Don’t Vault Your IP. Route It.

When cognition is cheap, ideas are abundant and matching is scarce. The value event is a match — and a vault with no edges is worth zero until something can reach it.

I was texting my brother. He works in traditional BI. His company, Altus, is trying to get into AI the way a lot of product-line firms do — scanning the catalogue for “AI features” they already sell and pushing those features at existing clients. Not much theory underneath it. Just product-line gravity.

I have been writing about something adjacent for a while: BI for soft data — treating meetings, mail, decisions, and organisational exhaust as a second data plane, compiled into a wiki-shaped worldview agents can actually use. Their workforce already knows ETL, source discovery, identity, and semantic modelling. The adjacency is obvious. The fit is almost unfair.

Gary’s instinct was immediate: you need to protect the intellectual property.

Mine was just as fast, and it surprised me a little by how automatic it felt:

IP is only useful if you connect the right idea to the right person or project at the right time.

That is not a legal opinion. It is a matching opinion. And once you see IP as a matching problem, the whole “vault the frameworks” reflex starts to look like a value destroyer dressed up as prudence.

Protect what, exactly?

When someone says protect the IP, the image is usually a pile of valuable stuff sitting still. Folders. PDFs. A private Notion. An encrypted drive with the good ideas inside.

But the value does not appear when the file is created. It appears when an idea intersects with a live problem, a person who can use it, a capability that can execute, a company that can absorb it, and a moment that has not closed yet.

Imagine the canon as a directory tree:

/terminal-value-doctrine/
/context-arbitrage/
/bi-for-soft-data/
/wiki-as-memory/
/soft-join/
/cognitive-time-travel/
/authority-infrastructure/
/memory-mate/

You can encrypt the drive.

Congratulations. You have protected a cupboard.

The actual value event looks more like this:

IDEA          BI for soft data
× PERSON      Gary
× CAPABILITY  traditional BI / ETL workforce
× COMPANY     Altus
× PROBLEM     “we need an AI strategy”
× FAILURE     selling AI features in incumbent products
× TIME        right now

Suddenly it is not “an interesting framework on a website.” It has a destination. Gary sees the fit — which is exactly why the next thought is protect it. The reflex I trust more is: route it.

When cognition is cheap, matching is scarce

This inversion only becomes obvious when idea creation stops being the bottleneck.

When ideas are expensive to produce, protection is rational. You made one good thing. Guard it. When ideas are flowing — because you think with models, write in public, and compound a personal canon — the scarce layers move:

• Matching — which idea belongs in this conversation?
• Timing — is the problem live, or merely interesting?
• Application — can this organisation actually absorb it?
• Execution — who has the skills and the client access?

Extreme protection can reduce value by preventing the edges from forming. Had “BI for soft data” lived only in an encrypted folder, the conversation with Gary does not happen the same way. There is nothing discussable. Nothing to map onto Altus. Nothing to route.

Protected but unreachable → VALUE: $0

That is harsh on purpose. Not every secret should be public. Client data, unreleased product mechanics, and true trade constraints still deserve locks. This article is not a brief against counsel. It is a brief against treating framework IP — the named ways of seeing you have already earned — as if the cupboard were the product.

The economic backdrop helps. When key inputs get cheap, structure reorganises around new scarcities. MIT Sloan’s “economies of unscale” argument is the long version: traditional advantages of size invert, and smaller operators can win by pursuing exact niches rather than defending mass-scale inventory.¹ In an AI-native practice, the niche is often a match — a specific company with a specific failure mode and a specific capability stack — not a generic feature SKU.

The same pressure shows up in the open-innovation literature: knowledge that is meant to travel across boundaries sits in permanent tension with instruments designed for exclusion.² You do not resolve that tension by shouting “share everything.” You resolve it by being precise about what you are optimising. For framework IP under cheap cognition, the thing to optimise is collision probability, not silence.

The IP agent is the Gmail agent’s twin

I already have a working proof of the pattern in another domain.

My Gmail agent used to be dumb and annoying. Same cheap model. Different substrate. Once it could read a compiled personal world — projects, people, what normal looks like, what I already ignored — it became the brilliant silent one. When it surfaces something, I take notice, because it has earned the right to interrupt. My AI router can go a month without talking, then say a mesh point has been down for two hours and needs a physical operator. That is not more chatter. That is selective cognition.

The Gmail agent answers: does this matter to Scott now?

The IP agent should answer: does this idea matter to this person, project, or company now?

GMAIL AGENT
new email
  → compare against Scott’s world
  → surface only meaningful intersection

IP AGENT
new person / project / conversation / company
  → compare against Scott’s canon
  → surface only meaningful intersection

So Gary says: Altus, BI, AI, ETL.

The bad agent dumps the catalogue — here are your 186 intellectual frameworks. The good agent quietly places two or three edges in view:

BI for soft data
Wiki as compiled worldview
Context arbitrage

And shuts up.

That is the same primitive as an edge surfacers for memory: at the moment of context, surface the smallest set of edges capable of activating the relevant human graph. Not a lecture. Not a chatbot monologue. A chink. Then recognition does the rest.

Publishing is edge creation

Here is the part that still makes people flinch.

I publicly write “BI for soft data.” Gary works in BI. The concept becomes discussable. He maps it onto Altus. Now there is a potential application. The publication did not “leak the asset.” It created an activation surface — a tentacle the live problem could grab.

If you treat publishing as a content engine problem — cadence, funnels, topic clusters — you have wandered into a different article. This one only needs one claim: for framework IP, making the idea discussable is often how the match becomes possible. The edge is the product surface. More meaningful edges mean more ways a future event can collide with the idea.

The graph increases the probability of valuable collision.

That is sharper than “better retrieval.” Retrieval assumes the question already knows what to fetch. Matching assumes the world is full of almost-relevant conversations that will never form the right query string unless something in your system is already joinable.

This is also why a compiled canon beats a trophy folder. The durable layer is not the encrypted artefact. It is the context, design, and relationships around the idea — the same shift people are making when they admit code is ephemeral and the surrounding judgment is the asset.³ Your frameworks are soft data too. Soft data is not inactive because it is unstructured. It is inactive because it is poorly connected to the moments when it could create value.

A wiki is an IP matching engine

Put the pieces together and the corporate-sounding sentence becomes personal and practical:

A wiki is an IP matching engine.

Not a museum. Not a graveyard of PDFs. A graph of claims and edges sitting ready so that when Gary (or a client, or a random Tuesday conversation) appears, something can soft-join live context to latent idea.

The architecture is almost embarrassingly simple once you stop optimising the cupboard:

RAW IDEAS / NOTES / DRAFTS
        ↓
COMPILED CANON (pages + edges + names)
        ↓  latent
LIVE CONTEXT (person × company × problem × time)
        ↓
AGENT ROUTES → 2–3 intersecting frameworks
        ↓
HUMAN MATCH / CONVERSATION / APPLICATION

I have written the substrate version of this as The Index Is the Data: do the expensive relationship work ahead of query time; own the map.⁴ Context arbitrage is the economic twin — pay once to compile comprehension, then amortise it across cheap decisions. The Proposal Compiler / Marketplace of One is the commercial twin: when bespoke output gets cheap, advantage moves to matching a compiled kernel to one specific opportunity.⁵

This article is the IP-facing face of the same shape. Your canon is not a vault inventory. It is a matching surface.

What changed in one conversation

Before breakfast, roughly:

• Before: “BI for soft data” is a framework on a site — interesting, latent, unrouted.
• Context arrives: Gary + Altus + ETL capability + weak “AI features” strategy + a live need for direction.
• Match: the idea has a destination; protect-the-IP is the wrong first optimisation; route-the-IP is the right one.
• After: the framework is no longer abstract. It is an opportunity candidate — still not a signed deal, still not a product launch, but no longer value-zero inventory.

That is the minimum proof this piece needs. Not a revenue chart. A before/after on the match formula. Yesterday the idea existed. Today it became commercially relevant because the correct tentacle touched the correct live object.

Rebalance: from possession to collision engineering

If you only change one default this week, change this:

Stop asking “how do I protect this framework?” as the first question.
Start asking “what edge would let the right person collide with it while the problem is still live?”

A practical rebalance

• Name it. Untitled genius in a folder cannot be matched. Give the framework a short, speakable name.
• Edge it. Link it to industries, failure modes, capabilities, and sibling ideas in your wiki — not just to related notes for your future self.
• Publish selectively. Enough surface that the idea is discussable. Not a content factory. One clear public articulation can create the Gary-shaped edge.
• Instrument matches. Keep a crude log: context → frameworks surfaced → whether a real conversation activated. Optimise that, not page count in the vault.
• Build the twin agent. Same discipline as the silent Gmail agent. Two or three edges. Then quiet.
• Vault what deserves vaulting. Client material, credentials, unreleased mechanics. Do not confuse those with the named ways of seeing that only work when they can travel.

What this is not

A few scope fences, said once so they stay said:

• This is not legal advice about patents, NDAs, or trade secrets. Those instruments matter in their own lane. Hire counsel for that lane.
• This is not the full BI-for-soft-data product thesis — how a BI firm should redesign services around organisational soft joins. That is a different article.
• This is not a content-engine playbook. Publishing appears here only as edge creation for matching, not as a media business model.

The lane this piece owns is narrower and sharper: when cognition is cheap, your IP system should be a router, not a vault.

The cupboard or the edge

Gary said protect the IP. The old script agrees with him. The new scarcity does not.

Frameworks rotting in encrypted folders are not “safe assets.” They are disconnected soft data. They have no tentacles. They cannot find Altus. They cannot find the Tuesday conversation. They cannot find you, three months from now, when the same pattern reappears under a different name.

Route the idea. Give it edges. Let the graph raise the odds of a valuable collision. Teach the agent to surface two or three frameworks and then earn silence.

Don’t vault your IP. Route it.

If a framework of yours is currently living in a cupboard, name the person, company, or problem it should meet while the window is still open — and make one edge that could get it there.