Archive.

You don't bring automation to a job. You bring it to a process. The role most companies haven't hired for yet, and why I'm built for it.

AI made it almost free to look like a craftsperson. The Japanese idea of shokunin, lifelong devotion to a craft, is more useful now than it has ever been.

AI slop is what you get when you let the model run without much wrangling. In R&D, that trade is often worth it: speed and cheap learning now, cleanup once the shape stops moving.

Engineering is no longer the hardest part of building software. The compounding hire now is a product thinker who chooses what to build, in what order, and why.

I built a private UI kit called Couch Kit so coding agents stop improvising my interfaces and I can ship consistent product UI without re-deciding every detail.

I built an AI orchestration API and accidentally 10x'd my understanding of prompting.

VCs are betting everyone will build their own software. History suggests otherwise: better tools expand what builders can do, not who becomes a builder.

Most people confuse job risk with career risk. The safest-looking paths cap your upside. The weird paths? That's where the math gets interesting.

Stop babysitting your AI coding agent. The Ralph Wiggum technique runs your agent in a loop until the work is done.

True original thinking is rare. Not because it's hard, but because most people won't go where it lives.

You've heard 'reinforcement learning' dropped in every AI conversation. You've nodded along. Let's actually explain it.

After two decades online, I've stopped conflating 'having a voice' with 'building a brand.' They're not the same thing, and the difference matters.

At the starting line, you have almost nothing. Only a spark and a hunch. Here's how to use AI tools to explore ideas before you know if they'll work.