This piece opens a new section of ProductHarness, Operation Scarecrow, where everything about Company and Enterprise Brains gets worked out in public. What follows is a short restatement of the argument "The Backplane Isn't One Memory. It's Four." already made in full, and the case for why what comes after it needed a room of its own rather than a slot in the usual rotation. Sources at the foot of this piece point back to the fuller versions.
Cognitive science doesn't describe memory as one system. It describes several — a transient working memory for whatever's in front of you right now, an episodic memory for specific, dated events, a semantic memory for facts with the originating moment stripped away, and a procedural memory for skills encoded as behavior rather than something you'd read. Organizational theory found its own version of this thirty-five years before AI agents made it urgent: Walsh and Ungson's 1991 framework, five retention bins an organization stores what it knows in, most of them retrieved by humans tolerating ambiguity well enough to ask a colleague or walk the building and just find out. Agents can't do that. They have no tolerance for the ambiguity a person shrugs off without noticing, which is the actual reason "Company Brain" turned into a funded category in 2026 — not new idealism about organizational learning, but a new kind of reader who can't route around bad memory the way people always have.
The argument built from there is that most of what's sold and built under that name bundles four different kinds of memory into one thing with one governance rule, the same mistake the original ProductHarness backplane made when it filed a dated decision and a settled firmwide fact under the same review process. A decision is a precedent, not a law. A fact that's been independently confirmed across separate contexts is worth enforcing everywhere; a hunch that seems generally important is not, no matter how confident the person voicing it is. Getting a runbook wrong looks like a bug, which is why it belongs under engineering review rather than content review, and working memory should barely be governed at all, because ceremony on something meant to be disposable is the wrong ceremony entirely. The fix undoing all of this isn't a product. It's an org-design decision — a real steward for firmwide facts, a mandated filing practice for decisions, engineering discipline extended to the procedures that currently get a pass — because the four kinds of memory don't need the same kind of ownership, and no vendor selling one undifferentiated index is going to notice that on your behalf.
Why this needed its own room
That argument was one essay. Everything that came after it wasn't going to fit in a second one.
The four memory types alone turned out to need separate, full treatment each — what a settled fact actually costs to govern at scale is a different question from what a runbook costs to maintain, and flattening both into subsections of one long page was already how the original backplane essay lost people halfway through. Underneath that, a set of questions kept surfacing that didn't belong to any single memory type at all: the mechanism by which a decision earns the right to become a firmwide fact, what happens when an organization wants to unmake one it already settled, whether anyone has actually built this end to end — Palantir came closest and still repeats the same governance mistake one layer down — and where something as ordinary as a project budget actually lives once you stop treating "budget" as one object instead of four. Each of those turned out to be a real, standalone piece of writing, not a paragraph borrowed from the main argument.
There's a second reason, less about volume and more about honesty. Two Cards and the Demo Gap were argued once and are finished; rereading them next year won't require an update, because the claims don't depend on a market that's still moving. This one does. The Company Brain category is being built in public, in real time, by vendors whose products change every quarter, against a framework that's still being stress-tested against new cases as they come up — which is exactly what the last several weeks of this looked like from the inside. Filing that under Ahead would misrepresent it as a settled forecast. Filing it under The Work would misrepresent it as finished practice. It's neither, and pretending otherwise for the sake of matching the site's existing two sections would cost more than it saved.
The third reason is just that it doesn't sit on the same axis as the rest of the site. The Work is about what a PM actually does differently right now. Ahead is about where that role goes next. This is about something a company has to build regardless of who ends up owning it inside the org chart — infrastructure, not craft. Readers who came here for either of the existing sections don't necessarily want this, and readers who want this specifically shouldn't have to wade through unrelated PM essays to find where it continues.
What Operation Scarecrow actually is
The name is the obvious one: an organization chasing a brain it doesn't have yet, the same way the Scarecrow spent a whole movie wanting one, only to find out afterward it had been doing the reasoning the entire time. Most of what gets sold as a Company Brain in 2026 has the same structure — a lot of confidence about wanting one, not much rigor about what it's actually made of.
This section is where that rigor gets built, piece by piece, and it's going to look less finished than the rest of the site while it's happening. Each page here will carry its own note about how settled it is — a working page is an unfiled thought, an episodic page is a dated argument that hasn't been tested against enough independent cases yet, and a semantic page is the small number of claims that have actually earned the right to be cited as done. That's not a formatting choice. It's the same distinction this entire section argues an organization needs to make about its own knowledge, applied honestly to the knowledge on this page.
Sources and further reading. This piece restates the argument made in full in The Backplane Isn't One Memory. It's Four., which itself answers a question raised in Three Pillars, No Fourth and revises the product backplane defined in When Execution Becomes Abundant. Full citations for the cognitive-science and organizational-memory research live at the foot of the Backplane piece rather than repeated here.