Everyone is building AI for screens. Chat interfaces, dashboards, copilots embedded in software. But I keep thinking about what happens when AI meets the physical world: not robots in a warehouse, but platforms that coordinate real things moving between real people.
Blood donation is one of these. You have a perishable biological product with a 42-day shelf life. You have unpredictable supply driven by human willingness. You have demand that spikes without warning: a car accident, a surgery, a natural disaster. And you have a logistics chain that still relies heavily on phone calls, spreadsheets, and gut instinct.
This is where AI gets interesting to me. Not as a chatbot that answers donor questions, but as the coordination layer underneath. Predicting which blood types will be short next Tuesday. Matching that forecast to donors nearby who are eligible and likely to respond. Routing collection to the locations where it has the most impact. Adjusting in real time when a hospital's needs change.
I think of this as the physical AI platform model: AI that doesn't replace a person or a screen, but orchestrates a system that exists in the real world. The same pattern applies to food banks, organ matching, disaster response, even local energy grids. Anywhere supply is human, demand is urgent, and the thing being moved has an expiration date.
The software-only AI companies will get funded first. They always do. But the ones that figure out how to coordinate atoms, not just bits, will build something much harder to replicate.
The moat isn't the model. It's the mess. And the mess is where the real problems live.