For about a thousand years, Damascus steel was the best in the world.
Blades that held an edge no European smith could match, with that unmistakable watered pattern running through the metal. Everyone could see it. Nobody could copy it.
Then, sometime in the 1700s, it stopped. The smiths who knew how to make it died, and the knowledge died with them. Not the recipe. We have versions of the recipe. What died was everything the recipe didn’t say: the feel of the fire, the timing, the thousand small judgements that only lived in the hands of the people doing the work.
Metallurgists spent the next two centuries trying to reverse-engineer it from the finished blades. They mostly failed.
I think about this a lot, because every business I’ve ever worked in has a version of it.
The owner’s code. How they decide who to hire and who to walk away from. How they price a job that doesn’t fit the usual shape. The thing they will not compromise on, even when it costs them. The bar a piece of work has to clear before it goes out the door.
None of it is written down. Most of the time the owner doesn’t even know it’s there. Ask them how they made the call and they’ll shrug and say they just knew. That’s not modesty. It’s steel with no recipe. The judgement is real, it’s good, and it exists in exactly one place.
Which is fine, right up until it isn’t.
It isn’t fine the day they want to hand something over and find there’s nothing to hand. It isn’t fine when a key person leaves and takes the standard with them. And it isn’t fine now, because everyone is being told AI will carry the load, and here’s the part nobody mentions:
You cannot train a machine on knowledge that was never captured.
If the code is in your head, AI just gives you faster, more confident ways to work around the fact that it’s missing. It will fill the gap with something plausible that has nothing behind it. If the code is written down, AI gives you a way to apply it a hundred times without it drifting.
Technology multiplies a good system. It also multiplies the absence of one.
The watered pattern was the easy part. Anyone could see the pattern. The hard part, the part worth protecting, was the part no one could see.
Get it out of your head and onto the page while the fire’s still lit.
The recipe survived. The steel didn’t.
P.S. The Damascus steel line isn’t mine. I borrowed it from Mike Pino, who used it to answer a question I put to him at the launch of his book, HumanCorps, written with Andrew Lopianowski.
Mike has spent years mapping how organisations evolve, and lately how they adopt AI, and the book makes this case better than one newsletter can. It also practises what it preaches: they trained an AI on the book itself, every claim sourced, forty thousand words of endnotes, so it can actually answer for the work.
The knowledge was written down first. That was rather the point. Worth your time if any of the above landed.