You Bought the Shop, Not the Furniture
A table saw will cut a board dead straight for anyone who plugs it in. No questions about experience. What it won’t do is tell you which board to cut, or why the drawer you’re about to build is going to stick in August.
Everybody understands this about woodworking. Nobody walks out of a hardware store with a router and a box of clamps and calls themselves a furniture maker, and if they did, their first bookcase would correct them inside a weekend. Owning the shop and knowing how to build are separate things, and only one of them can be bought.
AI is being sold the other way around. A company buys a stack of seats, and a steering committee starts waiting for the furniture to show up. But getting a model to do what you actually want is a craft. You learn it the way you learn joinery: slowly, by making bad things and working out why they’re bad.
The chair that looks finished
A beginner at the bench fails in public. The joint gaps. The table rocks the first time someone sets a coffee cup on it, and everyone at dinner can see exactly where the mistake is.
A beginner with an AI tool gets handed a finished chair every time. Upholstered, even. It reads well and it has headings. Whether it holds weight is a separate question, and the person least equipped to answer it is usually the one who just asked for it.
So most of the craft isn’t in the asking. It’s in the inspection: knowing, for example, that the posting behavior an answer describes isn’t how your ERP actually posts, or that the “best practice” it quoted belongs to a feature deprecated two releases ago. That’s domain knowledge, built up over years. No prompt template supplies it, and I’m wary of anyone who says the prompt is the whole skill.
Jigs
Watch someone who’s good in a woodshop and you’ll notice how little they freehand. Anything they’ll cut more than twice gets a jig, maybe a crosscut sled or a template screwed to a scrap of plywood. The skill gets stored in the jig, so the tenth cut doesn’t depend on whether it’s a good morning.
Skills are the AI version of that. I work across Claude, ChatGPT, Midjourney, Copilot and VS Code, and each one has built up its own library of skills for specific jobs. They don’t move between tools any better than a crosscut sled fits a saw it wasn’t built for.
A writing-style skill carries my rules (no em dashes, among others), so a draft doesn’t come back sounding like a product launch. Another builds my Midjourney prompts, so every AD&D365 image lands in the same golden-hour sourcebook look instead of whatever the model felt like that afternoon. And there’s a rack of more than forty cartography skills that take one source map of a town and redraw it as a watercolor atlas or a tactical grid while every landmark stays where the source put it.
Nobody builds a good jig before they’ve ruined a few boards, and nobody writes a skill worth keeping before throwing out a pile of mediocre answers and working out what each one got wrong. Which means a skill library is mostly a record of failures. Copy someone else’s and you get the shape without the reasons.
Grain
Wood has grain. Plane against it and the surface tears out in little ridges; work with it and the board comes up smooth. Models have grain too, and it’s consistent enough to learn. They agree with you faster than they should. Push one for a source it doesn’t have and you may well get one anyway, nicely formatted and entirely made up.
A skilled user stops being surprised by this and plans for it. Ask for the strongest objection before asking for agreement. Ask what the model is unsure about before asking what’s true. None of that is clever. It’s the same reflex that makes a woodworker look at the end of a board before running it through the planer.
What carries over
Woodworking tools don’t stand still either. The brace and bit gave way to the electric drill, and the electric drill eventually lost its cord. Nobody stops you from boring holes the old way (some people still prefer it), and most AI tools will let you go back to an older model too, at least until it’s retired. Or you learn the new one and keep the elbow grease for something else. What carries over between generations is the eye, the ability to look at a joint and know it won’t hold, whatever tool cut it.
None of this is news in ERP. Nobody who has lived through an implementation believes that buying the licenses means owning a working system, and the consulting side of this business has spent years explaining to executives that the software is the shop and the implementation is the furniture. Some of the organizations that learned that the expensive way are buying AI seats in bulk now, on a rollout plan that fits on a single slide.
My book, Gamifying the Enterprise: Game Mechanics for Continuous Proficiency, is on Amazon. All my books are on my Amazon author page.
The AD&D365 configuration guides are at adnd365.com/start.
Connect with me on LinkedIn.
