Precision agriculture AI: Saepum, Taranis and John Deere by the numbers

Will Drewes

Founder, Fern Strategy · LinkedIn · 2 min read

The field results are coming in

Precision agriculture has been a "coming soon" story for years. The hardware ran ahead of the software and the software ran ahead of the outcomes. A few named companies now have numbers.

Saepum: disease detection at scale

Saepum, a crop-diagnosis platform, reports a 95% detection rate for crop diseases, pests and abnormal signs across roughly 8,000 farms and 200,000 hectares, and 18% higher cultivation volume alongside it.

Most field scouting is periodic and sampled. A system catching 95% of anomalies continuously across 200,000 hectares feeds a farm's decision cycle with a different kind of input, and the yield lift follows from earlier, more accurate intervention.

Taranis: targeted application instead of blanket spraying

Taranis reports that AI-guided drone monitoring let farms cut chemical fertilizer application by more than 52% compared with uniform application. High-resolution early detection lets operators treat specific zones.

Input costs drop and runoff liability drops with them. Where agrochemical rules are tightening, the record of what was applied, where and why becomes a compliance asset.

John Deere: AI in 60% of new equipment

John Deere announced in 2025 that over 60% of its new equipment would embed AI for precision farming tasks, with machine vision and real-time analytics across planting, spraying and harvesting.

That is an adoption number rather than a farm outcome, and it means farmers buying new equipment will get AI recommendations whether they asked for them or not. Few operations have answered who reviews the recommendation, who overrides it, and what gets logged when the system is wrong.

What this means if you farm or advise farmers

The Saepum and Taranis results come from the same setup: stable agronomic rules, messy field inputs. Disease criteria and application thresholds change slowly. Reading a field accurately at scale and speed is the work, and that is what the models are doing.

Deere's fleet numbers add a layer. Recommendations embedded in the equipment are harder to audit after the fact than a separate analytics tool, and tracing a bad outcome back to a specific model decision gets murkier.

Before committing to a build or a buy, ask whether you can see what the system decided, why, and what happened next. If you can't, someone else's 52% input reduction and 18% yield lift won't transfer to your operation in any way you can verify.