# Your Moat Just Became a Download By Peter Plötner (https://www.towardsjoyandpeace.com/about/) · Published 2026-08-05 Canonical: https://www.towardsjoyandpeace.com/essays/your-moat-became-a-download/ Google's Gemini Robotics 2 can teach a robot a body it has never seen from fewer than 200 examples, turning a skill that once needed a PhD into something close to a download. For technical founders: why a hard skill is a head start and not a moat, why the same cheapening is also your lever to disrupt, and how the real moat splits in two, the company's (what you own that is scarce) and yours (how fast you learn the next hard thing). --- NASA's Robonaut 2, with the Space Shuttle Discovery reflected in its visor. Photo: NASA, public domain. Google can teach a robot a body it has never seen in an afternoon. For a founder, that splits into two questions: what keeps the company hard to copy, and what keeps you hard to copy. By Peter Plötner. Aerospace engineer and Wayfinder Life Coach. More about Peter → Google's newest robot model, Gemini Robotics 2, can take a robot it has never seen and learn to control its whole body from fewer than 200 examples. Getting a machine to run an unfamiliar body was, until recently, a problem you needed a PhD and years to touch. If your startup's edge is a hard skill like that, the ground under it just moved. Whether that is a threat or your best opening depends on what you do next. On July 30, Google's DeepMind team showed Gemini Robotics 2. Hand it a robot that was never in its training data, a new set of arms, legs, and joints, and it adapts to that body with fewer than 200 examples and a few hours of practice. They ran it on a humanoid built by another company, Apptronik, and it walked, crouched, bent down, and handled objects while working through the task in real time. The skill it just made cheap has a name. Getting one robot brain to control a body it has never used, and to walk and handle things at the same time, has been one of the genuinely hard, data-hungry problems in the field. Not the hardest. But hard enough that doing it well was a whole career. I know, because a smaller version of that skill was mine. ## The skill I loved was never the whole game I have a PhD in steering tiny robots through the human body. The problem was navigation: how to guide a micro robot up through the maze of blood vessels in the brain to the exact spot it needed to reach. I spent years on it, deep enough to be one of a small handful of people who really understood it. I loved it. I would not trade those years for anything. There is a specific joy in knowing one hard thing better than almost anyone, and I still chase it. But that skill got me in the door. It was not what made me good once I was inside. The moment the work started to involve other people, leading them, convincing them, judging who to trust, the degree went quiet. None of what mattered then had been on the syllabus. Reading the room. Having the conversation everyone was avoiding. Knowing which problem was the real one and which was just the loudest. I had spent years mastering the kind of robot control a model can now pick up in an afternoon, and almost no time on the skills the work actually turned on. ## The wave that erodes you is the one you ride All of that sounds like a threat. Your hard-won edge is turning into a download, so panic. It is only half the story. The same cheapening that eats your skill is a tool you get to pick up. You do not even have to be first. The founder who puts Gemini Robotics 2 to work can build things that used to need a robotics lab and a team of specialists. Commoditization does not only take a moat away. It hands you a cheaper shovel to go dig somewhere new, often right under a slow incumbent still protecting the old way. But do not chase every new tool. Change is not free. Alex Hormozi's rule of thumb is that a real change sets a team back around a fifth of its performance while everyone adjusts, so a new tool has to clear that cost, and the risk that it does not pan out, before it earns its place. Most do not. They make the product different, not better, and different is just churn. So the question is never “is my skill safe.” It is “would this make us clearly better, by more than the change will cost me to make.” Chase the ones that clear that bar. Leave the rest alone and keep shipping. ## So where is the moat? If no single skill stays rare, hard or soft, then the moat cannot be a skill you own. It splits in two, and it helps to hold them apart. **The company's moat** was never your skill. It sits on the things that do not change. Jeff Bezos likes to say people always ask him what will be different in ten years, but almost never what will stay the same, and the second question is the one you can build on. Customers will always want it faster, cheaper, and with more to choose from. They will always want more of the result they came for, more reason to believe it will actually work, and less time and effort to get there. No model release, from Google or OpenAI or anyone else, touches that. And value does not just float up to whoever owns the smartest software. It goes to whoever owns something scarce. Look at phones. For years, Apple has sold only about one in five of the world's smartphones and still taken home roughly eighty percent of the whole industry's profit. The software layer, Android, is free and runs on most of the rest. The profit went to the company that owns an integrated product and a brand people trust. Your company's moat is whatever you own that a rival cannot download: distribution, a brand, a dataset, real switching costs, a product so woven together it is hard to copy one piece at a time. **The founder's moat** is different, and yes, it is still a skill. Call it a meta-skill, and it has two parts: spotting which hard thing is worth learning next, and then breaking it into a stack you can climb faster than the field does. It is the skill that makes all the others, which is why it does not expire when any single one gets cheap. That is why “I'm just not a people person” is such an expensive sentence. It treats a skill you have not broken down yet as a wall you were born behind. The founder who says it and stops has handed you the game. The one who takes “people person” apart, noticing who goes quiet, saying the hard thing kindly, asking the better question, and then practices, will pass them. Not because that skill is rare, but because the decision to keep learning is. ## Two questions, not one A moat is not a thing you dig once and sit behind. It is two questions you keep asking as the ground moves. What is your company's moat this year? And what is the next skill you will improve? ## Frequently asked questions ### Isn't a hard technical skill still a real advantage? Yes, especially if you use the new tools to build with it. It just is not a moat you can sit behind anymore. Treat it as a head start, not a wall. ### So should I adopt every new AI tool? No. Every change costs you something before it helps, so most are not worth it. Adopt one when it would make the product clearly better in something a customer feels, by more than the disruption costs you. Otherwise leave it alone and keep shipping. ### What actually can't be downloaded? For the company: what you own that is scarce, like a brand, distribution, an integrated product, a dataset, or real switching costs. For you: knowing which hard thing to learn next, and learning it faster than the field once the last one gets cheap. ### Aren't some people just naturally good with people? I don't think so, and it is a costly thing to believe. “Good with people” is a stack of smaller skills you can name and practice: noticing who went quiet, saying the hard thing kindly, asking better questions. Call it a gift and you will never train it. Break it down and you can. ### Does this mean my PhD or deep expertise was wasted? No. Deep expertise gets you in the door and sharpens your judgment for years. It just stops being the thing that sets you apart, not that it was ever worthless. Mine was some of the best years I have had. --- If this hit something, The Problem You Want to Solve Isn't the One in Your Way is a companion piece on finding the constraint that actually binds you. And the Essential Self Diagnostic is fifteen questions, sixty seconds, a quick way to see which of the things you are defending are still worth defending.