# The Recipe Was Public. The Cooking Took 11 Years. By Peter Plötner (https://www.towardsjoyandpeace.com/about/) · Published 2026-08-20 Canonical: https://www.towardsjoyandpeace.com/essays/the-recipe-was-public/ China's LandSpace just landed an orbital-class booster on solid ground, the 2nd company ever to do it after SpaceX. Every step was public: SpaceX livestreamed the maneuver in 2015, the physics is in textbooks, and for years an AI would walk you through it in 10 seconds. It still took 11 years and 1 crash. For technical founders: why the knowledge was never the moat, what AI actually compresses and what it cannot, and why simple and easy are different things. --- The last seconds of the landing, 390 kilometers from where it took off. Every step of this was public for 11 years before anyone else could do it. Photo: LandSpace. A rocket standing upright in the Gobi Desert is quietly good news for anyone worried that AI just erased their moat. By Peter Plötner. Aerospace engineer and Wayfinder Life Coach. More about Peter → AI can explain how to land a rocket in 10 seconds. Doing it took China's LandSpace 11 years and 1 crash. If you believe AI just killed every moat, including the one under your company, look at what happened in the Gobi Desert this week. On August 19, LandSpace launched a rocket called Zhuque-3 from the edge of the Gobi Desert. It is 66 meters of stainless steel and it burns methane. 137 seconds into the flight, the first stage, the big lower half, let go of the rest and began to fall. For almost 6 minutes it fell. Then it relit an engine, snapped out its landing legs, and set itself down, upright, on a pad 390 kilometers from where it started. 8 minutes after liftoff the booster was standing still in the desert, and the upper stage was carrying a satellite to orbit. Before this flight, exactly 1 company in the world had ever landed an orbital-class booster back on solid ground: SpaceX. LandSpace just became the 2nd, and the first from China. Now the strange part. None of it was a secret. SpaceX landed its first booster on December 21, 2015, on a livestream, in front of the whole planet. The physics has been in textbooks far longer than that. The maneuver has been replayed frame by frame on YouTube for a decade. And for the past few years, you could ask an AI to walk you through every step, and it would, politely, in 10 seconds. LandSpace was founded in 2015, the same year that livestream aired. The recipe was public for the company's entire life. The cooking still took 11 years. ## Why didn't the recipe help more? Because the recipe was never the hard part. On paper, landing a booster takes 3 sentences. Turn the falling stage the right way up. Relight the engine. Arrive at the ground and at 0 speed in the same moment. Every rocket engineer on Earth has known those sentences since 2015. Between the sentences and a rocket standing in the desert live all the things the sentences do not say. How the leftover fuel moves when the stage flips over. How an engine behaves when you light it in a fall. What desert wind does in the last 100 meters. None of that is in the recipe. All of it gets a vote. And any single vote can end the flight as a fireball on video, or, as SpaceX likes to call it, a rapid unscheduled disassembly. In December, one of those votes went against LandSpace. On this rocket's first flight, the landing burn went wrong. The combustion turned abnormal during the relight, and the booster crashed at the edge of its own landing pad. 8 months later they flew again, and the same maneuver worked. The company's chief commander says the data from the crash is part of why. The failed attempt was not a detour from the recipe. It was an ingredient. And it is an ingredient you cannot buy, download, or ask for. You can only get it by flying. ## What about AI? You know the 2026 fear. AI writes the code, the deck, the copy. Anything made of information can be copied in an afternoon, so whatever you know, a competitor knows by Friday. That fear is not wrong. When a product is made of information, the recipe and the product are close to the same thing, and a public recipe might mean a copied product. Even there, the copying is not equal. People who know a field deeply pull more out of the same AI than beginners do, because the tool answers the questions you know to ask. Rockets just gave us the cleanest test of the other kind of work. LandSpace had the best conditions any copier ever had: 10 years of public video, published research, a whole industry of suppliers, and lately an AI that answers any question in seconds. The copy still took 11 years and 1 crash. In the physical world, the learning loop does not run through your head. It runs through metal, fire, weather, and an 8-month rebuild. AI shortens the distance between you and many possible answers. It does not shorten the number of times reality gets to test them. Yes, AI will speed up the next company. Better simulations, faster design loops. Say it cuts the journey in half. Half of 11 years is still a 5-year head start. And SpaceX and everyone else at the front get the same tools, so while the follower speeds up, the finish line moves too. ## The 10-second answer and the 11-year skill Your own skills follow the rocket rule, not the download rule. Ask AI how to stay calm when your co-founder is angry. You will get a genuinely good answer in 10 seconds. You will nod. Two days later a real voice rises in a real meeting, and your chest does what it has always done. I know this gap from the water. The recipe for controlling your depth while diving is 1 sentence: breathe in and you rise, breathe out and you sink. I could recite it before my first dive. My body learned it much later, 1 dive at a time, and the water graded every single one. Simple is not the same as easy. Simple means the recipe is short. Easy means you need only a few reps, and each rep costs you little. Landing a rocket is simple, and every rep costs a rocket. Staying calm is simple, and every rep costs a real moment with real stakes. Neither one is easy, and a better recipe will not change that, because the recipe was never the missing piece. ## Count reps, not recipes If you are a founder, this flips the moat question. The old question was: what do we know that nobody else knows? In 2026 the honest answer is usually: nothing, not for long. The better question is: what has reality tested us on 500 times that a copycat has not survived once? The batches shipped. The customers who left, and the reason. The launch that crashed at the edge of the pad, and the fix that flew 8 months later. That ledger is the moat. And it grows at exactly 1 speed: the rate at which you put real things in front of the real world. The same ledger runs on you. Pick the skill you keep meaning to build. The calm. The hard conversation. The clean no. Count the recipes you have collected: the books, the videos, the AI chats. Then count the reps, the times reality got to grade you. If the score is 12 recipes to 0 reps, you do not have a knowledge problem. LandSpace's next step is to fly a landed booster a second time. Yours is closer. 1 skill, 1 real rep, this week. Which skill have you been studying instead of flying? ## Frequently asked questions ### What did LandSpace actually do on August 19, 2026? LandSpace, a private Chinese rocket company, launched its Zhuque-3 rocket from the Gobi Desert, put the Honghu-03 satellite into orbit, and landed the first stage upright on legs, 390 kilometers away, on land. That made LandSpace the 2nd company in the world, after SpaceX, to land an orbital-class booster on solid ground, and the first Chinese commercial company to recover one. ### SpaceX showed how to land a booster in 2015. Why did copying take almost 11 years? Because the public part, the recipe, was never the bottleneck. The slow part is the learning loop that runs through the real world: build, fly, break, fix, fly again. LandSpace's first landing attempt crashed in December 2025, and the company says data from that crash helped the next one succeed. That kind of ingredient cannot be downloaded. It has to be earned in flight. ### Doesn't AI make copying much faster now? For anything made of information, yes. Code, text, and designs copy almost for free, because there the recipe and the product are the same thing. AI also speeds up the research and simulation around physical work. What it does not compress is the number of real-world tests a physical system, or a person, has to survive. Even if AI cuts a follower's path in half, half of a decade is still a 5-year head start, and the leaders are using the same tools to move further ahead. ### What is a moat made of reps? It is the advantage stored in everything reality has already tested you on: products shipped, failures fixed, hard conversations survived. A competitor can copy your plan in a day. They cannot copy your ledger of survived tests. They have to run every test themselves, in real time. ### How do I use this for my own skills? Pick 1 skill that matters to you and count 2 numbers honestly: how many recipes you have collected (books, videos, AI answers), and how many reps you have done where reality could grade you. If the recipes are winning, schedule 1 small rep this week. The recipe is not the missing piece. Reps are. --- One thing that closes soon. I just earned my certification as a Wayfinder coach, and to mark it I am giving away my full 3 month coaching package to 1 person, free. It is the same package paying clients get: a session every 2 weeks, support in between by message or voice, and the method itself, so you can run it again on your own later. No homework. I read every application myself and they close on September 20. If you have been collecting recipes for a skill you keep meaning to build, the application is here and takes about 5 minutes. The companion pieces are Your Moat Just Became a Download on the kind of advantage that does copy in an afternoon, and Fail Faster to Get There Sooner on making the reps cheap enough to run often. If you want a first reading on where your own reps are missing, the Essential Self Diagnostic is 15 questions and takes about a minute.