A working code for AI at work.
Ten reasons to bother. Eleven things you do.
This is a code, not a course. You can read it in twenty minutes. The point is to keep it nearby and use it until it stops being a list and becomes how you work.
It comes in two layers. The Spirit is the character behind the work: ten reasons to bother, the things that keep the practice honest. The Moves are the practice itself: eleven things you do at the desk.
Hold both. The Spirit without the Moves is a poster. The Moves without the Spirit is a checklist nobody keeps. Together they're a way of working.
Why you bother. The character behind the work. None of these is a thing you do; each is a reason the doing is worth it.
What you actually do. Eleven things, in this order. Each runs four pages in the book: the idea, why it matters, in practice, and a worked example.
Method
Work to a structure.
Most variable results come from variable prompts. A structure removes the variation. Use the same four parts every time — Role, Context, Task, Format — and the output gets predictable enough to improve.
Typing the first thing that occurs to you, then blaming the tool for a vague answer.
W. Edwards Deming, The New Economics
Toolbox
Write down what you use and how you work.
Keep a written record of the tools you reach for, what each is for, and the standing instructions that make the AI sound like you. The toolbox is what stops you starting from scratch every time. Written down, they compound.
One favourite prompt used for jobs it was never meant for.
Abraham Maslow, The Psychology of Science (1966)
Focus
Mind the work, not the tools.
Stay fixed on the tasks you do and the skills they need. Tools change every few months; tasks and skills last. Organise your learning around the work and the tool churn stops mattering.
Knowing every new feature and finishing no actual work.
Steve Jobs, WWDC 1997
Review
Check it before it leaves you.
Nothing the AI makes leaves you unread. AI output is plausible before it is accurate. The review is where your judgement is added — and your judgement is the only thing the reader is paying for. Skip it and you are not using AI; you are forwarding it.
Pasting the first answer straight into the thing that matters.
attributed to James A. Michener
Understand
Use only what you can explain.
If you can't say the point, the reasoning and what's been left out without looking back at it, you don't understand it, and you're not ready to put your name to it. Understanding is what separates using a tool from being used by it.
Presenting an analysis you can't explain when someone asks 'where did this number come from?'
Richard Feynman, last blackboard, Caltech, 1988
Validate
Make sure it landed, both ways.
Two directions. Check the AI understood your prompt before it runs. Check your reader understood the output before you call it done. Sending is not the same as landing. Both gaps are invisible unless you check.
Assuming the message arrived because you sent it.
Russian proverb, used by Ronald Reagan, 1987
Lists
Keep three of them.
Three lists carry the practice: prompts that work, failures worth learning from, and a weekly note on what changed. They are your memory, kept outside your head. Without them you repeat your own mistakes and learn nothing from the reps.
Solving the same problem from scratch every month.
David Allen, Getting Things Done
Retrospective
Look back so the next session starts better.
End each session in a state that makes the next one easier. Save the prompts that worked. Log anything that failed. Clear the rest. Reps without reflection just bank the same habits — good and bad.
A great result you can't reproduce, because you didn't note how you got it.
Søren Kierkegaard, journals, 1843
Experiment
Try the unsafe thing, expect a third to fail.
Run real tests, not safe ones. If everything you ask AI to do works, you're only asking for what you already know it can do. A healthy practice has a failure budget. Those failed sessions are not waste — they are the search.
A zero per cent failure rate, which means you stopped exploring.
Samuel Beckett, Worstward Ho (1983)
Persist
A little every day, deep dives when you have time.
Adoption is a habit, not an event. Short and regular beats one big sitting. Ten minutes most days outpaces a weekend crash course you never repeat. The curve is the same for any new skill: slow, then sudden.
Trying it hard for a week, getting mixed results, and stopping.
Anthony Trollope, An Autobiography (1883)
Delegate
Hand over the right work.
The skill is no longer doing everything. It is deciding what to give the machine, what to keep, and what shouldn't be done at all. Hand over the wrong things and you'll spend longer fixing AI output than the job would have taken. The judgement of what to delegate is the new craft.
Automating a task efficiently that you should have stopped doing.
Peter Drucker, Harvard Business Review, 1963
This is a starting point, not a destination.
After ninety days you'll know which Moves matter most for your work and which Spirit words you keep returning to. Write your own version. Keep the two layers. Name the Moves you actually use, the review standard you hold, and the failure rate you'll accept. Revisit every ninety days and ask what you've learned.
The code is the floor, not the ceiling. Once you have one, you can build on it.
Advise, then make the thing that proves the advice.
Craig Stanley · Craig Stanley Studio · First Edition 001
Ugly Designs That Work
The same Spirit + Moves framework, applied to a different question: what happens when an AI designs something that looks wrong but outperforms every human solution? After Urania — the AI that redesigned gravitational-wave detectors and quietly rediscovered Soviet physics nobody had tested in forty years.
Ten reasons to bother. Eleven things you do. Different content, same code.