What Growers Actually Need From AI: Field Notes After WCCBE Frankfurt
The Short Version
I spent two days in May chairing and speaking at World Class Cannabis Business Europe (WCCBE) in Frankfurt, on the subject of visual AI for cannabis cultivation. The biggest takeaway wasn't about the technology. It was about the gap between what the AI industry likes to talk about – autonomy, end-to-end automation, “AI runs the grow” – and what growers in the room actually asked for, which was something much more grounded. They want a tool that helps them decide, not one that decides for them. They want it to be specifically good at plants, not generically good at everything. Here's what I heard, and what it means for where PlantLab goes next.
The setting, and an honest mismatch
WCCBE Frankfurt, May 19-20, 2026. I chaired a day and gave a talk on visual AI for cannabis – the kind of plant-by-plant diagnosis problem PlantLab works on.
The first honest thing to say is that a deep cultivation-diagnosis talk is not the same animal as a policy, regulation, and investment talk, and a conference that spans all of those is going to have rooms with very different centers of gravity. The business and policy sessions are genuinely useful for credibility and for understanding where the industry is heading. But the product conversation – the one where someone tells you what they'd actually use on a Tuesday in their grow room – happens with operators, not in the M&A track. That's not a criticism of the event. It's a note to myself about which rooms the product story belongs in.
With that said, the technical lessons that came out of the operator conversations were sharp enough to be worth writing down.
Lesson 1: Growers want a co-pilot before they want an autopilot
The pitch the AI industry loves is autonomy: the system watches, decides, and acts, and the human steps back. Almost nobody I talked to wanted that as the first step.
What they wanted was a tool that makes their own judgment faster and more confident. Flag the plant that looks off. Tell me what you think it is and how sure you are. Then let me decide. The grower stays in the loop because the grower is accountable for the crop, and because they have context the camera doesn't – what they fed it last week, what the room did overnight, what this strain always does at this stage.
This maps cleanly onto a design choice: the most valuable output isn't a decision, it's a well-calibrated suggestion plus an honest signal of how much to trust it. Autonomy can come later, on the narrow slices where the tool has earned it. Human-in-the-loop is not a stepping stone to be skipped. For most growers it's the actual product.
Lesson 2: Generic AI keeps disappointing growers
This came up again and again, usually as a half-frustrated aside: people have tried asking a general-purpose AI assistant what's wrong with their plant, and it gives them an answer that sounds authoritative and is often wrong, with no signal that it might be wrong.
This is the wedge, and it's a fair one to talk about publicly because it's about outcomes, not methods. A general model trained on the whole internet knows a little about cannabis among a billion other things. It will confidently call magnesium deficiency on something that's actually a pH problem, because the visual distinctions between plant health issues are subtle and the general model never had to get them right. Worse, it presents every answer with the same smooth confidence, so the grower has no way to tell the good answers from the bad ones.
A purpose-built tool earns its place precisely here: it's narrow on purpose, it's measured on how well it does the specific thing, and it can tell you when it's unsure. “Specifically good and honest about its limits” beats “generally capable and uniformly confident” for anyone making a real cultivation decision.
Lesson 3: The near-term home for visual diagnosis is controlled indoor cultivation
Visual AI diagnosis works best where the inputs are consistent: stable lighting, repeatable camera angles, a known set of plants, a controlled environment. That description is controlled indoor cultivation, and it's also where the people most willing to adopt new tooling tend to be.
Outdoor and greenhouse settings introduce variability – weather, mixed lighting, scale – that makes consistent visual diagnosis harder. None of that is unsolvable, but it's the second problem, not the first. The honest near-term answer is that indoor, controlled grows are where this technology delivers reliable value today, and that's where the product should aim first.
Lesson 4: The automation companies need plant-state signals, not another app
A recurring theme in the hardware and automation conversations: the companies building controllers, sensors, and grow-room automation don't need another consumer-facing diagnosis app. They need a plant-state signal they can pull into what they already build.
That's a different shape of product. It says the diagnosis capability should be available as something an automation system can consume – a clean signal of plant condition and how trustworthy it is – rather than only as an app a human opens. The growers running those rooms have already chosen their controllers and their dashboards. The useful move is to feed plant-state into those systems, not to ask everyone to adopt yet another screen.
Where this points PlantLab
Four lessons, one direction. Take the product to the rooms where operators are, not only the rooms where the industry talks about itself. Keep the diagnosis specifically good and honest about its uncertainty, because that's the thing generic AI can't do. Aim first at controlled indoor cultivation, where visual diagnosis is reliable today. And make plant-state available to the automation systems growers already run, instead of competing for their attention with another standalone app.
None of these are surprising in hindsight. That's usually the sign of a good conference – it doesn't hand you a new idea so much as it sharpens the ones you arrived with and tells you which were wishful thinking. The wishful one was autonomy-first. The sharpened one was that being narrowly excellent and honest about it is the whole game.
PlantLab is free to try at plantlab.ai. Three diagnoses a day, results in milliseconds. If you build grow-room automation and want a plant-state signal to integrate, the API documentation lives at plantlab.ai/docs.
FAQ
What is WCCBE?
World Class Cannabis Business Europe (WCCBE), held in Frankfurt in May 2026, covering cultivation, policy, regulation, and business across the European cannabis sector. I chaired a day and spoke on visual AI for cannabis cultivation.
Why not just use ChatGPT to diagnose my plants?
A general-purpose AI knows a little about cannabis among everything else, and presents wrong answers with the same confidence as right ones. A purpose-built diagnosis tool is narrow on purpose, measured on how well it does the specific task, and can signal when it's unsure – which is exactly what a general model can't do.
Does PlantLab automate my grow room?
PlantLab provides the diagnosis and a trust signal; what you do with it is your call. Most growers want a tool that informs their decision rather than one that acts for them, and the API is built to support that human-in-the-loop use first. It can also feed plant-state into automation systems for the cases where automatic action makes sense.
Related reading: – What's Wrong With My Cannabis Plant? A Visual Diagnosis Guide – The grower-facing diagnostic hub – Confidence Is Not Reliability: Trust Signals for Automated Plant Diagnosis – How to know when to trust an automated answer – Build an Autonomous Plant Health Monitor with AI + Home Assistant – Feeding plant-state into automation