AI Adoption Is Not a Tools Problem
Most stalled AI adoption I see is diagnosed as a software problem. I think it is usually a trust problem wearing a software costume.
Working hypothesisFrom experienceDraft
Independent research practice
I study how businesses succeed, how they fail, and how AI changes the conditions beneath both.
As AI takes on more routine work, curiosity, taste, and connection become more valuable.
Above the line
The surface is honest
about the wrong things.
Revenue ↑Headcount ↑Pipeline fullAwards shelf dusted
Every one of these signals is real. None of them tells you whether the structure underneath can absorb a change in conditions. I ran a business that looked strongest the year before it broke.
The conditions beneath
- Customer patience
- Key-person dependence
- Financing assumptions
- Concentration risk
- Trust inside the team
The conditions that decide outcomes rarely appear on a dashboard. Finding them — and watching how AI moves them — is the work of this site.
Research
Follow the thinking as it forms.
Arguments published with their evidence labeled — including the ones that later turn out to be wrong.
What a Patrol Aircraft Taught Me About Weak Signals
Maritime patrol work is the discipline of finding quiet evidence under a loud surface. I keep meeting the same discipline — and the same failures of it — inside businesses.
From experienceMy interpretationDraft
The Sequence Before the Stack
Owners keep asking me which AI tools to buy. I think the ordering of adoption decisions matters more than any item on the list.
Working hypothesisCurrent observationDraft
Method
Four lenses on every case.
- Principles
- What must stay true for this business to deserve to survive?
- Decision logic
- How choices actually get made — and how they adapt when conditions move.
- Implementation
- Where intent meets Tuesday afternoon: sequences, checkpoints, and proof.
- People and culture
- Who is trusted, who is heard, and what changes when AI joins the crew.
Learning Lab
Practical guides for careful operators.
Field methods you can run this week — each one scoped by risk, time, and what you walk away holding.
Turn a Repeated Workflow Into an AI-Assisted Process
One workflow running with AI assistance, a written checkpoint design, and a go/no-go decision based on your own evidence instead of a vendor demo.
Draft
Choose an AI Model by Risk, Not Hype
A simple risk map of your AI use cases, and a defensible model choice for each one.
Draft
Before You Put Company Data Into an AI Tool
A one-page data boundary — what may, may not, and must never go into an AI tool — written in your own operating language.
Draft
The human position
As AI takes on more routine work, curiosity, taste, and connection become more valuable.
This is the working thesis under everything here. Not that people must race the machines — that the distinctly human work is about to be worth more, and most businesses are not organized to notice.
On the desk now Draft
Open questions, currently under the instruments.
- How do leadership teams keep cross-checking alive when one crew member is a model?
- Which AI adoption sequences produce their failures cheapest and earliest?
- What does trust cost to rebuild after a bad automation call — and who pays it?
The line so far
Success and failure, both from the inside.
- Military aviationMaritime patrol: weak signals under a loud surface
- Building a businessRenewable energy, built from zero
- Visible growthThe years the numbers looked best
- Examined failureConditions changed faster than the structure
- Coaching studyHow people actually change their minds
- Building with AIDaily, hands-on, unglamorous practice
- Research practiceThis site — the thinking, in public
Bring a question.
The most useful thing you can send me is a hard question from a real operation. Not a pitch, not a brief — a question you are actually carrying.