Field Notes
Observations from
the operational edge.
Posted Monday, Wednesday, and Friday. The work, the patterns, and the things worth saying.
-
November 5, 2026
States are starting to regulate clinical AI. Who on the building team notices?
California's clinical-AI laws take effect 1 January 2027, and Texas has its own. More and more, the teams building health tools make the compliance calls.
Read → -
October 22, 2026
What Jev's wrong answers told me about my own records
Jev scored 44% on my work items. Twenty of its 27 misses were the same mistake, caused by category definitions that lived only in my head.
Read → -
October 20, 2026
The use case picks the model
A model's published numbers tell you what it does. Only your own data tells you whether it gets your decision right. So I tested Jev on 48 of my work items.
Read → -
July 17, 2026
Agents Don't Replace Your Rules. They Wrap Them.
Let your people build with AI and the rules seem to loosen. Done right, it's backwards — they get enforced more reliably, not less. Here's the inversion that does it.
Read → -
July 15, 2026
Bring Your Shadow AI Into the Light
Half your team already uses AI at work — and most of them hide it. That's not a discipline problem; it's a readiness gap. Here's how to close it.
Read → -
July 10, 2026
Doctrine, Not Permission
The person who does the work can now build the tool that fixes it. When something goes wrong, the instinct is to lock it down. The military solved this a long time ago — and the answer wasn't permission. It was doctrine.
Read → -
July 8, 2026
Onboard It Like a Hire. Own It Like a Commander.
Framing an AI agent as an 'employee' made managers catch 18% fewer of its errors. You can delegate the work — never the ownership of the outcome.
Read → -
July 7, 2026
Buying the Model Isn't the Same as Being Ready
Part of what we do at Coyote & Quill is track the frontier and tell you what matters. Here's 28 minutes worth watching — OpenAI's president on why the model was never the hard part — and how it maps to the readiness framework we run.
Read → -
July 6, 2026
The Integration Was Me
76% of small businesses use AI. 14% have built it into how they operate. I launched Coyote & Quill in that gap — after six months of getting integration wrong.
Read → -
June 29, 2026
A skill isn't a document. It's a drill.
Researchers catalogued how developers configure their AI coding agents across 2,853 repositories. The most neglected mechanism — skills — is also the most misunderstood: almost all of them are static notes. The leverage is in the ones that actually run.
Read → -
June 26, 2026
You don't need to code. You need to give orders.
A site tested for ten users crashed under fifteen thousand. Everyone blamed vibe coding. It was a requirements failure — which makes it a leadership problem, not an AI one.
Read → -
June 24, 2026
The skill isn't prompting
AI fills in blanks you didn't know existed. The real skill in vibe coding isn't prompting — it's knowing which assumptions to shape before you type the first message.
Read →