On the 8th of May 2026, a closed-door conversation in Zürich brought together deep-tech investors, AI founders, and operators building for the physical economy. Louis Millon of Systemiq Capital hosted, and the room produced one of the most commercially useful diagnoses of Edition I.
One founder said that if everything in your pitch and on your website is AI, it is like building a company around the fact that you breathe. Nobody disagreed. Not the investors, not the other founders, not the operators who buy this stuff and deploy it into factories and fields. Physical customers do not care if it is AI or not, he said. They have a problem. The AI is plumbing. Nobody buys a house because they are excited about the pipes, and nobody in the room needed that explained to them.
What follows is what the room produced. The named companies and figures come from research conducted afterwards to test and extend what was said. Where a claim is attributed to "a participant" or "the room," it was said in the conversation. Everything else is verified independently.
The wrong pitch
There is a particular kind of pitch that the room had all seen too many times. Every slide says AI. The website says AI. The product name says AI. The founder cannot get three sentences into a description of the company without invoking the model, the training data, the intelligence.
The founder who made the breathing comparison was making a diagnostic point about focus rather than a stylistic point about marketing. When a founder leads with the AI, it usually means they have fallen in love with the capability rather than the problem. The room's observation was that the companies gaining traction in industrial environments lead with a named, quantified problem, and the technology is incidental to the sale.
The room's test for whether this has been achieved was simple: if a customer buys something and only later realises it was powered by AI, the product worked.
The customer in the physical economy does not care that you breathe. They care whether the thing works, whether it saves them money, and whether it survives contact with a real operating environment.
Sector depth beats technology breadth
That led to a second argument in the room. The instinct of many AI founders is to build a horizontal platform, a capability that can be applied across many industries, because that is the story that promises the largest market. The room's experience was that this instinct is usually wrong for the physical economy.
The obvious objection is that the most valuable AI companies in the world are horizontal. OpenAI and Anthropic sell a general capability to everyone, and it is working. The room's answer was that the line becomes harder to cross when software meets atoms. Horizontal products thrive where deployment is repeatable and the cost of trying them is low. In the physical economy, deployment often depends on specific equipment, workflows, safety constraints and failure modes, and every failure is expensive. The more expensive the deployment and the failure, the more sector knowledge becomes part of the product.
Some in the room questioned whether "physical AI" is even a useful category. What determines whether a company breaks through is the depth of the founding team's understanding of the specific sector they are selling into, whether that is port operations, agricultural machinery, automotive manufacturing, or industrial inspection. The buyers in each of these worlds have different procurement processes, different definitions of risk, and different relationships with data. The room's experience was that founders who had lived inside an industry were better equipped to navigate its buyers, procurement processes and definitions of risk than a technically superior but sector-agnostic team. Technical capability alone is becoming easier to replicate. Knowledge of how to make it work inside a specific operating environment is harder to copy.
The room's phrase for the shortcut to this was customer obsession, which it distinguished sharply from customer research. Sales cycles in industrial environments are structurally longer than in software, and the founders who compress them are rarely the ones with the best technology. They are the ones who spend disproportionate time with customers before building, listening for the problem rather than pitching the solution. One participant put it plainly: customers will always describe their ideal solution as a better version of what they already use. The job of the founder is to hear the problem underneath that, and solve it in a way the customer would never have imagined asking for.
Two companies that hid the technology
The first is bolted to the back of a tractor. In 2017, John Deere paid $305 million for Blue River Technology, a Silicon Valley startup that had built a computer-vision system to tell a weed from a crop. Deere did not market the result as artificial intelligence. It marketed it as See and Spray, a sprayer that puts herbicide only where the weeds actually are. Boom-mounted cameras scan the field at speed, onboard processors decide in milliseconds whether each patch is crop or weed, and individual nozzles fire accordingly.
And to the farmer, none of that is the point. The farmer sees one thing: in the 2025 season, according to John Deere, See and Spray was used across more than five million acres, cutting non-residual herbicide use by an average of nearly 50% and saving around 31 million gallons of chemical, while company-sponsored field trials across seven US states showed a yield increase of roughly two bushels per acre. The chemical bill roughly halved, and the yield went up. The farmer can put both numbers on a spreadsheet and calculate the payback.