ATAILA Newsroom · Budapest · 2026-08-06

Seven data points after one spraying — this is where AI adoption stalls

The Hungarian news site telex.hu profiled an AI-based crop-monitoring system developed at the University of Szeged — and explained why technology is not what holds back its adoption. Anna Farkas’s article describes, precisely, a pattern we see every week in industries far removed from agriculture: the model is ready, and the rollout stalls.

What telex.hu describes

According to the article, the Szeged system recognises stress affecting crops — water or nutrient deficiency, for instance — before visible symptoms appear, so the farmer can intervene before yield is lost. The technology side, in other words, is in place.

The constraint is not the model. According to Péter Miklós Varga, general vice-president of the Hungarian Precision Agriculture Association, the success of new developments is determined far more by users’ openness and digital readiness than by the artificial intelligence itself. The electronic farm journal requires recording plant-protection treatments, nutrient replenishment and other field operations. “We are talking about seven data points in total, to be recorded after one spraying,” Varga says in the article — adding that for many farmers even that is an obstacle.

Commercially, though, this is the article’s most important sentence: Varga argues for solutions that integrate into the systems farmers already know — and the article adds that it would help a great deal if the interface answered each entry immediately with a suggestion or a warning, because far more people would then use it in their daily work. In other words, not yet another standalone application. The article also notes that many producers still rely on their own experience, or on other farmers’ advice, more than on the data.

Not a rebuttal — the most accurate description of the problem

We have no quarrel with this. It is the cleanest statement of the problem we call the provisioning gap: the capability is ready, but there is no path to where the work actually happens.

One thing, however, we would put more sharply. If the bottleneck is located in users’ openness and digital readiness, it is easily read as the fault sitting with the user — as if the market will eventually grow into the technology. Our experience is different. When a model arrives as a standalone application — its own login, its own data entry, its own operations — the developer has shifted the cost of integration onto the user. The seven data points are not hard because there are seven of them. They are hard because the person recording them gets nothing back in that moment.

That is where our thesis comes from — and the Szeged example supports it exactly:

What ATAILA does about the gap

The article’s findings map point by point onto how we work.

That the gap can be crossed is best shown by our own practice: our references are our own products running in production on this same platform, and the newest of them went live in hours by reusing existing modules.

What we do not know, and what we will not take on

We are not an agriculture company. We have no plant-physiology model of our own, and we could not tell you what the Szeged system sees in a field — the Szeged researchers understand that, and by the article’s account they understand it well. Providing access to AI is one profession; agronomy is another.

Nor do we claim that a platform solves a question of trust. Someone who trusts decades of their own experience more than the data will not be convinced by a deployment. Good software can do this much: it makes trying cheap, and it gives something back immediately for what the user typed in. The rest is a matter of time and evidence.

And, to be honest about our own boundaries: a two- or three-person family farm is not our customer. ATAILA is for regulated or data-sensitive organisations of 20–200 people, where an expensive, recurring workflow runs on confidential data — contracts, patient records, financial data. A field’s plant-protection log is not in the same confidentiality category, and we will not dress it up as such just to make our own thesis look better. If you have no such workflow, we will say so up front, before anyone spends anything.

If the pattern sounds familiar

If you have a confidential, recurring workflow you have long wanted to automate — one you would never paste into a public AI — tell us about it. We will say honestly whether it can reach production, and what getting there costs. No pressure, in a confidential conversation.

Get started →