/about
// staff platform engineer · open-source builder · writer
// what I work on
If you've tried to get real work out of AI, you've met the firehose: confident, fast, and wrong often enough that you can't tell what to trust. That's the problem I work on.
Day job: staff platform / SRE work in defense autonomy, with a Master's in Cybersecurity. Reliability, release discipline, and operating systems for autonomy infrastructure, at a production bar. Public writing here stays generalized and privacy-safe.
Side lane: open-source tools that make AI agents reviewable, testable, and operable; self-hosted patterns that make agents, RAG systems, and local models safe to run; and teaching that translates the discipline for non-engineers.
// what I believe
three rules I trust because they cost me to learn.
Stay under 40% of an AI's context window. Past that, in my runs, error rates climb fast enough that I treat it as a hard ceiling. The most important rule I've learned turning AI work into reviewable delivery.
Every AI session produces byproducts: decisions, patterns, learnings. Most people let these evaporate. I capture and compound them. Citations track what knowledge helped. Each session leaves the next one better-informed. The full picture →
Embrace variance (parallel agents). Filter aggressively (validation gates). Lock progress permanently (merge to main). The ratchet only turns forward.
// what I'm building
open-source tools for reliable AI delivery and self-hosted AI operations.
Coding-agent plugin for Claude, GPT, and other models. Skills run research, planning, validation, implementation, and learning extraction. Knowledge compounds across sessions.
Steve Yegge's context-orchestration ecosystem, run and extended in my own stack. How agent workflows become durable operating systems.
Git-backed issue tracking that survives agent sessions. Durable workflows that persist across runs and keep coordination explicit.
Case studies and the operating loop are at /work and /method. The methodology has its own home at 12factoragentops.com.
// the bigger picture
For the first time, AI puts the power of software in reach of people who can't code. The abstraction climbed from assembly to high-level languages to, now, plain words: describe what you want, and you can put software to work. But the benefit lands unevenly, mostly with engineers, because making AI reliable and safe takes discipline most people never learned.
So I'm taking the practices and safety I use in production and translating them for everyone else. It started with my own family. It's the step beyond The Gutenberg Moment for Code: taking that moment past code, to everyone else. /ai-partner →
// how to reach me
Building in reliable AI delivery, platform engineering, or agent-driven workflows? Or you want your own AI set up safely and taught to you? I want to hear about it.