The Cup of Coffee Wish (Asking a Genie for Small Things)
The Problem
The first question almost everyone asks a new AI tool is the same: what are its limitations? What can't it do? So they go looking, and they find them, every time. Then they stop. The people who stay mostly ask small, polite things. Summarize this thread. Make this email nicer. Clean up this spreadsheet. Useful, a little, and nothing about their week changes. Meanwhile the automation team is a third of the way through a hundred pages of requirements for a problem someone has carried for years. Look at this field. Compare it to that field. If they don't match, check a third thing. Every step written down, because that's how we've always told computers what to do.
The Principle
Ask for the big thing first. Tell it what you want at the end, not the steps to get there, and then check everything. If you found a genie in your garage, your first wish wouldn't be a cup of coffee. Start with the problem you'd fix if you had a month to spare. Describe the outcome the way you'd brief a smart new hire. Talk instead of typing, rambling included, because the rambling is where the context lives. Then be the expert. It will get things wrong, sometimes confidently. Your name is on the work, so you check it. Checking an answer is still much faster than building one from scratch.
4 Steps to Take Action
- 1
Ask the genie question
Before your next session, ask everyone: if this could do anything, what would you ask for? Write down the answers. Start there, not with a list of limitations.
- 2
Describe the outcome, not the steps
Write what you want at the end, who it's for and what good looks like. Leave out the how. If it gets the how wrong, tell it, and let it try again.
- 3
Talk, don't type
Put your earbuds in and explain the problem the way you would to a new hire over coffee. A short, polite question gets a short, polite answer.
- 4
Check every line
You're the expert and your name is on it. Check it the way you'd check a new hire's first week of work.
💡 Inspired By
Chapters 2 and 4 of The Second Factory, "Licenses and Lip Service" and "The Front Row." And a real session where a woman near the front said, loudly and to nobody in particular, "I think I just saved a million dollars."
This is what Analytics Pitfalls & Principles is about: people helping people.
In Practice
Evan Brandt sells lumber. For nineteen years he did quotes by hand: a contractor texted him a photo of a deck sketched on the back of an invoice, and Evan sat down with a calculator for most of a morning. On the drive home one night, he asked the AI to build him something where a contractor sends a photo and gets a quote back. It asked him questions. Half of it didn't work. He told it what was wrong. Repeat about forty times. When Nora Berg, newly in charge of AI for a 900-person company, saw it on his tailgate, her first question was where it got things wrong. What were its limitations? Evan laughed. "That's what everybody asks. Say you're cleaning out your garage and you find an old bottle. You rub it, and a genie pops out. What's the first thing you say?" He still checked every quote before it went out. "I'm checking a quote now, not building one from scratch. I went from a morning to ten minutes." A few weeks later, ClassicMotors' VP of Finance came to Nora's first session. For four months her team had worked with IT on a hundred and twelve pages of requirements to automate a reconciliation that ate six months of their year. In the session she gave the AI three months of exports from all three systems and told it, out loud, what she wanted at the end: find every transaction that doesn't match, explain why, group them by cause, and tell me which ones we can still recover. Fifteen minutes later she said, loudly and to nobody in particular, "I think I just saved a million dollars."
