Every law firm owns almost everything it needs to know about itself.

The documents. The matters. The ledger. Every client. Every counterparty. Every relationship. Every piece of experience. Every judgement its people have made over decades.

And yet ask a firm a question that crosses those things — a real question about itself — and in most cases it cannot answer.

Not because the information is missing.

Because the understanding of that information does not belong to the firm.

Its experience is curated inside one application. Its client relationships sit inside another. Its documents are searchable through another. Its financial truth is somewhere else. Its risk position is assembled inside systems it licenses every year.

The firm owns the raw material.

It rents the understanding.

And for thirty years, that was entirely rational.

The problem was technically extraordinarily difficult. Different systems describe the same client differently. The same matter differently. The same person differently. Permissions conflict. Structures conflict. Records are incomplete. External reality changes.

To create one coherent view of the institution, firms historically had two choices.

Move everything into another system.

Or buy another application that assembled one particular part of the truth.

So that is what the industry did.

One application for conflicts. Another for CRM. Another for experience. Another for pricing. Another for documents. Another for business development. Another for knowledge.

Each vendor assembled the world it could see.

And the boundary of the product became the boundary of the intelligence.

That constraint is disappearing.

Which changes much more than legal software.

It changes who can own the intelligence of the firm.

Here is why this has suddenly become urgent.

Law firms are deploying AI agents.

Not copilots that write an email.

Agents that perform work.

And the moment an agent performs real work, it stops living inside one application.

A simple task can cross the document-management system, practice management, finance, CRM, email and external information before it is finished.

Which creates a problem the industry has not solved.

Who governs that agent?

The document vendor cannot, because it does not own finance.

The finance vendor cannot, because it does not own the documents.

The CRM vendor cannot, because it does not own the matter system.

And the model provider cannot, because it does not understand the firm's institutional rules.

Nobody owns the whole path.

So nobody owns the governance of the whole path.

Neil Cameron at Legal IT Insider described this missing layer as the spine between the applications a firm buys and the data those applications consume.

The important point was not the metaphor.

It was that nobody owned it.

And there is a very simple architectural truth underneath.

You cannot govern what you have not resolved.

You cannot enforce a client wall around a client if that client exists as six different identities across six systems.

You cannot ask an agent whether the firm can act for somebody if the relationships required to answer the question sit across conflicts, CRM, finance, matters, documents and the outside world.

You cannot create firm-wide AI on application-level context.

So the enforcement layer everybody suddenly needs sits on top of a resolution layer nobody previously built.

Which is our opening.

And it is why we start with conflicts and anti-money-laundering.

Not because we are building another compliance product.

Because conflicts and AML are among the first places where the economic value of resolving the entire estate becomes immediately obvious.

The answer does not exist in one system.

It exists in the relationships between all of them.

And now something very important is happening.

The market is independently arriving at the same architectural conclusion.

Within six days of each other, the two leading document-management vendors announced context layers.

NetDocuments called theirs a Legal Context Graph.

iManage called theirs a Context Fabric.

Think about what that means.

Two incumbent vendors. One week. Almost the same architectural response.

The repository is no longer enough. AI needs context. Agents need relationships. The flat application has to become something closer to a graph.

This is not competition invalidating our thesis.

It is the market validating it.

The argument is no longer whether a persistent contextual layer is required.

The argument is where that layer starts.

And that distinction is fundamental.

They start with the application and move toward the estate.

We start with the estate and move toward the applications.

Their graph begins with what their product owns.

Ours begins with what the firm owns.

They ask: how are the things inside our system related?

We ask: what is true across every system the firm operates — and beyond them?

Those sound superficially similar.

Architecturally, they are opposite.

Because an application vendor ultimately unifies by expanding the application.

Every new problem becomes another module, another capability, another reason for more of the estate to converge on that vendor.

LegalFab does not require the firm to converge on us.

We converge on the firm.

The records remain where they are. The DMS remains the DMS. The practice-management system remains the practice-management system. The CRM remains the CRM. The ledger remains the ledger.

We discover the estate, catalogue it, resolve the identities and relationships across it, apply governance to that resolved layer, and allow intelligence to operate over the result.

No second institutional copy. No requirement that the firm surrender its data estate. No big-bang migration.

The firm finally owns the intelligence layer above the systems it already owns.

Which is the difference between a graph of a vendor's repository and a graph of the firm.

And there is a larger way to understand what follows from that.

For thirty years, the legal technology industry sold firms applications because applications were the only practical containers for intelligence.

You bought the application because you needed the intelligence trapped inside it.

AI changes that. Resolution changes that. Agentic execution changes that.

Once the institution itself can become the foundation, intelligence no longer has to live inside the application.

And when intelligence becomes independent of the application, the application has to justify its existence again.

Some will. Some will not.

We call it Technology Darwinism.

We are not asking firms to rip out their systems.

We do almost the opposite.

Keep them. Resolve across them. Govern above them. Build better capability on the shared layer.

And over time, if an incumbent application still adds enough value, it survives.

If a thin capability on the shared foundation does the job better, the firm simply stops renewing the old application.

No migration programme. No replacement day. No enormous transformation project.

The estate evolves.

And you can see it in the capital markets.

Intapp was worth five point seven four billion dollars in February last year.

It is worth about three point three billion today.

Down forty-two percent.

This is not a story about Intapp becoming a bad company.

Quite the opposite. It is still growing. A hundred and forty-two clients still pay it more than a million dollars a year.

A healthy incumbent can keep growing while the market changes its mind about which layer is strategically valuable.

That is what repricing looks like.

The value is migrating.

And then look at what the customer is doing.

Kirkland and Ellis has committed five hundred million dollars over three to four years to build a proprietary, model-agnostic AI capability.

More than a hundred million of it this year alone.

Over a hundred and eighty AI engineers and data scientists. Two hundred and fifty attorneys involved. A Palantir relationship at the centre of it.

Why?

Because at sufficient scale, firms understand that merely adding another AI assistant does not solve the institutional problem.

Eventually you need the foundation.

Kirkland can spend half a billion dollars building one.

Almost nobody else can.

We productised the foundation.

And once the foundation exists, the economics of software itself change.

Traditional enterprise applications are thick. They have to be.

A vendor serving hundreds or thousands of firms has to carry its own data model. Its own store. Its own permissions. Its own audit framework. Its own search. Its own integrations. Its own interface.

Then it spends months configuring all of that generic machinery until it approximately resembles the individual firm buying it.

And somewhere inside that enormous structure is the thing the customer actually wanted.

The judgement that performs the work.

We think about it as eight layers.

Seven layers of scaffolding. One layer of actual domain intelligence.

The industry has historically charged enterprise prices for all eight, because there was nowhere else for the other seven to live.

But resolve the estate once — and the economics reverse.

The data model already exists. The permissions already exist. The audit exists. The search exists. The integrations exist. The interface framework exists. The institutional context exists.

So when somebody wants to build a capability, they do not need to rebuild the company underneath it.

They need to encode the judgement.

Which is why an implementation that historically takes nine months can become something built in weeks.

And it is worth being explicit about how these things connect.

Because they are usually presented as three separate arguments — a wedge, a moat, a flywheel.

They are one chain.

We enter through one high-value workflow. Conflicts. Anti-money-laundering. Onboarding.

Work with a budget holder, a deadline, and an obvious cost of getting it wrong.

Answering that workflow requires resolving the systems it crosses.

So the firm ends that first engagement with something it did not have before.

A resolved view of itself. Which it owns.

Which means the second workflow is not another procurement cycle.

Seven of the eight layers already exist. It is weeks.

And as more work runs on the resolved layer, governance moves with it. Walls, permissions, lineage, audit — enforced in one place rather than approximated in six.

The account becomes difficult to unpick.

Not because we locked anything.

Because the firm's own institutional state now lives there.

Then the last two steps, which are the ones people miss.

What was built for one firm can be offered to the next. Not the data. The capability.

And every deployment lowers the cost of the one after it.

One workflow in. An operating layer out.

Not copied blindly. Not imposed as generic software. Acquired as capability, then re-specified to the second firm's own data, rules, permissions and institutional judgement.

Firm one does not just become a customer. It contributes to what can exist for firm two. Firm two contributes to what can exist for firm three.

And this is the part I think is easiest to underestimate.

Our north star is not “build a marketplace someday.”

The marketplace logic is why the architecture looks the way it does today.

It is why we resolve rather than migrate. It is why records remain in place. It is why the foundation belongs to the firm. It is why utilities are thin. It is why anyone should eventually be capable of publishing.

Because nothing valuable compounds on infrastructure from which the customer can ultimately be evicted.

The moat is not simply the Knowledge Fabric.

The moat is what accumulates because the Knowledge Fabric exists.

And that brings us to the first ten firms.

Legal is an unusual market. Highly concentrated. Highly referential. And extraordinarily sensitive to what peer firms are doing.

Andreessen Horowitz described this dynamic in relation to Harvey.

Every major firm that adopts reduces the perceived risk for the next one.

The first white-shoe firm moves. Then another.

And the question changes from “should we do this?” to “why aren't we doing this?”

Harvey demonstrated that domino effect at the application layer.

Something similar happens at the infrastructure layer.

But there is one critical difference.

For us, the first ten are not simply ten contracts. And their significance is not their total contract value.

They are the reference set for an architectural decision.

After ten credible firms have examined the architecture and chosen to proceed, the eleventh conversation begins from a different place.

The architecture has been validated. Not because we say so. Because ten sophisticated institutions examined it.

The perceived risk falls. Peers are visibly moving. The category becomes legible.

And then there is the fourth effect.