The trap: making new things while your best work sits in a drawer
Most people meet AI and immediately ask it to make something new.
That's the expensive mistake. You're sitting on the most valuable training data in your industry and you're ignoring it. Thousands of emails where you explained something hard to a nervous client. Recorded calls where you talked someone off a ledge. A folder of screen recordings nobody has watched since you made them. Decades of judgment, already expressed in your own words, scattered across places nobody can reach.
The bottleneck was never that AI doesn't know enough. It's that your best thinking is unreachable. Fix the reach and the leverage shows up on its own.
Large language models need language
Here's the part people miss. These models run on language. Feed them yours and they learn to think in your patterns. Feed them nothing and they hand you the average of the internet.
That makes recorded conversation the single most undervalued asset in your business. Every call you take, every question you answer out loud, every explanation you've given a hundred times ... transcribed, that's gold. It's context engineering that costs you nothing extra, because you were going to have the conversation anyway.
And it cuts the other way. If your work happens in person, in rooms where nobody hits record, you're at a real disadvantage. You end up summarizing yourself for the model instead of feeding it. A summary is a photograph of a conversation. The transcript is the conversation. Start capturing what you can, however you can.
Start with the archive you already have
Before you build anything, find out what you're holding. Give AI access to your inbox, your documents, your recordings, and make it map the territory before it touches a single word. The move is a ladder, and skipping a rung is what makes this fail.
- 1Map itWhat kinds of conversations do you actually have? Where does your expertise cluster? You'll be surprised, and the surprise is the point.
- 2Build the baseTurn the map into a knowledge base that captures how you explain things. Your phrasing, your analogies, your order of operations.
- 3Point it at somethingDraft replies in your voice. Answer client questions without you. Train the team. The base is worthless until it powers something.
The archive-mining prompt
Now that you have access to my email and documents, here's my goal in three steps. First, research and categorize the kinds of conversations I've had over the last few years and give me a map of where my expertise actually lives. Second, turn that into a knowledge base that captures how I explain things, in my voice, not a generic one. Third, use it to draft in my voice so I stop writing the same answer for the fifth time. Don't build anything yet. Show me what you found and let's agree on the shape before you start.
That last line is doing real work. Give it the whole ladder up front, then hold it at the top of the stairs. The conversation that follows is worth thirty minutes of your time, and it's the difference between a knowledge base you use and one you abandon.
When it's in your head, not in a file
Sometimes there's no archive. The expertise is in you, and the outline you've got is thin. Don't ask AI to improve it. Ask it to interview you.
The extraction prompt
I want to get something out of my head and into a form we can both use. Here's my rough outline. Slow us down, and don't improve it yet. Ask me 20 questions about what I actually believe, what I've learned the hard way, and what most people in my field get wrong, so you can align with my perspective instead of a generic one. Don't stop until you're certain you've got it. Then tell me what I'm missing and we'll build the real version together.
Twenty questions you didn't want to answer will get you further than any clever instruction. You walk out with your own thinking, sharpened and written down.
Name what you find
Something happens when you mine your own archive. Patterns surface. The same five-step sequence you've run for years. The framework you never bothered to label because it was just how you do it.
Name it. Label it. Write it down as a method. That's not busywork, it's intellectual property. The knowledge base is useful. The named methodology is an asset ... something you can teach, license, build a book around, or hand to the next hire. Ask AI to watch for it while it's digging: flag any repeated framework or approach that deserves a name, because you want to know what you've built without realizing it.
Amplify, then verify
One guardrail, and it matters more here than anywhere else. When AI reaches into your archive and hands back something polished, it feels finished. It isn't. Models invent citations. They'll produce accurate information wrapped around a source that doesn't exist. That's a small problem in a draft email and a catastrophe in a book with your name on it.
Cross-check with a different tool. A model built for research will find sources one built for reasoning won't, and the reverse is also true. Don't let a model be the only judge of its own facts.
The verification pass
Check every factual claim and citation in this. For anything you can't independently verify, flag it and tell me you couldn't find the source. Don't quietly drop it and don't paper over it.
The real prize
Ask any owner what they'd do with another one of themselves and watch their face change. More clients served. More time back. The thing they've been putting off for six years.
That's what this shift is actually about. Not a better chatbot. Not a faster draft. You've spent a career building judgment that only exists inside one head, and for the first time you can put it somewhere else and let it work while you sleep.
You already did the hard part. You lived it. Now go get it out.
Your homework
Pick the single biggest pile of your own expertise ... the inbox, the recordings, the video library, the drafts. Point AI at it this week with the archive-mining prompt. Come back with the map: where does your expertise actually live, and what's the one thing you'd build on top of it?