The Matter Never Repeats. The Component Always Does.

Two rows, condensed from a table we already published:

Node 24 · US bank KYC package: formation certificate, EIN letter, operating agreement, FinCEN CDD beneficial-ownership certification · paralegal-ops · 2 hours · $190 (modeled) · ASSIST · "assembly automatable; bank-specific requirements vary." Node 59 · MX bank selection; KYC package (escritura, RFC/CSF, poderes, rep ID, comprobante de domicilio) · paralegal-ops · 2 hours · $190 (modeled) · ASSIST · "assembly automatable; bank checklists idiosyncratic."

Under the published automation model, an adopted ASSIST node returns 35 percent of its labor cost net of the review tax: $66.50 per run, modeled. "A Legal Transaction Has a Bill of Materials" summed every number like it across one matter's graph and printed the ceiling honestly: 23.5 percent of modeled matter cost at full adoption. That arithmetic divided $66.50 by one matter. It divided by the wrong thing.

And keep the second row's last clause in view: bank checklists idiosyncratic. The row that proves components repeat also printed, in our own handwriting, the reason practitioners say they don't. We will need both halves of that row before this is over.

Two prices for one task

The published rate card behind those rows is three lines: US attorney $450 an hour, Mexican counsel $250, paralegal-ops $95 · modeled market assumptions, stated in the open. The published automation model is four numbers: an adopted AUTOMATE node eliminates 80 percent of its hours, an adopted ASSIST node 50 percent, 30 percent of every saved hour comes back as human review priced at the node's own rate, netting 56 and 35 percent of labor cost respectively.

Read that model closely and it contains a claim nobody priced. The node's economics are a product of two terms: what the choreography costs to run, and the rate of the human it stands next to. The first term belongs to the component. The second belongs entirely to the practice around it. The KYC package in Node 24 · the same document choreography, the same 31 CFR 1010.230 beneficial-ownership certification logic, the same escalation paths when a bank's checklist deviates · exists in every closing that touches a regulated account. The component is identical. The rate card it sits inside is not. One task, two prices, and the difference between them was never an engineering variable.

If one task has two prices, then the capital map of legal AI was drawn on the wrong premise.

The entrants are evidence for, not against

The 2024-26 wave put its weight where repetition is visible at the matter level: Crosby's Sequoia-backed seed and $60 million Series B, Eudia's $105 million Series A, Covenant's 2025 seed round. That is a claim about where the capital concentrated, not about where all of it went, and the intellectual warrant for the whole trade was written early: Sarah Tavel's August 2023 case that AI founders should sell the work itself rather than software, because the market prices work against the cost of the human who would otherwise perform it, not as a productivity gain.

Now look at what the complex-work entrants actually productized, because none of them productized a matter. Covenant, built for private-market investors, productized LPA review and MFN election analysis · across funds that are never the same fund twice. Crosby, a law firm in its own right with in-house AI, productized the negotiation of MSAs, DPAs and NDAs across clients that share nothing but the document types. Eudia bought recurring process capacity outright, acquiring the ALSP Johnson Hana and its 300-plus legal professionals to serve Fortune 500 departments. The fund is unique; the MFN election analysis is not. The client is unique; the DPA turn is not. Every one of these companies is priced as proof that complex legal work resists productization, and every one of them is operating proof of the opposite: complexity lives in the wrapper, repetition lives in the components, and the components are where the product went.

Which collides head-on with a number we published.

Dividing by the right thing

The collision resolves cleanly, and resolving it is this memo's job. Every finding in the BOM memo stands untouched at its own denominator. The 23.5 percent ceiling is real: on one matter, automation nets a quarter of modeled cost, the 40 percent floor never moves, and the durable position sits at the institutional nodes on the notario line. The memo also found that the AUTOMATE drafting column is where competition will be fastest and margin thinnest, because every entrant gets the same models. Nothing below relocates any of that. The moat does not move in this memo.

What moves is the denominator. A matter is the client's unit of account. A component is the operator's. Build the KYC choreography once and the matter lens sees $66.50, a rounding error inside one engagement. The portfolio lens sees the same component running in every formation, every credit facility, every subscription close and every signing the practice touches in a year, each run returning its spread against whatever human it displaced in that wrapper. Matter-level automation is a cost story, and the BOM memo told it to its honest end. Component-level automation is a revenue line that the matter lens is structurally unable to see, because no single matter's economics justify building any component · and a firm's matter-by-matter accounting never has to confront what the portfolio would have paid for it.

Two distinctions keep this honest. Margin capture is what this memo prices: the entity operating the components across the portfolio earns the spread per run. Defensibility is what the BOM memo priced, and it stays where that memo put it: at the statute-backed institutional nodes, not in any component. And whether the operator keeps the spread at all is a fee-clause question that belongs to "The AI Dividend Has an Address" · one line here, the full mechanics there.

The nearest-human price

Tavel's pricing rule, taken one step she did not take: if AI-delivered work is priced against the human who would otherwise perform it, then the anchor is not a constant. The component is identical everywhere; the nearest human is not. Displace paralegal hours and the component's revenue anchors at $95 an hour. Move the identical choreography into a wrapper where the displaced hours belong to Mexican counsel and the anchor is $250 · 2.6 times the spread for zero additional engineering. Where the displaced hours are US attorney hours, $450, and the multiple is 4.7. Her warrant for why the component can hold that anchor is also already written: an AI system with consistency and service levels "should be vastly superior to an outsourced offering," which is to say the anchor holds because the component outperforms the human at the human's own price point, not because anyone is fooled.

Call this the nearest-human price: every automated component is priced against the nearest human it displaces, the component is constant, the human varies by neighborhood, and the spread between run cost and nearest-human price, summed across a portfolio, is the income statement the matter-level lens cannot see.

One discipline, before the arithmetic gets enjoyable. The multiple is earned only where the displaced hours genuinely sit with the attorney. Assembly hours do not: a document checklist is staffed to the cheapest competent person in every band of practice, in corporate towers as in boutiques, and a model that prices KYC assembly at $450 an hour is a billing fantasy. What sits higher on the card is the judgment slice of the same component · who counts as a beneficial owner through two trust layers, whether the control prong reaches the independent manager, which deviation in a bank's idiosyncratic checklist is noise and which is risk. An honest ledger anchors the assembly hours at the paralegal rate everywhere and lets only the judgment slice scale with the wrapper. The ledger below does exactly that, and the practitioner's version of this objection gets its full hearing two sections down.

The component ledger

One component, traced across four wrappers; the other three components named by the sweep get one summary line each. The reference portfolio is stated first, because a ledger without a stated portfolio is a number waiting to be backed into: a modeled reference portfolio of 12 cross-border formations, 6 credit-facility closings, 2 fund subscription closes of 40 subscribers each, and 3 M&A signings per year · proportions set to mirror the corridor mix of the practice Project Saltillo modeled, scaled to a boutique's annual throughput. Every count is a model input, printed so it can be disputed. All figures use the published rate card and the published 56/35 automation model; rows marked published are measured against the BOM table; everything else is modeled and says so.

One component, four wrappers · KYC choreography (assembly + 1010.230 certification logic + escalation)

The other three components, one line each (modeled, same portfolio)

Reconciliation, in one line: same component, two denominators · $66.50 per run inside one matter's published 23.5 percent ceiling, $9,429 a year as a portfolio margin line at blended corporate anchors, modeled under the flat-fee regime "The AI Dividend Has an Address" analyzed; under cost-plus, the same automation deletes the line.

Read the total honestly: four components at boutique throughput model out to roughly $14,800 a year. That is a real margin line where the matter lens reported a rounding error, and it is not a headline number, and both facts matter. The line scales on exactly three knobs, each inspectable: runs (portfolio throughput), catalog (Saltillo alone contains 54 automatable nodes, and the cross-matter catalog of a full practice is dozens of components, not four), and anchors (every component whose judgment slice genuinely sits with attorneys). We decline to print what the ledger yields at mid-market throughput with a full catalog, because deriving toward a pre-announced number is the exposure-percentage sin the BOM memo was written to prosecute. The model is on the table. Run it at your own scale.

Notice also what the ledger does that a thesis cannot: it sorts. One component carries blended anchors, one scales clean at the attorney rate, one is half fake and printed as such, and one fails the corporate-anchor test entirely and stays in the table anyway. A component ledger that contains no null rows was backed into.

Three old maps

Susskind. Legal decomposition in the Susskind tradition is intra-matter and sourcing-oriented: take the matter apart, send each piece to its cheapest competent producer. It is a procurement map, and it is correct as far as it runs. But it cannot tell you which parts are worth owning, because that answer lives in the portfolio · how many times this component runs per year, against whose rate · and a firm's matter-by-matter accounting has no ledger on which the question can even appear.

Christensen. Modularization theory predicts value migrates to modules once the interface standardizes. The interface of complex legal work never standardizes: every wrapper stays bespoke, which is precisely why the field reads complex practice as automation-proof. The ledger above shows components repeating beneath an interface that never standardizes. This is not low-end disruption climbing upmarket on a standardizing product. It is component capture inside a product that never modularizes, a case the theory does not have a shelf for.

Greenberg. A March 2026 Bloomberg Law analysis argues that AI-native firms will disrupt the cash-strapped legacy law firm model · a gradient that runs from the volume end upward. The rate arithmetic points the gradient the other way. The spread per run is widest where the nearest human bills the most, so the economics pull componentization toward complex practice, not away from it. The disruption literature watched the cheap end of the market because that is where matters repeat. Components repeat everywhere, and they pay best at the expensive end.

The practice pushes back

Here is the strongest version, in the practitioner's voice, conceding nothing:

Components do not exist outside the matter. The reason no two matters repeat is the reason no component is truly identical: my disclosure schedule is shaped by this deal's reps and this counterparty's counsel; my KYC choreography turns on which bank officer answers · your own table says so, "bank checklists idiosyncratic," your words. Standardize the component and you have manufactured the malpractice case: the reused escrow mechanics that miss the one term that differed. I supervise under Rules 5.1 and 5.3, and ABA Formal Opinion 512 makes me review the output regardless, so your component arrives on my desk as more review, billed at my rate. Clients buy outcomes from a lawyer whose name is on the opinion; they will not pay for an assembly line. And your arithmetic is worse than your ethics. Nobody staffs KYC assembly to a $450 lawyer · commodity work goes to the cheapest competent person in my shop exactly as in yours, so the nearest human for a checklist is always a paralegal and your multiple never existed. And whatever spread is real: what protects it? You published the finding yourself · drafting margins go to thinnest fastest because every entrant gets the same models. Your component spread compresses the same way, and you have already admitted the moat lives at the notario line, not in your components.

Six answers, conceding what must be conceded.

First, supervision is priced, not waved at. The published model adds back 30 percent of every saved hour as review and escalation at the node's own rate; Opinion 512's review duty is inside the unit economics, not adjacent to them. The 35 percent net is the post-supervision number.

Second, conceded in full: the component belongs on the operator's income statement and never on the client's invoice. The client buys the matter, the outcome, and the name on the opinion, full stop. Nothing here proposes selling clients an assembly line. The ledger is the operator's economics for delivering the same matter, and "The AI Dividend Has an Address" governs whether the operator keeps any of it.

Third, the variance objection is the moat argument wearing a frown. "Bank checklists idiosyncratic" is not evidence against the component; it is the specification of it. A template fails on exactly that idiosyncrasy, which is why a component is not a template: it is choreography plus the escalation judgment about which variance matters, and that judgment is learned across runs no single matter can teach. The practitioner who has closed with one bank knows that bank. The component that has run across forty knows which deviations are noise. The variance is the part worth owning.

Fourth, conceded plainly: some claimed components are fake. Where the variance lives in the component rather than the wrapper · a disclosure schedule that is the deal · it is not a component, and an honest ledger says so in print. Ours does: the schedule's content is excluded, only the choreography slice is priced, and apostille logistics is carried as a null row.

Fifth, the staffing objection is half right, and the ledger is built on the half that is. Conceded: assembly hours are paralegal-priced in every band, and a flat 4.7x on a checklist is a billing fantasy; that is why every corporate row above blends. Not conceded: that the judgment slice is a fantasy too. The 1010.230 certification call, the escalation on the deviant checklist, the release-condition logic in an escrow · in corporate practice those hours sit with attorneys today, which is the only thing the nearest-human price requires. And the record on what corporate practice actually bills for commodity work is not flattering to the objection: clients found enough attorney-billed routine work on their invoices to go to war over it, and the outside-counsel-guidelines genre exists because the billing existed. The honest anchor is blended; blended is what we printed.

Sixth, the compression objection deserves a straight answer, because our own series sharpened it. Conceded: the component spread is margin, not moat. Defensibility stays at the institutional nodes, exactly as published, and the generation slice of every component compresses toward zero because every entrant gets the same models. Two things slow the spread's decay, and neither is a fortress. The escalation judgment in answer three is portfolio-trained: it does not ship with the model, and an entrant starts its run count at zero. And the spread reprices at the speed of disclosure, not the speed of model releases: complex practice prices outcomes, the parts list is opaque, and the BOM memo already explained why every incumbent is structurally committed to keeping it that way. So treat the component spread as "The AI Dividend Has an Address" taught us to treat the dividend: a term annuity, with a duration set by run-count lead and rate-card opacity, not a perpetuity. An operator should hold both maps · the ledger for income, the notario line for position · and should never confuse them. We just spent a memo not confusing them.

Ask for the run count

The diligence instruction, series style, is one sentence: do not ask whether the practice repeats; ask which components repeat, how many runs a year, and against whose rate. A practice that answers in matters has not looked. A deck that answers in exposure percentages has been answered already, by our own earlier memo. Where the durable position sits is "A Legal Transaction Has a Bill of Materials"; who is contracted to keep the spread is "The AI Dividend Has an Address." This memo adds the middle term: the matter never repeats, the component always does, and the entity built to run components across a portfolio · at the nearest-human price, on its own income statement, behind a disclosed fee clause · is the MSO.

---

This memo is published by Jopese, a legal management services organization operated by HIRO PARTNERS LLC, a Texas limited liability company. It is offered for educational and analytical purposes only. It is not legal, tax, or investment advice, and it is not an offer to sell or a solicitation of an offer to buy any security or service. Jopese is not a law firm and does not provide legal advice or legal services; legal services are delivered by an independent law firm under a separate engagement in which Jopese does not participate. Project Saltillo and the reference portfolio are modeled, representative constructs, not client engagements; no client information appears in this memo. The component ledger is a model built on stated assumptions, printed so that any reader can recompute or refute it. References to specific funds, firms, investors, publications, transactions, and regulatory developments are drawn from public sources and are provided as market commentary, not as an endorsement, a recommendation, or a representation of any relationship. Questions and competing ledgers: rene@jopese.com.