WRITING & TALKS
Essays, field notes,
and one talk.
What I have learned building AI systems for real teams, written down so someone else can use it.
01 / ESSAYS, FIELD NOTES & THE TALK
Shorter, and closer to the work.
Two gates over the same content beat one careful pass
The build gate on this site caught its own scrub list being wrong twice. Then I pointed it at a page I had already cleared, and it came back with sixteen matches. Careful is a mood. A gate is a property of the system.
Read →ESSAY · 6 MIN · 2026The intake form is the product
The first artifact of an AI engagement is not a model or a demo. It is a structured set of questions, sixty fields, each one mapped to the file it produces. And every question left unanswered becomes a tracked item instead of a guess.
Read →ESSAY · 6 MIN · 2026Adoption is the unglamorous half
Nobody writes about the half of the work that decides whether a system still gets opened in month three. A diagnostic question instead of a pitch, a working artifact before a proposal, and a handoff that ends with their account configured.
Read →TALK · 5 MIN · 2026The room I almost did not get to stand in
DJ Von Frank reflects on returning to Hide In Plain Sight as its keynote speaker, and how targeted support shaped the systems he builds today.
Read →ESSAY · 7 MIN · 2026Why most company AI rollouts fail
It is not the model. Three failures I watched up close, the mechanical fix behind each, and the two problems every dead rollout shares: the system never knew the company, and nobody trusted it enough to hit send.
Read →FIELD NOTE · 2 MIN · 2026A control nobody has watched fail is a belief
The most dangerous state for any safeguard is reporting clean while broken. The only honest test is planting a violation and watching the thing die.
Read →FIELD NOTE · 2 MIN · 2026Never ask a client to organize their knowledge. Give them a shape.
The intake trick that makes non-technical owners succeed on the first try: build every folder in advance and let the structure do the sorting.
Read →02 / REFERENCE ARTICLES
The longer pieces.
From experimenting with AI to operating with it
Why companies that buy AI get experimentation instead of transformation: the access theory of adoption, the seven things a login does not include, and what it looks like to operate with AI as one system rather than dabble in it.
Read →REFERENCE · 8 MINHow to build a company AI knowledge layer
The architecture of a company AI knowledge layer, from four production builds: the constitution, the routing table, single-source-of-truth facts files, and why the folder structure is the architecture.
Read →REFERENCE · 8 MINAn enterprise AI guardrail checklist
A working guardrail checklist for enterprise AI, from four production systems: frame rules as refusals, wire them into the pipeline, scrub what must never ship, and negative-test every gate before you trust it.
Read →REFERENCE · 7 MINWhy enterprise AI should say UNKNOWN instead of guessing
Why enterprise AI systems should answer UNKNOWN rather than guess: a plausible guess in a facts file becomes a company claim within a week, and a numbered question register is worth more than fluency.
Read →REFERENCE · 8 MINThe AI adoption operating model
An operating model for AI adoption from four production rollouts: the intake, build, train, measure loop, why the bottleneck is trust rather than capability, and what to measure in month three.
Read →REFERENCE · 8 MINBrand-system architecture for AI-assisted teams
Brand-system architecture for teams producing with AI: tokens as law, contrast ratios measured and their failures published as rules, build gates that reject off-brand values, and reference-first skills that hold variance down at volume.
Read →REFERENCE · 8 MINA static-site QA and build checklist
A static-site QA and build checklist from a working system: run QA inside the deploy, resolve every link against files on disk, verify quoted code against its source, and leave tripwires for the day someone forgets.
Read →REFERENCE · 7 MINWriting evidence-backed implementation case studies
How to write implementation case studies that survive scrutiny: the problem, build, outcome shape, a single source of truth for every figure, leading with the smaller true number, and going qualitative where numbers cannot be defended.
Read →