Writing

A proof of concept that ships is permanent. It never gets finished, because finishing is never on the roadmap and there's always a next thing.

Subtraction · part 3

Every feature you ship costs you something, usually speed. The problem is that nobody ever writes the price down, so you end up paying it without ever deciding to.

Subtraction · part 2

Nobody's blocking the merge. That's usually the problem, not the good news.

Subtraction · part 1

When a team says AI doesn't work for them, it's usually the setup, not the model. Here are the four levels I'd work through, in order, and how to tell which one you're stuck on.

Heroism is a painkiller you hand the company. It looks like a cure until the team can't function without it.

Long-running agents don't need micromanagement. They need a way to know when they're done.

How I reverse-engineered RuneScape's obfuscated client and built a bot framework that survived weekly updates.

I didn't set out to learn programming. I just wanted to stop clicking.

Years of algorithmic contests left habits that still show up in how I approach engineering problems today.