/ writing
The essays I finally started writing.
I've had these in my head for years and never the time. So they live here now: half-finished, honest, updated when I have something worth saying.
I built the AI pivot from the inside. It didn't work.
In late 2024 every crypto company suddenly had an agent strategy, and I was inside one of them, building ours with my own hands. The agents mostly died. This is the honest anatomy of a hype-wave pivot (why we did it, what we built, why it failed) and why the failure is the direct ancestor of everything I build now.
read →contents · 10 more pieces
We rebuilt the whole game once. The mistakes survived the rebuild.
The Book of Aaru took five years, one full restart, and a small studio to ship. I already wrote the goodbye; this is the build: where the years actually went, why restarting fixed the assets but not the decisions, how the pace of AI made me rethink the whole path, and the 1%-loop lesson I'd run any project on now.
Everyone builds local-first second brains. I'm building the opposite.
Local-first is the sacred cow of second brains: files on your disk, plain text, outlives every app. I think it's the wrong architecture for the agent era, and I'm betting LoomBrain on the opposite: a brain that lives on the internet so agents can reach it while my computer sleeps. Build log: three ground-up rebuilds in four months, the WHY-annotation idea, and an empty user count.
Everyone says taste is the moat. That's the cope.
The comforting line is that taste is the one thing AI can't take from you. But nobody can define it, teach it, or prove who has it, and stripped of the mystique, taste is just an evaluation function: what tells a feedback loop better from worse. We already built the machines that learn one. The edge was never having taste. It's running the loop.
Stop typing. Don't stop understanding.
The keyboard is the part that depreciates, not the thinking. Stop typing code, direct the loop in English, and move your understanding up the stack. The trap is confusing "stop typing" with "stop understanding," and that is where careers and codebases die.
UI isn't dying. It's splitting in two.
The argument, steelmanned and stress-tested, with the diagrams to see it. The task interface is being demoted to something the agent generates and throws away. The trust interface is being promoted. Whoever confuses the two builds on the wrong half.
18 years, four bets, one pattern
Rails, React Native, Web3, agents. What catching a wave early actually costs, and why I keep doing it anyway.
The graveyard is the portfolio
On shelving most of what you build, and why the dead projects taught me more than the shipped ones.
Building the plumbing, not the interface
Memory, sandboxes, reasoning, routing. Why the least glamorous layer is the one worth owning right now.
Small models, big reasoning
I spent a while collecting 16 research-backed reasoning strategies and building a router that picks one per task. Notes from teaching small local models to think.
I shipped a game, then walked away
Five years, a small studio, a narrative game on Steam. What building Book of Aaru taught me that no SaaS ever did, and why I closed the studio anyway.
More as I write them. The feed on the home page shows what I'm actually doing this week.