Context Engineering: Scaling Judgement in Agentic Systems
Getting more work done with AI agents means helping the next agent understand the decisions you've already made. Here's how I think about that problem.
I build ambitious things with AI—and show you what works, what breaks, and what I change. From local hardware to agents doing real work.
The films. The experiments. The guides to try it yourself.
Getting more work done with AI agents means helping the next agent understand the decisions you've already made. Here's how I think about that problem.
My AI workflow classified a medical association as a shooting range, and two AI models approved it. How it happened, what we found auditing 2,484 facts our AI accepted, and the free guide to hand to your AI.
We tuned four R9700s on 100 real USA Arms jobs and got almost 9x more right answers an hour. How we did it, how to tune yours for your own work, and the free guide to hand to your AI.
Why I started moving the AI behind USA Arms onto our own hardware, a video of building three quad-R9700 servers, and the full build guide to hand to your AI.
What I’m building
A directory helping people find shooting ranges, instructors, firearms dealers, clubs, and other local resources.
Explore USA ArmsThis is where the lessons come from.
Messy sources. Ambiguous answers. Agents that get things wrong. I’m building the systems behind USA Arms and sharing what it takes to make them useful.
The failures become tests. The experiments become films and guides you can use in your own work.

I’ve spent nearly 20 years building data-driven businesses, working on programmatic SEO and lead generation. Now I’m exploring how much more one person can build with AI.
Expect real projects, useful lessons, and the occasional piece of hardware I probably didn’t need.
More about meTwenty years of this site. Writing on business, distribution, and doing your own thing.