The AI Pilot Is Not the Unit of Value
Companies count pilots. Boards count outcomes. What survives a quarterly review is a workflow that runs, not a demo that dazzles.
A pilot is a probe, not a product.
JULY 2026 / thesis / governed work beats intelligent demos
Make AI do real work, not just impressive demos.
I help leaders close the gap between what AI promises in a demo and what it delivers on Monday morning — with clean data, encoded judgment, workflows people can trust, and a straight answer on cost.
What I've shipped
Things that are actually running.
Anyone can name a framework. Fewer people have systems in production behind them. A short list, kept honest.
A four-layer AI FinOps system, live in production
Reasoning, governance, orchestration, and execution kept in separate lanes — so the AI's guesses never touch a system of record without a policy checking the move first.
A consulting engine that works on real audit data
Reads audited enterprise data and produces the artifacts a board expects: RACI, roadmap, cost model, executive deck. People still make the call. The machine just stops making the deck the bottleneck.
A policy check between the AI and the action
Every move an agent wants to make is checked against written policy before it happens. This is the layer most enterprises need before they trust AI with anything that matters.
How I think about it
Three ideas do most of the work.
AI Value Engineering
Turning pilots into production. The unglamorous middle where most programs stall.
Cost of Work
A cost number a CFO can actually use — per outcome, versus what it costs today.
Governed AI Systems
The plumbing that connects data, judgment, policy, and workflow so the whole thing holds up under review.
Don't scale the pilot until the operating model is written down.
Recent writing
Short essays from inside the work.
The Missing AI ROI Number
Tokens tell you what you spent. Seats tell you who has access. Neither tells you whether the work got done cheaper. Cost of Work does.
Measure cost per outcome before you claim ROI.
Why Autonomous Agents Need Written Rules
You can't put a human in front of every AI decision. But you can put a written policy there. That's the difference between an agent you trust and one you don't.
Autonomy without written rules is liability.
Playbooks
Reusable ways of working — not slideware.
Playbook
Pilot-to-Production Gap
Why AI pilots stall right before they turn into everyday work — and what to do about it.
The board approved the money. The pilot works. It still hasn't crossed into production.
The pilot-to-production gap is an operating problem.
Playbook
Cost of Work
A simple cost number: what does one AI outcome cost, and how does that compare to what it costs today?
Token bills don't tell a CFO whether AI is worth it.
Outcomes first. Tokens second.
Playbook
AI Readiness Check
Seven questions that place your program on the gap and hand back one thing to do next.
Leadership teams need an honest baseline before they spend more.
Where is your program, really?
AI Readiness Check
Seven questions. One straight answer.
Not a description of a check. The check. Answer seven questions and get back where your program stands and the one thing to do next. Screenshot it, forward it, argue with it. No sign-in.
AI Readiness Check
Where is your program on the pilot-to-production gap?
Seven questions. Answer honestly and get back where you stand, what's still open, and the one thing to do next. No sign-in.
Books
Longer work on how enterprises actually create value.
One book out, three in progress — on startups, vertical AI, AI cost, and the data engineering that has to sit under agents.
See the shelf →- Startup Predicament
- The Vertical AI Playbook
- AI Cost Management