“We loaded thirty years of projects into it. It answers everything.”
The one saying it is Tecla — who runs technology at the company, snorkelling mask on her forehead since this morning: the gear is always ready, the right moment never comes. The one answering is Rocco, the expert from the sun lounger next door, who already has the solution. “Then you’re sorted. Throw everything in and let it get on with it.”
It’s the request that comes to us most often, and it isn’t naive: give us everything you’ve ever done, AI learns how you work and does the next project itself.
If the knowledge is in there, why not use it?
Because what’s in there isn’t knowledge. It’s a deck of cards.
A thousand projects in a folder are a thousand cards at the same level. The one that made margin and the one that cost six months both weigh the same. Old revisions sit next to new ones, and the contradictions were always there: as long as a person was reading them, they knew which card to discard without even noticing they were doing it.
Then comes the machine, and it does one thing only: it searches by meaning. It hasn’t studied your projects and it hasn’t put them in order. It finds the piece that most resembles the question and brings it back, detached from everything else. It’s a very fast librarian. It isn’t an engineer with twenty years of craft.
And the part that matters isn’t in the deck. Why you chose that solution and not another. What constraints that client had and the others didn’t. What you never do, and since when. You never wrote it down because there was no need to: there was someone who knew.
Last week, different department, same rule: the management system didn’t imprint a method where the method wasn’t there. AI doesn’t create what was never written.
This way of working has a name — RAG — and the word is starting to circulate. One thing only matters: even the day you had written everything down — constraints, reasons, style — RAG would still be what it is. It draws from the deck.
The path isn’t a tidier deck. It’s a designed agent: one that knows where to look, in what order, and what never to use. The word “agent” matters little. What matters is that someone decided those three things, one by one, before switching it on.
Start small. Take the last project you delivered and write a single page: the three decisions that shaped it and where they came from — a client constraint, something that went wrong years ago.
That page doesn’t go in the folder. It goes into the hands of whoever will design the thing that will use it.
Knowledge first. Then the agent. Then, perhaps, the folder.