Putting AI to work in your
company without losing
control of it
Cover for a book that will never go to press.
For companies putting AI to work
Revised monthly. Never printed.
Twelve chapters on the decisions a company actually has to make about AI, in the order they come up: what to build, what to forbid, what it costs, and how you know it worked.
A book on this subject is out of date months before anyone reads it. Look at what is already on the shelf. Courses age the same way and cost more, so this one gets revised instead of reprinted, and every revision is logged.
Agents read the field every day and draft the revisions. They cannot publish, they cannot reach anything else on the machine, and nothing here has gone up without me reading it first. This site is built the way the book says to build.
The current landscape revised today · Governance tooling added as its own category
From a novelty people typed party tricks into, to systems that do the work. How the question changed.
Why the split between the team that builds and the team that approves is the expensive mistake.
Nobody can tell you where the time goes. Here is how to find it without a survey.
Most of the value is subtraction. Stop at the first honest yes.
Four layers around every model, which one breaks, and which one you own.
Where company information goes, who keeps it, and how to decide before someone asks.
What it can reach, what it can change, and how much damage a bad run can do.
Where the company's knowledge lives, how a system gets at it, and why permissions do not survive the trip.
The arithmetic behind self-host versus API, and why most people argue it instead of calculating it.
Golden sets, graders, and stop rules. How you know before you trust it.
Symptoms, the layer that owns each one, and the fix that actually applies.
Pilots, owners, and getting adoption by pull rather than push.
Writing the rules last, on purpose, so they describe something real.
Tool categories and who occupies them. This page changes most.
Every term used three ways, and which meaning to keep.
The policy, the register, and the incident procedure, for this site. The chapter thirteen artifact, published.
AllThe guide was re-cut for companies rather than for beginners. Twelve chapters rewritten under one argument: enablement and governance are the same decision.
Ch. 5New chapter on the data boundary. Tiers, the four questions to put to any vendor, and the untrusted input triangle.
Ch. 7New chapter on company memory and retrieval. Including what happens to document permissions when you index everything.
Ch. 8New chapter on models, hosting, and cost. The arithmetic for self-hosting, including the two lines everyone leaves out.
Ch. 12Policy moved to the end of the book. It now describes decisions already made rather than predicting risks nobody has seen.
I take a small number of engagements at a time. If any of this maps to something you are dealing with, tell me about it and I will tell you honestly whether I am the right help.