I lead product and engineering, and I still ship.
I build systems that show their work.
- 2M+monthly visitors on search I built
- $6Mplan audited before it was built
- 99.9%uptime, from several crashes a day
- 90+engineers trained
- 2patents pending
- 8,731commits shipped at GARDE1
You only find what’s universal by living where everything else is different.
Tell me what you’re working on For hiring managers: the record
So I built a company where AI has to show its work.
01
GARDE1
CMMC tooling was stuck in 2010: checklists, and software that needed an expert to use it. 110 controls, 320 assessment objectives. GARDE1 does the work.
PWTL (Prompt Workflow & Task Language)
A task language for AI judgment. Instead of writing an answer, the model chooses one, cites the evidence it relied on, and keeps its reasoning, including what it ruled out.
It can’t invent an outcome, and an uploaded file can inform a decision but never instruct the model, so prompt injection has nothing to grab. Patent pending. It runs every decision GARDE1’s AI makes.
The GARDE1 document system
Compliance documents written only from what an organization’s facts require, with a receipt for every sentence.
Every statement can answer an auditor’s “why is this here?” Nothing is invented. Same facts, same policy, every run, and changing one fact changes only what it affects. Patent pending.
Locus
Lets people get around the product by chatting, without the chat ever navigating them away.
GARDE1’s customers run machine shops, not IT departments, so they shouldn’t have to learn the software to use it. Locus sees only the rendered tree, what’s actually on your screen, and a transformer running in the browser in WebAssembly matches what you ask to it. Anything sensitive stops at the product’s own confirmation.
It took nine years and ten steps to get there.
02
The path
Founder of GARDE1. Two patents pending. Nine years of production systems, from a Y Combinator startup to AbbVie.
Along the way, I write about what AI will actually change.
03
Grounded
“AI made producing the report free, left the judgment exactly as expensive as it was, and then multiplied how much judgment was needed.”
From The Wrong Layer of Abstraction
IP and Novelty in the Age of AI
“Notice that none of these moats is knowledge alone.”
Learning the Language of AI
“The suggestions shaped not only the wording but the perspective.”
Seven Tests for Consciousness
“If you can replay the exact same stimulus and get the exact same response every single time, the system is executing, not experiencing.”
Beyond Vibe Coding
“Effective vibe coding actually requires more expertise, not less.”
None of it makes sense without the places.
04
Lived
Understanding humanity, through the systems of our lives.