How do you capture tribal knowledge with AI agents?
You capture tribal knowledge by working alongside the people who hold it until an agent reaches the same answers they do, then running that judgment on every case. HomoDeus did this for Galapagos Capital's house view, a São Paulo firm's precedents, Grupo Raízes's contracts and a US brokerage's filings.
How you know it worked
The agent starts finding what the people could not. Keystone read one sponsor's model against what the building really spent over twelve months and surfaced four expense lines under real spend, insurance by a factor of two and a half. The desk's finished model had carried them through. Read against the real numbers, the return fell by more than half.
Keystone, Galapagos Capital
Keystone captures the house view of Galapagos Capital's US real estate desk, the judgment that lived in a few analysts' heads, and runs it against every sponsor package the day it arrives. We came to read sponsor packages. The hard part sat somewhere else: capturing the house view itself, the instinct that tells an analyst a number is wrong, and giving it to something that runs on every deal. A sponsor's model is written to sell the deal. The agent reads it against what the building really spent over the last twelve months, and on one deal it surfaced what the desk's own finished model had carried through: four expense lines under real spend, insurance by a factor of two and a half. Read against the real numbers, the return fell by more than half.
Tribune, A São Paulo Law Firm
Tribune turns fifteen years of a law firm's own proposals and drafts into one live base any lawyer can query, so a client email comes back as a firm-formatted, precedent-cited proposal in minutes, behind the firm's walls. A law firm's real asset is its own paper: every fee proposal, every draft, every opinion it has written. None of it was in one place, so a proposal the firm had effectively written ten times got written again from a blank page, from whatever precedents one lawyer happened to keep.
Ledger, Grupo Raízes
Ledger reads every one of Grupo Raízes's custom client contracts, learns what each variable means, and computes the month's invoice against the operation's real numbers, so the billing team stops chasing branches. A billing analyst knew that a column called CTE meant one client's constant, and why. That meaning lived nowhere the computer could read it, so no system could ever compare the right numbers.
Scribe, A US Brokerage
Scribe turns a brokerage's client onboarding from a month of compliance paperwork into a packet built in minutes by a team of agents, with a compliance officer signing at the end. Each custodian wants the packet its own way, with its own quirks. What a clean filing looks like, and the firm's compliance rules, lived in the senior analysts' heads.
Questions
Where does this knowledge usually live?
In a few people's heads and the spreadsheets they rebuild: a desk's underwriting, a firm's pricing, the meaning of a billing column, what a clean filing looks like.
Does the knowledge leave when a person leaves?
In Tribune the know-how that used to leave with each lawyer stays in the base, and every proposal the agent drafts re-seeds it.
Can the agent learn new terms?
Yes. Ledger asks two short questions when a term is new and keeps the answer, so it asks less every month.