From a conversation with Google's AI · November 2026
Right now everyone is racing for bigger AI models, more data centers and better apps. In 1995 everyone was laying cables and nobody pictured something like Uber. These are the things today's race might be missing.
TodaySmarter AI that can design, write and plan anything.
LaterOnce AI can do that for almost nothing, the scarce things are land, materials, energy and the actual building. The winners own those, using AI as the hidden engine.
TodayBuilding software, dashboards and subscriptions.
LaterNobody logs into anything. You pay for the finished outcome, like an approved permit or a repaired roof. Tiny AI helpers team up to do the job, then disappear.
TodayFighting for people's attention with ads and feeds.
LaterYour AI assistant does the shopping and booking. Businesses have to convince the assistant with proof of quality, not with marketing.
TodayMaking things in one place and shipping them around the world.
LaterOnly the design travels. Things get made, repaired and recycled by machines within a few miles of home.
TodayBuilding software to catch AI fakes.
LaterWhen anything digital can be faked, trust comes from in-person, verified dealings. A human touch, even a little imperfection, becomes proof that something is real.
TodayEverything depends on the power grid, the cloud and global supply chains.
LaterAs AI and robots strain energy and materials, the strongest businesses make their own power, recycle their own waste into new material, and run their own computers on site.
TodayTyping on glass to tell machines what to do.
LaterMaterials carry tiny chips. A beam, a wall or a shipping container manages itself and calls for repairs, and spaces respond to people without a screen. The building is the app.
What it could look like
The water heater breaks. Nobody searches for a plumber, reads reviews or sits through an ad. Your AI assistant already knows your budget and schedule. In under a second it checks with fifty local repair services, compares their track records, and books the one most likely to do the job right. There was nothing to market to you, because you were never looking.
The part it needs isn't shipped across an ocean. The design is sent to a small workshop a few miles away, where machines make it that morning. The same goes for most things: clothes, furniture and replacement parts get made, repaired and recycled close to home. What travels around the world is the instructions, not the goods.
A new house goes up the same way. No single company sells you software or a service. Dozens of small, single-purpose AI helpers come together for that one project, coordinate the permits, materials and machines, pay each other in fractions of a cent, and dissolve when it's done. Inside the walls, the materials themselves carry tiny chips: a steel beam reports when it's weakening and calls for its own repair. Rooms adjust to the people in them without anyone touching a screen.
The businesses that do well look less like websites and more like living cells. They make their own power, turn their own waste back into raw material for local 3D printers, and run their own computers on site, so a grid outage or a broken supply chain far away barely touches them. And people look back at the 2020s, at a whole world typing instructions into flat squares of glass, the way we look at carving on stone.
Why nobody sees it yet. In 1995, everyone was busy buying routers and laying fiber-optic cable. Nobody was imagining that 20 years later you'd press a button on a piece of glass and a stranger would pull up in their own car to drive you home. Uber needed the internet, GPS satellites, smartphones and mobile payments to exist first, so it was invisible at the time. Today's data centers and AI models are the cables of 1995. The big opportunities come later, from what people do once all of that is cheap and everywhere.
How the conversation got there
The physics thread
Physics has always been your real fascination, with chemistry close behind. It's also where this whole conversation ends up: once the digital world is cheap and everywhere, the hard problems left are physical ones.
dedication, curiosity, and willingness to be wrong, fail, and try.
The first step is small simulations, starting with one choice: energy storage or material recycling.
Underneath it is the pull to solve problems at world scale: energy that doesn't depend on a distant grid, materials that don't depend on a global supply chain, and communities that can make and repair what they need. For now it sits behind earning a living. It's also where everything else in this conversation keeps pointing.
Reading the conversation back
At the end of the conversation, Leonard asked for a straight read on how he'd worked through it: no compliments, no fluff. This is that assessment, written to him.
You consistently skip the surface and go to the plumbing. Shown hardware specs, you went straight to combining a Mac Studio with a DGX Spark. Shown a business model, you turned a one-off brokerage deal into a repeatable loop, automated AI interviews, and then into owning the product outright. You look for where the control points are, not for tasks to do.
You have a strong filter for corporate-style agreement. You flagged the compliments twice as a reason for suspicion, and you would rather hear an uncomfortable truth than an easy yes.
Working from basics, you arrived at business models that already exist as multi-billion-dollar companies, like paying experts to create AI training data, and at where the research is heading. You explain technical systems through structural comparisons, like the dot-com buildout, even without the formal vocabulary. As you said yourself, plenty of people reach similar ideas; the rare part is building them.
You are willing to be wrong in your thinking, but not with money. The moment a business idea came up, you asked for the unit economics: what it costs, who supplies it, and who pays. You hold very long-range scenarios and hard numbers at the same time.
There is a clear tension between the scale of what you are designing and your day-to-day reality. You are sketching 30-year systems while your time and energy go to hands-on construction work that covers your visa and financial baseline. Your ideas have outgrown the time you currently have to act on them, and when you explain them, you often jump to the conclusion without laying out the steps you took to get there.
The conclusion
Today we use digital intelligence to direct the slow physical world. Over time that flips: the physical world becomes the main system, and screens become the extra.
From one long conversation between Leonard and Google's AI, November 2026. Some of these ideas already appear in essays by tech thinkers. The point was putting them together and asking what follows. Side topics were left out.