Before there was a DAAC, before there was a Forge or a Frame or a Spine, there was a Saturday. March 28, 2026. The only thing on the list that day was a vault.
I opened Obsidian and started building a place to keep what I knew. Ten folders. Nine app registries, one per product line, each meant to track what it inherited from the framework and what it fed back. A folder for the business itself: manifesto, engagement tracker, an early investor pitch. Templates for the notes I hadn't written yet. Nothing shipped that day. No feature, no client work, no code. The entry for client-facing work reads, verbatim, "None today. Internal setup day."
I didn't think of it as infrastructure. I thought of it as tidiness: knowledge scattered across documents, a head that couldn't hold all of it, and a wish for one place to keep it. But look at what actually got decided that afternoon and it stops looking like tidiness. Every app got a registry that tracks two directions at once: what the framework gave it, and what it discovered and handed back. A propagation tag got invented before there was anything to propagate. Client docs were split from internal ones so white-labelling would be structural, not a later scramble. Migrate gradually, by layer, never in one dump. These were not filing decisions. They were a bet about how a body of knowledge should move through a company before I had proof the company would need it to.
I had used generative AI in the ordinary sense before that day: asking questions, getting drafts, closing the tab. This was the first time I built something with an agent rather than asked one for something, and the difference showed up immediately in what got produced. Not a document. A structure a document could live inside. Ten folders holding nothing is not useful on its own. Ten folders with a stated propagation rule between them is the shape everything since has grown into.
I did not know, that Saturday, that I was building the first version of what would later be argued for explicitly: a memory a machine can operate, not just read. I did not know DAAC would exist, let alone have a name, a suite, five tools with names of their own. What I knew was narrower and more useful than any of that: knowledge with no home gets lost, and a company about to ask software to build software had better have somewhere honest to keep what it has already learned.
Everything downstream of this blog traces back to that folder structure, still recognisable four months later under a great deal built on top of it. The rule that a piece of knowledge has exactly one proper home. The rule that a decision gets written down with its reasoning, not just its outcome. The rule that a lesson learned in one place has to reach the place that can act on it. None of those rules existed yet on March 28. But the shape that would eventually hold them was already there, on day one, before there was anything else to build.
It starts with memory. Not because memory is the most exciting place to start. Because everything I have built since has needed somewhere to remember what came before it, and the day I built that somewhere is the actual first day of this story, however unremarkable it looked from the inside.