The Method
All legal work, everywhere, decomposes into six primitives.
Name them and you can see your practice clearly for the first time. Systematise them and your practice stops depending on memory. Automate them and you have an AI-native practice. That is the whole game, and it plays out in exactly that order.
Before you read on
This is the thinking behind how we build. For what we actually build and what it costs, see Services.
The fixed order
Name
Write every primitive down, in your own language. This is what makes a practice legible for the first time.
Systematise
Turn what you named into templates, playbooks, and role charts - so the practice stops depending on memory.
Automate
Only now do agents get deployed, on primitives already named and systematised. Skip ahead and you get faster chaos.
How the six fit together
Matter
The container every other primitive attaches to.
Primitive 01 of 06
MMatter
The atomic unit of legal work.
Most practices are organised around something else - the inbox, a person, a folder. The MATTER position is that the matter itself is the container. Every action, deadline, person, document, and note attaches to it. One matter, one place, one state. The test is brutally simple: could a competent stranger open a single matter and understand its complete state in ten minutes, without asking anyone? In a systematised practice, every day, for every matter, the answer is yes.
Why it matters
Every other primitive depends on this one having an owner. Actions belong to matters. Time is measured against matters. Teams are assigned to matters. Evidence is anchored to matters. Get the container wrong and every later system you build sits on sand - you can automate a badly-defined matter, but you'll only automate the confusion faster.
Ask yourself: If the handling lawyer went on leave tomorrow, could someone else open the matter and know exactly what happens next?
Read the full primitive - how to systematise it, what automation looks like, common failure modes→Primitive 02 of 06
AActions
Named, repeatable moves.
Within any practice area, actions repeat. A property dispute in one city and a property dispute in another are not the same case, but they are overwhelmingly the same sequence with different facts poured in. Senior lawyers experience this as expertise; it is really an action library stored in a human being, retrievable only by them and lost when they leave. Extract the library. Once actions have names, the lawyer's attention is finally free for the parts that are not routine.
Why it matters
"We use AI" is a meaningless sentence until actions have names. "Actions 1, 3 and 4 of our standard sequence run automatically, and a lawyer reviews before anything leaves the building" is an operating model. You cannot make the second statement until the first draft of the action library exists.
Ask yourself: Could a new hire follow a written sequence for your most common matter type without asking anyone a question?
Read the full primitive - how to systematise it, what automation looks like, common failure modes→Primitive 03 of 06
TTime
Managed, not merely billed.
The billable hour trained the profession to record time rather than manage it. Deadlines, which are absolute, and durations, which decide profitability, both go largely unmanaged. Treat time as a property of the matter, not a product sold to the client. Every matter carries its dates as structured facts. Every action, once named, accumulates duration data. Fixed pricing becomes arithmetic on data, not a guess.
Why it matters
Fixed-fee pricing is the single clearest external marker of an AI-native practice, and it is only survivable once time is genuinely measured. A firm that knows how long each named action actually takes can quote with confidence; a firm that only records time can only guess.
Ask yourself: If a deadline is three weeks out, does it surface itself, or does someone have to remember to check?
Read the full primitive - how to systematise it, what automation looks like, common failure modes→Primitive 04 of 06
TTeam
Roles attached to actions.
For every action on every matter: who does it, who reviews it, who decides. Traditional practice answers by hierarchy and habit; responsibility is discovered retroactively, usually when something goes wrong. Make it explicit and 'team member' stops being a category limited to humans. An agent drafts, a lawyer reviews, and the delegation structure can hold - because it was already there for people.
Why it matters
The reason most firms can't safely delegate to AI is that they never had a delegation structure for humans either - no defined doer, no defined reviewer, no explicit handoff. A practice that has systematised Team can hand one named action to a machine and keep review with a lawyer, and state exactly who is responsible for what. That sentence is the entire governance model of safe AI practice.
Ask yourself: For your five most common actions, could you name the doer and reviewer right now, without checking?
Read the full primitive - how to systematise it, what automation looks like, common failure modes→Primitive 05 of 06
EEvidence
Anchored, self-verifying work.
Everything a lawyer asserts stands on something: a document, a testimony, a precedent, a fact. In most practices, the connections between claim and source live in one lawyer's memory. The MATTER position: the connections are the asset. Every claim in every draft is anchored to its source. The practice becomes self-verifying - and the legitimate fear about AI, fabrication, is answered not by 'better AI' but by a standing rule that no assertion enters work product without an anchor.
Why it matters
This is the primitive where the AI question actually gets decided. An AI drafting inside a strict evidence-anchoring rule is safer than a tired human drafting outside one - because the rule, not the technology, is what makes the work trustworthy.
Ask yourself: Could any reader open every citation in your last major filing in one click and find the exact supporting passage?
Read the full primitive - how to systematise it, what automation looks like, common failure modes→Primitive 06 of 06
RResearch
A compounding stock, not a flow.
The traditional model treats research as an event: a question arises, someone researches it, an answer comes back, and the work evaporates into a folder. Next year the same question is researched again from zero. Treat research as a stock. Every answer compounds into a searchable body of firm knowledge, versioned and current. Machine speed on a curated stock is the compounding edge no rival can buy.
Why it matters
Machines are extraordinary at retrieval and synthesis over a well-structured stock, and dangerous at invention over nothing at all. A firm that treats research as a compounding asset gets faster every year it operates; a firm that treats it as a one-off event resets to zero every time a lawyer leaves.
Ask yourself: Could a colleague find your firm's prior answer to a recurring legal question in under two minutes?
Read the full primitive - how to systematise it, what automation looks like, common failure modes→
The claim, restated
"You cannot bolt AI onto chaos and get anything but faster chaos."