2025: The year I reinvented myself

Iterations

I have been a certain way for as long as I can remember—or at least for the past three years: a purist of sorts and a self-proclaimed perfectionist. 2025 threw me into such chaos that I had to look within myself more deeply than ever before.

I realised that the perfection and quality I admire in my idols aren't the result of some internal drive. They're the result of iteration. The first Logan Paul video wasn't as entertaining, the first Apple product wasn't perfect, and the first version of Linear wasn't this intuitive. They got there through iteration.

In my pursuit of these perfect things, I had become so bottlenecked that my rate of iteration had slowed to almost zero. I wouldn't do anything until I knew the output could withstand intense scrutiny.

I have now learned that I need to build the muscle required for anything I want to get good at. To do that, I must go through the process again and again until it becomes simple enough that I naturally take on the challenge of levelling up.

Advice

I stopped taking raw advice and started thinking for myself. I used to take everything people said to me as advice tailored to what I needed. Looking back, it was naive to think that people understood enough about my situation to give the right advice, or even knew enough about the subject matter to determine what that advice was.

Oddly, along the way, I also seem to have encountered a few snake oil salesmen. I'm much happier now that I treat someone's advice as an account of their experience—a mere data point in my decision-making, weighted by their relevance to me and the subject matter.

Startups

I took a step back from active startup involvement this year. After Mars closed in April following an acqui-hire—which went terribly, for what it's worth—I decided not to jump into anything and instead take life one week at a time. I worked on projects that interested me in the moment and took time to learn some of the theory behind them.

As a kid, I learned by trying, studying the theory, building depth, and then spreading wide. I think I'm now in the depth-building phase, learning how software companies are built. Luckily, I now have tools that let me build any software I want within minutes.

I can't begin to tell you how much Claude Code has accelerated my work and how the bottleneck between thinking of an idea and implementing it has shrunk to almost nothing.

This now gives me the opportunity to concentrate more on the next aspect: distribution. I have fallen short in this area, and I am actively working to improve. I think I got lucky with Mars: because we couldn't accommodate more than three or four companies in a batch, I never really understood my limitations here. Distributing the software I make and earning money from it will be at the top of my 2026 agenda.

Picking a niche

I have changed my mind on this topic, oddly enough, after listening to the Dwarkesh Podcast and using Linear and Greptile. All three offer purpose-built experiences, and they have taught me that I should pick a niche and go to great lengths to serve it.

There's very little point in being generic in the domains I want to conquer. There is an idea going around that niche experiences will no longer exist once AI software becomes prolific, but I don't believe that the majority of humankind knows exactly what it wants. It's my job as a builder to draw that knowledge out and do the hard work of shaping it. People pay me for this work, not the final output.

For example, if you look at a tool like Linear, you'd imagine that it isn't the toughest thing to build. However, had I asked you to make Linear without ever seeing or using it, would you have been able to? Probably not. That's why we pay them—not for the software itself, but for the great choices they make.

I'm actively looking for a niche to serve. I haven't found one yet, but it's only a matter of time before I get there.

Closing

If I'm being vulnerable for a moment, I did not expect 2025 to go this way. I thought I'd be making rapid progress towards building a new kind of computer. With the benefit of hindsight, though, I think it went exactly as it should have gone. If it had gone the route I wanted, I'd probably have had a harder fall. So I'm in a better position now.