A Goodwin corporate partner is launching a US practice attached to a Swiss company that raised seed money as research software and is now trying to sell legal outcomes. Leon Johnson left Goodwin to build that US arm of Omnilex. The résumé will draw the attention. For anyone who prices expertise by the hour, the sharper question is what happens when the first draft stops being the expensive part.
He laid out the model on 14 October 2026, in an interview with Richard Tromans of Artificial Lawyer. The US practice, he said, had already been established. This is a product company turning a tool for lawyers into work for clients. A century-old partnership with a chatbot taped to the intranet is the wrong comparison.
Two chapters, one name
Johnson said Omnilex was founded in Switzerland a couple of years before that interview. It began as a legal-tech platform sold to attorneys. The founders later decided the better use of the technology was to deliver the work themselves. He described that turn as a shift still underway, not a finished history.
The funding write-up belongs to the earlier chapter. LegalTechTalk reported a $4.5 million seed round, about €3.8 million, for the Zurich company. The money came from investors in Switzerland, the United States, the United Kingdom, and Germany, including Founderful, Plug and Play Tech Center, Tiny Supercomputer Investment Company, Angel Invest, and Pascal Mathis. No lead was named.
At that point LegalTechTalk called Omnilex an AI-powered legal research startup, built to simplify research for law firms and in-house teams. Founder Marco Henri’s stated ambition was “legal commentary that surpasses existing literature in scale, actuality, AI fitness and undiscovered niches.” That is a pitch about commentary. It is not a US law firm. A seed round for research software is not a book of deals.
You are buying the review, not the verb
Johnson’s line for how the work moves is short enough to sell, and loose enough to hide the hard part.
takes in your matter, solves your matter and then escalates it to a lawyer for review.
That is the platform’s job, in his telling. In the same answer he said the lawyer stays in charge from start to finish, checks accuracy, signs off, and delivers. Those are not the same claim. “Solves” is a sales verb. Responsibility is a professional one.
No public figure here measures how often the lawyer rewrites the first pass, which systems sit under the platform, or what review consists of beyond the promise that a human remains accountable. If you are buying AI automation, you are purchasing that review standard, not the verb.
He is starting where the documents are most standardized. The US practice, he said, would open with venture capital financings, startup governance, and commercial contracts. He called that a position of strength because US venture work leans on NVCA forms, with comparable forms in the UK. He did not describe those forms. He did describe the old assembly line: term sheet to documents in 24 to 48 hours, a junior drafts, a mid-level reviews, a partner signs, and the client pays for several billable hours.
Anyone who has funded a company will recognize the portrait. The documents rhyme. Both sides usually want the round to close. A first draft that used to eat a night is where software looks brilliant. It is also where a faster template gets mistaken for a new kind of firm.
The billable hour is the business model
Johnson invokes Clayton Christensen’s The Innovator’s Dilemma for the reason he left: a billable-hour pyramid will not readily adopt a tool that erases the junior hours it sells. He has praised Goodwin. The argument is about the pricing system, not a claim that his old firm is asleep.
Clients, he argues, want an outcome at a set price. A contract review. A venture financing. A closing. A pyramid that bills the junior’s draft has a structural allergy to a machine that drafts for free. Speed does not drop to the bottom line. It evaporates, or the firm finds another task to bill.
Outcome pricing is the proposed cure, and the hard parts are still missing. A set price has to say how many turns are included, what counts as out of scope, and who eats the fourth draft. None of that has been published. Predictability is the pitch. The revision policy is the product.
His exit is also not proof that Big Law has failed to adopt AI. Using a drafting tool inside a traditional firm is no longer exotic. Rebuilding the offer around it is the part that collides with how partners get paid. He left a system he thinks cannot survive a thinner staff. This interview does not show that incumbents are frozen.
Tromans calls firms built this way AI-native, and sometimes “new model.” Those are his labels, not a census. His harder point is the one the venture wedge does not answer. In a cooperative financing, both sides want to sign. In litigation and in ugly M&A, software that surfaces more issues can create more negotiation, not less. The efficiency gets bargained away. No one has published data showing which way that cuts.
The cap table nobody demos
Tromans stated, and Johnson agreed, that the US setup is an MSO, a management services organization. In the interview’s shorthand, a regulated law firm links to a technology company through a licensing agreement, so investors can fund the tech company where they cannot simply buy the practice.
Do not let that shorthand harden into law. Arizona and Utah are the main American experiments in letting non-lawyers own law firms. Britain has a formal regime for it. Australia is a patchwork, not a yes. An MSO is a different, widely used structure. It is not a license to ignore the bar where the lawyers actually practice, and nothing in this interview blesses Omnilex’s arrangement under California rules or any other.
Holland & Knight, in an October 2025 commentary that does not mention Omnilex, sketched the fences these deals are supposed to respect. Non-lawyer equity stays in the MSO. The law firm keeps 100% of the legal entity and of the legal fees. The MSO is paid a flat fee or fair-market value, not a cut of those fees, and it does not direct professional judgment, control the client relationship, or loosen confidentiality.
If a vendor’s system can see client files, access rules are part of the ethics question, not an IT afterthought. The same piece cited Texas Committee on Professional Ethics Opinion No. 706 (2025) and said Texas was then the only jurisdiction with an ethics opinion aimed squarely at law-firm MSO partnerships. That snapshot predates the Omnilex interview by about a year. Treat it as context, not a clearance.
Serious deployment of this kind costs real engineering money, which is why an MSO can matter. The structure is not right for every firm. For a general counsel, the decisive facts are still the missing ones: who owns the law firm, who owns the MSO, how the license is priced, and which bar rules govern the lawyers signing the work.
Questions worth asking before you hire one
- Who owns the law firm, who owns the MSO, and who signs? Ask for that chart before you send a file. Outside capital is supposed to sit beside the practice, not inside the legal judgment or the fee.
- What does the set price include? Demand the number of turns, the out-of-scope line, and who pays when the documents come back a fourth time. Without those, outcome pricing is a slogan.
- Does the software replace the lawyer? No. Johnson says the platform takes the first pass and a lawyer reviews, stays in charge, and signs, so the thing you are hiring is that review.
- What will the US practice do first? Venture financings, startup governance, and commercial contracts, where the forms are repetitive and both sides usually want to close. Do not treat that wedge as proof the model survives a contested deal.
- What has not been published? Rewrite rates, the systems in use, a definition of review, a fee card, a revision policy, and the license terms. Those missing pages are the diligence.
Reputation will ration this market faster than model quality will. When a deal can end someone’s job, buyers hire the name they do not have to explain to a board. Unfamiliar firms earn that trust only by doing routine work visibly and having it hold up. A clean first draft on NVCA paper does not purchase it. That is a market effect, not a feature you can demo.
For anyone weighing AI for business inside a professional firm, the stress test is boring. If the first draft becomes cheap, the pyramid loses both cost and apprenticeship at once. The firms that last will be able to show a human still owns the judgment, a price that survives the second and third revision, and a cap table that does not quietly own the client. Johnson has put a Goodwin résumé on that bet. The résumé is real. The operating proof is still mostly ahead of the adjectives.