Adam. Writes
about building
with AI, mostly.
Building with LLMs, agents, and whatever else ships this week. Essays, build logs, and experiments — shared raw so I can look back and cringe.
Human in the Loop Is a Trap
Turn-taking is the rendering. Memory is the runtime. We solved multiplayer for the machines and left the humans alone in a corner.
Read essay →I'm a software engineer (other fancier titles too) in Sydney building with AI. I have thoughts and opinions from time to time, and thinking them out loud here is cheaper than therapy — essays on what I'm reading, mulling over, build logs from things I'm shipping, and experiments good, bad or ugly.
Write
Short essays on what I'm learning. Usually born out of an argument I'm having with myself.
Build
Small things that teach me something. Agents, MCP servers, eval harnesses. Most last a weekend.
Share
In public, rawer than feels comfortable. This is the archive I wanted to find when I started.
Marking My Own Homework
The 2026 AI predictions nobody asked for — graded at the halfway line, no curve.
Burn After Reading
Why I built an encrypted AI you can cryptographically destroy — with a receipt to prove it.
Dangerously Ship
Ideas got valuable, execution got free, and the only thing still scarce is knowing gold from slop.
- 2026
Notewell
My daughter couldn't take notes. So — as one does — I built her Notewell: an app that teaches the science of note-taking and quietly runs on it.
Read → - 2026
Resumatron9000
Resumatron9000 — a JD-tailored résumé and a recruiter-facing interview agent, engineered so neither can invent a fact, figure, or credential.
Read →
- —Reading: The Scaling Hypothesis (Gwern), and re-reading Rich Sutton's essay.
- —Building: a smaller, faster version of evalharness that runs on a laptop.
- —Writing: an essay on why evals are the product, not the model.
- —Avoiding: premature abstractions. Again.