Writing
Design practice, education, AI governance, and how they turn out to be the same thing.
Engineering Intent
Spiral curriculum. 23 chapters.
Twenty-five years of the same operation across different materials. Construction sites, print shops, enterprise platforms, brand systems, AI governance. The methodology existed before the tools did.
Read →
How I Got Here
Construction sites, SVA, screenprinting, twelve years on an enterprise platform. The through-line was always the same.
Read →
Why Does My Brain Turn Everything Into a System
Construction sites, screenprinting, software, breakfast. The same operation across every material. I only saw it when I had to put it all in one place.
Read →
Lens Extraction
"Act as Vignelli" gives you a caricature. Extracting evaluative frameworks from bodies of work gives you testable diagnostic lenses. The difference matters.
Read →
I Needed a Better Tool
Thousands of ChatGPT conversations in two years. Then I found Claude Code, and the thing that made it stick was that it could read my files.
Read →
Voice Governance
AI copy all sounds the same because it learned from published writing. Published writing is performing. Real voice is in conversations.
Read →
The Integrated System
Coordinators dispatch independent diagnostics in parallel, surface where they agree and disagree, and leave the decisions to the maker.
Read →
Reading the Room
Everyone's been on the receiving end of something that just knew. A DJ set, a teacher, a product that didn't make you feel stupid. That's not magic. It's a skill. And it transfers to everything.
Read →
I'm Compilative, Not Generative
Most people use AI as a generator. I use it as a compiler. The distinction matters because compilation preserves authorship. Generation replaces it.
Read →
The Site Is the Proof
A blind evaluator read this site and concluded it was unequivocally human-written. It wasn't. Every page was compiled from three years of conversations by the system described on the site.
Read →
What I Built This Year
Each tool exists because the previous one failed. SavePoint because context evaporated. The voice protocol because the output stopped sounding like me. LensArray because evaluation was one question pretending to be twelve.
Read →
The IEP for AI Systems
I've been solving the same problem my entire career: decompose complex work for a system that can't process it whole, scaffold the structure, individualize the criteria. A self-contained classroom in Sunset Park is where I learned it.
Read →
What Does the System Actually Need?
I gave a language model four evaluation criteria in one prompt. The output degraded the same way a compound instruction degrades in a classroom. The fix was the same fix. I'm calling the framework accommodation design.
Read →
Processing Profiles
Your AI forgets the beginning of a long conversation for the same reason a student loses the third instruction in a compound prompt. A special education teacher already knows the fix.
Read →
The Unstructured Corpus
Everyone says structure your AI input carefully. I did the opposite: three years of talking out loud, brainstorming, arguing with myself. The messy version produced better results. Here's why.
Read →
Why Does ChatGPT Get Worse the More You Type?
Your AI gives sharp answers at the start of a conversation and vague ones by the end. It's a processing constraint that special education teachers have been designing around for decades.
Read →
What Special Ed Teachers Already Know About AI
You've written IEPs. You've decomposed compound instructions. You've built scaffolding that comes off when the skill is solid. You already know how to design for AI. The field just hasn't asked you yet.
Read →
Talk to It, Don't Type at It
Typing into AI is already filtering your thinking. You're editing as you go. Dictation gives the tools what they actually need: how you think before you've organized it.
Read →