Taranis Review

Crop scouting has always been a sampling problem dressed up as an inspection. An agronomist walks a field, looks at a handful of plants, and extrapolates from them to several hundred acres.

That works remarkably well and fails in a specific way: the problem is wherever you did not walk. A weed escape, an insect pressure point or a nutrient deficiency in the corner of a field is invisible until it is large enough to see from the road, by which point it has cost you.

Taranis replaces the sample with something closer to a census — imagery at leaf level across every acre — which is a different activity rather than a faster version of the same one.

Taranis: Products And Reach

What Does Submillimetre Crop Imagery Show You?

Taranis describes submillimetre image resolution of every acre, with imagery clear enough to count the spots on a ladybird, used to detect stand counts, weeds, insect damage, nutrient deficiencies, disease pressure and general field health.

The resolution claim is what makes the detection list possible, and the connection is worth spelling out. Satellite imagery shows you that an area of a field is stressed; it cannot tell you why, because the pixel is larger than the symptom.

At leaf level the symptom itself is visible. Insect damage has a characteristic appearance, nutrient deficiencies show as specific patterns of discolouration, and a weed is identifiable as a species rather than as an anomaly. That is the difference between knowing something is wrong and knowing what is wrong.

Stand counts are the least glamorous item on the list and possibly the most valuable. Knowing how many plants actually emerged, precisely, across an entire field, informs every subsequent decision — and counting by hand is a sampling exercise with all the same limitations as scouting.

The honest boundary is that detection is not diagnosis. An image showing damage tells you damage exists and what it looks like; deciding what caused it, whether it is spreading, and whether the economics justify treating it remains an agronomist's judgement, informed by things no camera records.

The greenhouse version of the same ambition works from sensors instead. Source.ag can measure a contained crop directly, which is why it forecasts where this one observes.

Satellite imagery tells you an area is stressed. Leaf-level imagery tells you why, because the pixel is finally smaller than the symptom.

Yield Impact Ties Images To Bushels

Taranis Yield Impact converts leaf-level insights into bushels, and this is the most commercially intelligent thing in the product.

The reason is that detection alone does not drive action. An agronomist told there is insect pressure in a section of a field still has to decide whether it is worth treating, and that decision is economic rather than biological: what does the damage cost, and what does the treatment cost.

Expressing findings in bushels answers the first half in the unit the industry actually uses. A grower who knows a problem is costing a specific yield across specific acres can weigh that against an input cost immediately, without an intermediate translation step.

It also changes the sales conversation for the platform itself. Software justified by capability has to be believed; software expressed in bushels can be checked against the season, which is a far stronger position.

The caveat is that a yield impact figure is a model output, and its accuracy is unpublished. It is an estimate built on assumptions about how a given problem affects a given crop under given conditions — useful for prioritising between problems, and not a number to treat as measured.

Ag Assistant And The Agronomy Question

Taranis describes Ag Assistant as the agriculture industry's first generative AI-powered agronomy engine, which is a substantial claim and worth reading carefully.

What a generative layer plausibly adds is interpretation. The platform generates an enormous volume of detections across many fields, and turning that into a prioritised set of recommendations — this field first, this problem matters, this one does not — is exactly the summarising and reasoning task these models handle well.

It also addresses a real bottleneck. Agronomic expertise is scarce, advisors are stretched across more acres than they can properly consider, and a tool that drafts the analysis leaves the human to check and decide rather than to compile.

The claim of being first is unverifiable and largely beside the point. What matters is whether the recommendations are good, and that depends on whether the model has agronomic knowledge grounded in something or is producing plausible-sounding advice from general text.

The question to put to the vendor is therefore simple and specific: what is Ag Assistant grounded in? Trial data, extension research, agronomist review — any of those would make the output defensible. Without an answer, a recommendation about a crop worth six figures is a recommendation you should verify yourself.

This is the pattern behind every image-recognition product that has to be trusted commercially — Restb.ai faces the identical question about property photographs, and the answer is always what the model was grounded in.

• Confirm your crops and region are covered before anything else, since the site does not state them.
• Treat a Yield Impact figure as a prioritisation tool, not a measured loss.
• Ask what Ag Assistant's recommendations are grounded in.
• Verify a sample of detections on the ground before trusting the whole set.
• Judge the platform on problems found early, not on total detections.

Running A Season With Leaf-Level Scouting

The value of aerial scouting is early detection, which means the evaluation has to run across a season rather than a demonstration.

Ground-truthing in the first weeks is what tells you whether to trust anything later.

1. Confirm coverage for your crops and your region before signing anything, since neither is stated publicly.
2. In the first flight, walk to five detections yourself and confirm what the imagery said is actually there.
3. Walk to two areas the system reported as healthy, since false negatives matter as much as false positives.
4. Compare Yield Impact figures against what you would have estimated, and note where they diverge.
5. Track how early a problem was flagged versus when you would have found it — that gap is the product's actual value.
6. At season end, compare treated and untreated areas where you acted on a detection, which is the only real measurement.

Who Buys Crop Intelligence, And Why

Taranis positions itself for advisors and growers, and those are meaningfully different customers with different reasons to buy.

For a grower the case is direct: find problems earlier, treat them more precisely, lose less yield. The value is measured against their own fields and their own season, and it is easy to assess after the fact.

For an advisor the case is leverage. An agronomist covering a large territory cannot physically walk every field often enough, and aerial coverage lets them prioritise where to be — which turns their scarce time into a resource allocated on evidence rather than on a rota.

The advisor market is also why the platform matters commercially. One advisor brings many growers, which is a far more efficient distribution model than selling farm by farm, and it explains the emphasis on dashboards and reporting alongside detection.

The gap in what is published is coverage, and it is a significant one. The pages we read name no crops and no regions, which are the first two questions any grower asks. The offices listed in the US and Israel suggest where the company operates rather than where the models are trained, and those are not the same thing.

The business side of a farm sits outside all of it. A ledger platform such as Puzzle handles the accounting that a saved crop eventually shows up in.

Taranis: Leaf-Level, Not Field-Level

• Submillimetre image resolution of every acre — the site's claim is imagery clear enough to count the spots on a ladybird.
• Detection across five problem classes: stand counts, weeds, insect damage, nutrient deficiencies and disease pressure.
• Taranis Yield Impact converts those findings into bushels, which is the number that decides whether anyone acts.
• Ag Assistant is described as the industry's first generative AI agronomy engine.
• Built for advisors as well as growers, with a web dashboard and the Taranis Connect mobile app for iOS and Android.