--- title: "Building for fun" url: "https://writing.colivetree.com/3/product/55/building-for-fun" --- Originally published Mar 1, 2026 ## A familiar question What should you build when everything is possible? How do you build when everything is possible? > In another moment down went Alice after it, never once considering how in the world she was to get out again. People today are throwing _vibe coding_ around still like it's a dirty word. It was actually not meant as such by Karpathy when he coined it - he, himself, one of the great minds behind the current technological revolution. It was meant to be described as fun, light, somewhat careless coding. A return to having fun with code rather than spending weeks planning something that most times would turn out inconsequential. Either because it didn't work, or because nobody wanted it. LLMs brought vibing back, and it _horseshoed_ in two groups of very different people who were aching for a simpler time. On one end of the spectrum, those for whom coding was like dark magic. Impossible to approach, yet mysteriously powerful. All of a sudden someone had invented a way to cross the threshold and a gleaming light revealed itself. You just had to talk to it and there it was, invented. Very quickly companies were being born, bought, pivoted and scaled. Cursor, Windsurf, Cline, Lovable, Replit. Come follow us down the rabbit hole, and we'll make your wildest software dreams come true. These companies often struggled with their ICP (Insane Clown Posse as a friend calls it). Who is it we're going after? The wide-eyed amateur, the professional dev wanting to 10x? The Technical PM? The designer wanting to push code? The Engineering Manager / Architect who's been craving to get their hands dirty? The horseshoe can fit everyone. ## An unfamiliar feeling For me, things felt very different immediately. Throughout uni, I had struggled with how much time I'd spent browsing shitty docs for systems I knew I would likely never have to or want to operate again. Oracle whatever the hell, Sun Microsystems go figure, holy shit the late nights fighting shit when even stackoverflow wasn't there to help (and what a blessing it was when SO started to come alive with people). Things had gotten drastically easier over time, but with ease came creeping complexity. Frameworks, abstractions, Virtual DOMs, Map-Reduce, Spark and Kubernetes, paradox of choice, millions of technological camps battling each other for relevance, trashing each other's solution for views. How is anyone supposed to get anything up and running these days? There was some twitter debate (aka flame wars), mostly triggered by the unparalleled indie success of Pieter Levels, with one camp arguing for the status quo. The level of abstraction and the prepared-for-everything attitude of the modern toolkit on the one hand. The just KISS, I run everything in a single server and mostly one file, attitude of Levels on the other. Both sides looked completely intractable to the other. Then, things started to change. The next big thing in technology came, and it started off looking like a toy. ## Everything looks exactly the same until it looks completely different ChatGPT is one of the most successful consumer product of all time. Its overnight success was unprecedented and just an acknowledgement of how the interface was perfectly right. For the time, for the need, for the use case. What if whenever you wanted to know something (ANYTHING!) you could just... ask? Of course, that's when the word hallucination penetrated the zeitgeist. It sounded plausible confident even when it was perfectly wrong, it kept its composure when corrected on critical details and was perfectly content with changing its mind on a whim if you directed it that way. People started testing it for work, for play, for coding, for research. It wasn't perfect, but it was fun. Especially if you didn't stretch it too far. Some of us tried to push it. We knew there was a point at which it broke, but we also knew we could design strong enough harnesses around it that we could make things others deemed impossible given the state of technology. Certainly, there was something of a leap of faith. We trusted models would get better, we had seen them get a little bit better with every release, and we thought this could go all the way. As with every bit of emerging tech, things seemed to plateau for a while. That made the non-believers trust they were right. See?, they pointed nonchalantly. Told you it was a toy. Maybe it was, everyone else doubted. Maybe this is as good as it gets. Even if it is, we're more productive than we've ever been so that's not bad. Yet, imagine if it isn't? Imagine this keeps getting better and better, and better. We're at a point now where things have gotten A LOT better. There was certainly an inflection point at the end of 2025 where models started to code well, spend unreasonable amounts of time on tasks, get a lot of hard things right, plan and do better with uncertainty and a higher degree of freedom and act more responsibly with destructive behavior. It still feels like early days, but the few people I've seen get productive with the tools now ship more than 20 / 30 people, and even certain 100 people teams. They ship usable software, more ambitious ideas, crazy high fidelity designs and design systems, complex hard-to-orchestrate systems. The infrastructure will only get better, and with each iteration, so does everything that uses it. Knowledge work and then all work, is on the line. I don't know where this is all going, and anyone that tells you they do - is bullshitting you. Our guesses are as good as the future says they will be, and the self-fulfilling prophecies they contain will often be true, but only to some extent. What I do know is that we need to keep building. If we build great things, and we build with genuine care for others, the future will be better off for them. Plus, it's great fun.