The agricultural sector consists of several types of work that have little in common. There is physical work in the barn, greenhouse or field, tied to season, weather and biological processes that cannot be sped up. There is administrative work around registration, subsidy applications, manure accounting and compliance with regulations that differ per crop, animal or region. And there is a third layer: planning, procurement, crop choices and the assessment of data from sensors, satellite images and market prices. That third layer is growing, while the first layer remains limited in scalability due to the nature of the work itself.
The conditions that steer the outcome are not uniform. An arable farming business with a lot of open data on soil and weather has different points of leverage for automation than a livestock farm where oversight of animal welfare is legally reserved for humans. A horticulture business with climate computers has been steering based on sensor data for years; an arable farmer who decides manually per plot has not. That difference determines where AI is already taking over tasks today and where that is still far off.
Three categories run through the work in this sector. First, tasks AI can carry out independently: interpreting sensor data for irrigation, recognizing disease patterns based on imagery, drafting initial versions of reports for subsidy applications. Second, tasks where AI makes a proposal and a human approves or rejects it with reasoning: a cultivation recommendation that is checked against local knowledge, a procurement proposal that is assessed for supply risk. Third, work that remains human work: physical oversight of animal welfare, on-site assessment of soil quality, negotiation with a buyer over a long-term contract.
This division is not a snapshot that shifts uniformly everywhere. Businesses that already work with sensor data and decision-support software see the first category growing, because the data structure is already in place. Businesses that still operate on experience and manual registration mainly see the second category emerging, because there someone first has to determine what is reliable enough to hand over to a system. The difference between businesses therefore does not lie in ambition, but in the degree to which the underlying work has already been captured in a form a system can read.
A strategy based on a fixed staffing requirement for administrative or monitoring tasks rests on an assumption no one has explicitly stated: that this work can only be done by humans, at a certain pace and a certain cost. As soon as AI takes over or supports part of that work, the assumption does not lapse with an announcement. It lapses gradually, while the plan on paper remains unchanged.
That is the pattern this sector shares with other sectors where AI affects task level rather than job level. A similar shift is taking place in the IT sector, where AI partly takes over coding and testing work, in financial services, where oversight of automated assessments is taking on a new role, and in the recreation industry, where planning and customer contact are being divided differently between system and human. The agricultural sector has its own pace, determined by season and regulation, but the mechanism is the same: an assumption that once held true is not withdrawn, it simply stops being tested.
This is not about the question of whether a business should reduce staff or deploy it differently. That decision falls under the applicable statutory requirements around labor law and employee participation, and rules apply there that are not addressed here. This concerns something that precedes that: knowing which work is today actually being done by AI, which work is done under supervision, and which work remains human work. That is a factual question, not a staffing decision.
The question of which work in this business can genuinely be taken over by AI is answered per task by the FTE TO AI work scan, so that assumptions about capacity no longer rest on estimation. Those wanting to know more precisely what a strategic assumption actually is will find that explained in the explanation of what a strategic assumption is and how you determine whether it still holds, and those wanting to know how a current strategy is checked against the present situation can read that in the explanation of testing an existing strategy.
The first step is not a fully revised plan, but a brief inventory of the assumptions on which the current plan rests. FTE TO AI offers a free assumption check for this: a short round in which you name your key assumptions and see, per assumption, when it was last confirmed. This does not deliver a judgment on the strategy itself, but it does provide insight into which parts need attention before they are adjusted. The full pressure test, with an external view alongside management's own self-assessment, is under construction.
Stel uw vraag. Vaak zit de echte vraag een laag dieper — daar mag ik naar vragen.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.