Central Lever of Tomorrow's Agriculture: Automation Faces Economic and Scale Barriers, Says Romain Faroux

Sensors, satellite imagery, decision support tools, input modulation, robotics: automation has been progressing for several years in agricultural operations. However, its impact remains uneven across sectors and regions, and struggles to deeply transform the entire industry.

For Romain Faroux, who has been working for over fifteen years on agricultural technologies and their operational deployment, the central question is no longer technical but systemic: agriculture is not facing an innovation deficit but a scaling-up deficit. In other words, solutions exist, but they are still too little disseminated to produce a systemic effect. Without massification, without robust business models, and without large-scale learning mechanisms, automation will remain a partially fulfilled promise.

In this interview, he discusses the different levels of automation, from agronomic decision-making to robotization, the economic and structural reasons that hinder their adoption, and the necessary conditions to make it a true lever for transformation. Between the diversity of agricultural models, the role of public policies, and the importance of collective learning mechanisms, he outlines the contours of an agriculture destined to become "less intensive in inputs and more intensive in knowledge" (according to Hervé Pillaud, in L'IA des champs).

How do you assess the current level of automation in agriculture?

It is clearly insufficient, and it could have been much more advanced if a structured vision had been maintained over time. We see a fairly classic phenomenon in France: a sprinkling of aids, numerous but poorly aligned initiatives, and ultimately slow diffusion.

Automation, at its core, is the ability to be more efficient with equal or lesser input consumption. It primarily improves the application of resources, particularly human and agronomic. But it does not necessarily reduce all expenses: the energy required for a pass will remain the same. However, we can optimize inputs or the organization of work.

The problem is that as long as solutions are not deployed on a large scale, they remain too expensive, thus poorly adopted. We remain in a circle where supply does not really meet demand.

" As long as solutions are not deployed on a large scale, they remain too expensive, thus poorly adopted."

Where does automation stand today on the ground?

We need to distinguish several levels. First, there is the automation of operations, with robots capable of performing tasks in place of a human. Then, the automation of application, for example, a towed tool capable of finely adapting its work thanks to sensors and measurements. Finally, there is the automation of knowledge, with decision support tools that guide technical itineraries.

Regarding the automation of application tools, we are around 20% of the market in France, which is still limited. And above all, the progress is slow: ten years ago, we were around 10 to 15%. It is clear that the acceleration remains moderate.

Is the automation of decisions more advanced?

Yes, clearly. Agronomic management tools, particularly regarding nitrogen through satellite imagery or sensors, are now quite well automated. We can measure the nutritional status of a crop and adjust inputs with a good level of reliability. The same is true for models predicting health risks, whose robustness and dissemination are encouraging.

This relies on a long process of aligning field measurements with imaging data (satellite, drones, onboard sensors). The calculation is largely automated, although human reference measurements remain necessary. In this area, France is actually quite advanced.

Why is field robotization progressing more slowly?

Because the conditions are much more complex. France is not a territory of large homogeneous plains. We have relief, hedgerows, varied plots, and strong regulatory constraints. This makes standardization much more difficult than in regions where one can work for kilometers in a straight line on homogeneous soil.

But above all, the challenge is not just to advance a machine on its own. The challenge is much more complex: for the tool to be able to adapt its behavior to agronomic conditions: working depth, input dosage, reaction to soil variations.

Can we imagine a shift towards agriculture largely under controlled conditions?

Controlled environment systems clearly represent one of the most favorable grounds for automation. Models like " Les Fermes Debout ", distinct from vertical farms) illustrate this logic: planifiable production, advanced automation of operations, fine management of inputs, and even automation of marketing. The example of the milking robot in dairy farms would also seem useful to add here, to highlight two successes of automation in controlled conditions. It was also too expensive during an initial development phase and took 15 years to find its balance before being able to accelerate significantly in the market).

These systems allow for the creation of very robust and manageable production units, but they rely on very high investment levels, often around one million euros per hectare. They are therefore relevant for certain productions, particularly local market gardening, but cannot constitute a model generalizable to all agriculture.

They will not replace large crops or the diversity of field systems. Controlled agriculture represents a complementary path that is set to develop, but not to become dominant.

Is the main barrier technical?

Not really. The real bottleneck is economic. Today, many farmers say that a robot costing 170,000 euros works very well but that they would be willing to buy it for 100,000. As long as we do not produce on a large scale, we do not reach these price levels. And as long as we do not lower prices, the market does not develop. Without scaling up at the European level, automation will remain the business of a few and will not transform agriculture in the short term.

" The real bottleneck is economic. Without scaling up at the European level, automation will remain the business

of a few and will not transform agriculture in the short term."

We often talk about a French delay. Is this the right reading?

The situation is more nuanced. On certain topics, particularly agronomic decision-making, France is very strong. Where it is slower is in machine automation, particularly in the field, for structural reasons related to production conditions. Comparing with highly specialized countries is not always relevant.

Is automation essential to improve agricultural competitiveness?

Automation is a necessary condition to improve the efficiency and resilience of farms, but it will not be sufficient on its own to resolve the economic imbalances in the sector, particularly related to rising costs and price pressure. It risks being a forward flight, serving a model that would focus its efforts solely on reducing production costs, without seeking the value creation that comes from diversification and resilience.

What role do public policies play in this dynamic?

A determining role. Agricultural innovation requires visions of fifteen or twenty years, while political cycles are five years. It would be necessary to massively develop reference farms allowing for large-scale test and learn.

If public money is not intended to support a market, it must accompany the emergence phases of breakthrough technologies, facilitate the rapid achievement of their industrial production thresholds, and help with their adoption. Public markets should be much more assertive levers in the current period of calls for sovereignty that we are going through.

" Agricultural innovation requires visions of fifteen or twenty years, while political cycles are five years.

If you had a magic wand, what would you do to accelerate automation?

It would be necessary to massively develop reference farms dedicated to learning and testing technologies in real conditions. The challenge is to create environments where one can experiment, measure, and accumulate robust data to build reliable models. These farms would align the various actors – sensor suppliers, machinery manufacturers, decision support tools, modelers – around common protocols. The goal would be to move away from a fragmented logic to create a true collective learning mechanism.

Today, many innovations remain difficult to deploy because each farm must integrate the solutions itself, which slows down diffusion. Well-equipped and closely monitored pilot farms would significantly accelerate the maturity of technologies and their scaling up.

Are we heading towards a polarization between agricultural models?

Yes, and this polarization is already largely structuring the debate. On one side, an approach focused on productivity and the maximum optimization of existing systems. On the other, a trajectory oriented towards agroecology and value enhancement with more diversified systems.

However, this opposition is often caricatural. In both cases, measurement and monitoring are common foundations. Automation can serve to optimize intensive systems as well as to manage more complex agroecological systems. We are moving towards a sustainable coexistence of several agricultural models adapted to territories.

What trajectory do you see for the coming years?

Automation will continue to progress; it is the direction of history. But the speed will depend on the alignment between technologies, business models, and public policies.

We are currently in the middle of the stream: technologies work, but the economic balance has not yet been fully found. As Hervé Pillaud says in L'IA dans les champs, "we will move from an input-intensive agriculture to an agriculture that will likely be even more intensive, but in knowledge. Digital technology provides us with the ability to aggregate all knowledge. And from this knowledge, we will build the agriculture of tomorrow".

Automation is a central lever, but its deployment depends less on technological prowess than on the ability to collectively organize its scaling up. Without this broad diffusion, it will remain a high-performing tool for certain systems, without producing the systemic transformation it promises.

"We are currently in the middle of the stream: technologies work,

but the economic balance has not yet been fully found."