A board deadline with no definition of “done” is a wish.
Learn how to turn a one-line mandate into a finish line you can defend.
“Make what done looks like impossible to miss, and then get out of the way.”
The mentor, in the novel
The Achilles Project · A teaching novel about how leading businesses use AI, by Jeremy Evans
AI doesn’t fail in the demo. It fails on the floor.
Learn to put AI to work in your operation. A business novel makes the lessons stick, and the full method ends the book.
You bought 500 AI seats. The P&L didn’t move.
Learn how to change the work, not just the tools, so the board sees the return.
“We were being billed for the meeting in which we were being shown the bill.”
The VP of operations, in the novel
Three years locked in off a slick demo. Your data too.
Learn how to keep the system and data yours, and leave any vendor when prices fall.
“No transfer of our configs, prompts, history, or logs. No second copy inside your walls.”
The VP of operations, in the novel
You banned AI tools. Security found customer data in personal accounts.
Learn how to build an approved route that beats the workaround.
“If you need to fire me, fire me. But walk the floor. Everybody’s got a tab open they don’t tell you about.”
The in-house builder, in the novel
Your 30-year veteran retires Friday. The reasons walk out with them.
Learn how to turn their judgment into instructions AI can follow, before the farewell cake.
“You can’t bottle a handshake. Write the magic down and you lose what made you.”
The founder, in the novel
$100k+ AI spend based on “vibe” doesn’t get past a CFO.
Learn how to show the returns on AI and, while you’re at it, every FTE.
“I can explain high spend that buys a controlled result. I can’t explain spend I can’t attribute.”
The board’s representative, in the novel
The cost of running GPT-3.5-level AI fell over 280-fold. Your roadmap missed it.
Build in days, not quarters, as costs fall.
The AI’s answer read perfectly. The regulator found the error first.
Learn how to verify the result, not the story, and hear about failures first.
“Our code turned an error into zero, and the model wrote a perfect answer to the wrong facts.”
The in-house builder, in the novel
Same question, two answers. Which one did the customer get?
Give AI the history, rules and one workflow.
“Capture how the best people actually grade a file, so the machine runs it the same way every time.”
The VP of operations, in the novel
When the AI fails, the board wants a name. It’s yours.
Learn how to sign for AI’s work with a record any stranger can verify.
“Give them a person, we contain it. Give them a system, the whole thing burns.”
The board’s representative, in the novel
The Goal and The Phoenix Project, for putting AI into production.
Nine stages of AI maturity. Find the one you’re on.
Rules and forms only
Your systems run on fixed rules and forms. Anything written in free text still needs a person to read it.
Read Stage 1Unsupervised AI pilots
Someone uses an AI model on real work, but nothing checks its answers and no one owns the result.
Read Stage 2AI tools for everyone, no shared method
Staff have licensed tools like ChatGPT, Copilot, or a vendor platform, but each team uses them differently and results vary from desk to desk.
Read Stage 3Shared playbooks on a setup you control
AI runs through a service you own, with written procedures and a named owner for each process.
Read Stage 4Costs and quality under control
You know what each AI task costs, you match models to jobs, and automatic checks catch bad answers.
Read Stage 5AI that knows your business
Customers, history, and past decisions live in one connected place the AI can look up.
Read Stage 6Precise context, measured accuracy
Each task gets exactly the information it needs, and accuracy is measured while costs fall.
Read Stage 7Proven work runs on its own
Routine tasks with a proven track record run without review, while high-stakes decisions wait for a person.
Read Stage 8Audit-ready and fully owned
Every AI decision is logged (what it saw, what it recommended, who approved it), and you could switch vendors tomorrow.
Read Stage 9
A result you can put in front of your board.
They said Stage 5. Their floor says Stage 3.
Your AI slide says one stage. Your floor says another.
Free, and no sign-up to see your result.
Effective — Question 1 of 15
The board wants an AI story by next quarter.
How did you pick what AI works on first?
- We turned on the AI built into software we already pay for, like Microsoft Copilot.
- A team lead offered up a painful process, and we measured it before and after.
- We ran an ideas contest and funded a dozen or so pilots at the same time.
- One pilot, on a backlog of emails and documents where a mistake is cheap to catch.
- Leadership picked a showcase project, like a chatbot our customers talk to.
Welcome back. See your last result →
Hardcover $32.99 · Paperback $24.99 · Kindle ebook $12.99, included with Kindle Unlimited. US list prices. Each store sets its own.
Before you begin.
Who's behind this, and what are they selling?
Jeremy Evans runs Thrice, an engineering firm that builds these systems for clients, and the book points readers to his firm's work.
Is the book out?
Yes. It's on Amazon now.
Why a novel, and not a playbook?
Because the failures that sink AI in production don't look like failures when they happen: a green dashboard, a fluent rationale, a great demo, a sensible ban. A story makes you sit through the week each rule didn't exist. You get the playbook too: the whole operating system ends the book, bare, for anyone who wants to run it on Monday morning.
I don't have time to read a novel.
Start where your problem is. Each of the twenty-nine chapters turns on one problem, and the titles say which: "It Ticked" is the AI bill, and "Whose Throat to Choke" is ownership. If you want the system first, the back matter stands on its own.
Is this another AI hype book?
No. In the novel the AI keeps failing until the company builds owners, checks and a meter around it. It grades a fatality as moderate, writes a perfect rationale on top of an error that became a zero, and one month costs $247,311.08. The book's thesis is that nearly every failure of AI adoption is organizational, not technical.
How is this different from books about AI strategy?
Strategy books end where this one starts. In the novel, the strategy is a single page handed over on the first morning. The rest of the book is the part that decides whether it works: who owns each answer, what checks it, what it costs per record, and whether the record holds up when an outsider pulls a file.
We're not an insurer. Will it apply?
The system rests on two tests that have nothing to do with insurance: can a machine now read this work, and is it cheap to be wrong? The last chapter makes the point in the story itself, when a freight-logistics leader maps the system onto her own operation: "Same ladder. Different nouns."