Build Something You Can Touch

Software moats collapsed. Physical ones didn’t.

Knowledge is getting close to free. You can read every research paper in a field, synthesise the findings, build a working prototype, and test it against generated data, all in a few days. Five years ago, entering a domain you didn’t already know meant hiring experts or spending months on the basics. Now you can turn an entire discipline into working code in a week.

Understanding it though, is a different story. Your code will work. The tests will pass. The logic is sound. But if you’re able to fight your Dunning-Krüger syndrome, you’re likely not sure whether the assumptions underneath it are correct, because you don’t have the gut feeling that a domain expert would. You built something that works perfectly, assuming your assumptions are correct. Assumption being the key word.

The only test that matters is physical

At my startup over ten years ago, we were building smart home technology with sensors, actuators, and low-power radios, and but we didn’t really know the radio technology part of the stack. So we tested things by brute force.

We lit actual fires to test smoke detectors. I once walked a full kilometre from a gateway holding a Z-Wave sensor, checking the connection every ten metres, because we wanted to double check our implementation and the only way to verify was to physically walk until the signal dropped. The max range on a Z-Wave sensor is about 100 metres in open air. It took a kilometre before we were sure there was a bug. Human-written code, highly skilled developer. When you don’t quite trust your own understanding, you need to see it fail in the world.

How do you know the assumptions are right?

You can get another LLM to verify your code does what it claims to do. That’s mechanical. The harder question is whether you built the right thing in the first place, whether the rules you extracted miss something practitioners know by instinct.

Physical testing closes that gap. You put a sensor in the ground, compare the data to what the literature predicted, and reality tells you what the papers left out. Not because the papers are wrong, but because the physical world has more noise than any document captures. The sensors tell the truth and battle test your assumptions.

The same AI that lets you synthesise a domain in weeks also means you can enter real-world domains that were previously hard to access. The knowledge barrier dropped. The hardware was already there, waiting. Industrial-grade sensors that can sit in a field for years are available off the shelf. The software to process the data is the easy part now.

Where this becomes an opportunity

The pattern repeats across a dozen verticals: the enterprise version exists, but the SMB one doesn’t. Precision agriculture has soil analysis for large commercial operations, but a family farm running 50 acres has nothing between a handheld moisture metre and a six-figure system. Water quality and cistern monitoring is the same gap. Waste bin fill-level monitoring, where the sensor is simple and the business model is obvious. Small-scale fleet tracking for someone running three service vans, not three hundred. Niche operations like swiftlet nest farming, where the monitoring needs are so specific that no enterprise vendor will ever build for it.

The domain expertise you’re missing is probably one conversation away. Does your neighbour make wine? Fermentation monitoring. Does your friend manage rental properties? Building moisture and energy tracking. Someone you know runs a marina? Water quality. A veterinary clinic, a small brewery, a poultry farm. The gut feeling you can’t synthesise from research papers is sitting in someone else’s head, and you’re the one who can build the thing they’ve been wishing existed.

Hardware costs money, and you’re investing before you have customers. That’s the honest difference between this and shipping another SaaS wrapper. But that cost is also the moat. Every dollar you spend is a dollar the next builder has to spend too, and most of them won’t. A million dollars a year in a physical niche is a great business for you, and it’s not worth a meeting at Google.

Most builders will keep focusing on software because it’s cheaper and easier to start. That’s the opportunity. If you’re wondering where the interesting problems went, they moved outside.