment that could identify the individual):

• “ESG Assessment / resilience management to consider better options protecting water, biodiversity, soil, etc.”

• “I expect it to improves ag sales people’s knowledge of farmers and prospecting.“

• “Faster analysis of policy changes or large bits of information. Also supporting those with dyslexia or ADHD to better focus and communicate clearly.”

• “Voice interaction with farmers to capture data and provide insights while farmers are on the go”

• “Today, as a business leader, I worry that the output I am getting contains less of the experience I hired / paid for and more AI. My concern is the cascading of thought and development of the next idea / strategy that comes from a human within the AG market. It helps tremendously with coding, verification / support for some of our copy / communications but still trying to figure out how to put it in functions where we can have "trusted" but verified output.”

• “"1-Acceleration of R&D through AI-Driven Data Analytics - Leverage advanced AI and data analytics to significantly reduce R&D cycles, improve formulation accuracy, and enable faster innovation based on large-scale agronomic and market data insights. 2-Real-Time Farm Monitoring, Intelligent Decision Support, and Future Automation - Continuous monitoring of farm activities through sensors integrated with AI-powered software to rapidly identify challenges and recommend optimal solutions for crop protection, plant nutrition, irrigation, and harvest operations. In the future, a large portion of daily farm operations will be executed autonomously by AI-driven agricultural agents, enabling automated actions, higher operational efficiency, reduced human intervention, and consistent precision farming at scale."

• “LLMs today solve 80% of the data interoperability problem. LLMs+will can solve 100% of the problem. That's HUGE. That means a large farm operation really can have a tech consultant who enables multiple tech platforms and layers to exist in the operation while pulling high quality, if disparate, data. That single person (or small team - you do need data science knowledge) then can truly process and "value add" disparate data sets to unlock competitive insights on an operational level. This could also be done on a cooperative level (see AgLaunch Data Commons.) The problem is, this means large farms are better served by tools than by platforms, so this further fractionates the agtech market from a VC vantage point.”

• “AI should be able to take a yield map, show planting pop, hybrid, trials with biologicals, etc all with a simple click. today you have to have special software thats cumbersome to even get an idea of what a grower did if he labeled it correctly as a Side by Side in the field. so much wasted data because we don’t know all of the variables and what can be drilled down on.”

• “The opportunity is immense. I see a utopian world where everything in and around agriculture can be done be LLM's + Machines+ Wearables+NLP+RAG where human intervention is required only for strategic decision making in next 1-2 decades.”

• “Faster software creation and delivery”

• “More access to structured data beyond agronomic and sales is an opportunity. ie: Farmer behavior”

The survey responses about what agribusiness professionals want from AI clustered around three themes:

1. better insights

2. less administrative work

3. and faster decisions.

These are all opportunities. I think there can be more, though.

Sangeet Paul Choudary argues in Reshuffle that the real value of AI isn't in optimizing individual tasks within a workflow. It's in becoming the coordinating layer that connects tasks, decisions, and actors across an entire company or value chain. I talked about this more in AI and Agribusiness Software: From Systems of Record to Systems of Action.

Tying it All Together

AI has gone from a curiosity or a toy to becoming critical knowledge infrastructure in agribusiness.

65% of respondents are saving 3+ hours per week, 86% report improved work quality, and nearly half would be "very disappointed" to lose access. While according to PWC, most CEOs (non-ag specific) say their companies aren’t yet seeing a financial return from investments in AI, the survey data suggests the foundation is being laid to improve worker productivity and effectiveness.

Investing in training and education is delivering a more AI-curious, and productive, agribusiness professional.

The gap between those with access to employer offered AI training and those without stands out. 55% of users that had access to training save 5+ hours weekly versus just 18% of users that did not receive training. 80% of trained users engage multiple times daily compared to 35% without training. 86% of users suggest the quality of their work improves with AI tool usage.

If your company is encouraging AI usage, providing training, and giving people the frameworks to experiment, you're building a compounding advantage. If you're not, your competitors are likely getting aheads.

People don't want AI to do the interesting parts of their job. They want it to do the parts they hate.

When looking at what agribusiness professionals wish AI could handle, the answer wasn't the “fun” stuff like strategic thinking. It was administrative tasks (50%) that allows time from higher-value activities. An interesting framework for thinking about AI tools to create or invest in in agriculture might be to follow the stuff that’s important, but tedious and time consuming.

The people extracting the most value from AI share a few things in common:

1. They've found their AI jobs-to-be-done. Consultants use it for technical research. Product managers split between communications and prototyping. Sales teams draft outreach. The highest-impact use cases are role-specific.

2. They treat AI like a collaborator. They use it as a sounding board rather than a tool to merely do all the work.

3. They acknowledge the tradeoffs and adapt accordingly. Generic outputs, hallucinations, time spent reviewing — these are realities. But the people getting the most value aren't waiting for these problems to disappear, they adapt their workflows that account for them.

The agentic future everyone's talking about has not made it’s way en mass to ag, yet.

Only 11% fully trust AI to execute actions autonomously. The vast majority will only delegate small, low-stakes tasks. But the interesting take away is that heavy users are more trusting. Intensive use builds confidence. As exposure increases across the industry, trust will likely follow.

About the Survey

The Upstream Ag Insights survey was available for agribusiness professionals to complete from January 7th until January 30th 2026.

198 individuals took the survey with the largest percentage being those operating in a company that is 51-500 people in size:

The roles skewed towards leadership, with over a third of respondents identifying as being an executive/leader/founder:

The individuals experience levels skewed heavily towards 11+ years:

For any questions on the survey, please reach out to [email protected]