The consent-first company brain. You decide what it knows about you, and you can take it back.
Except for one thing. None of them answer the question the company actually needs: who knows what, why they decided it, and whether that decision still holds. That requires context no document contains. It lives in people.
Every company brain today is built top-down. Connect Slack, index Drive, point a model at the pile, and hope the retrieval is good enough. Glean is worth $7.2B doing exactly this, on $300M of revenue growing 89% a year. Microsoft Copilot is at 30 million paid seats and doubled that in six months.
Pick a ring below, or click one in the diagram at the top.
At the core, a digital twin of each person. Not a chatbot wearing their face. A structured, permission-scoped model of what they know, what they decided, and why.
The twin sits inside a personal brain, and next to it sits a vault. 1Password for your context. You hold the keys. You grant scoped, revocable access outward. The company never sees what you did not share.
It earns its place the moment it starts automating work that needs your context. Not generic AI assistance. Work that only succeeds because the agent knows how you handle things.
The company brain emerges from below. It is what personal brains add up to, each one contributing what its owner chose to share, with the original citation and the original permission boundary intact. No silent drops. No god-mode retrieval.
The outer ring is what this becomes when the system learns from its own operations. Continual learning, reinforcement from real work, and a world model built from real organizational behavior, not training data.
A digital twin is a company brain with a person axis on the index and a persona on the output. Around 80% of the code is identical. The 20% that differs is not architecture. It is consent.
The brain belongs to the company. The twin belongs to the person.
Revocation, portability on exit, an audit log the individual owns. Nobody has shipped this as a product primitive. The incumbents cannot, because their entire business model assumes org-level consent granted once by IT. The individual's side of the permission boundary is unclaimed territory.
We build the brain. We ship the twin as a view of it. And we enter through consent, the one capability a seat-priced, org-indexed incumbent structurally cannot copy.
Every answer carries a citation. Every retrieval respects two independent authorities: the company's existing file permissions and the individual's grant. Where they conflict, the person wins. One cross-boundary leak ends the company. This is non-negotiable engineering, tested in CI on every build.
Search tools answer "where is the refund policy?" A company brain answers "how do we handle a refund for an enterprise customer past 30 days, who decided that, and is it still true?" Then lets an agent act on the answer. The moat is the "why" layer that no incumbent leads with.
Model-agnostic. Feeds context into Copilot, ChatGPT, Claude, your own agents over MCP. Replaces nothing the company already pays for. Lowers the rip-and-replace objection to near zero.
One wedge. One design partner. One use case where "how + why + execute" is obviously valuable and measurable. Land one mid-market company bottom-up, prove ramp-time and agent-task-success, then widen. Enter on the wedge, not the category.