The seventh revolution in law.

Seven revolutions, two patterns.

Software has compute per dollar. Legal work has never had an equivalent number, which is why arguments about legal technology end in anecdote. There is one, and it has fallen at every revolution: what a single authoritative answer costs, in the two currencies a practitioner feels. How far you went to reach the text, and how long it took to get an answer out of it. The first has hit its floor. The second has not.

The second pattern is the spacing. They arrive closer together each time, and the last three fall inside a single working lifetime.

Every one made lawyers more capable, not less necessary. The work moved up a level each time, and the firms that moved first set the terms.

The expert-witness market has not had its turn. A first strategic read still takes weeks and a specialist’s hours, for the same reason legal research once took a week in a reading room. That is the part a machine can now do. The parts it cannot, signing the report and taking the stand, stay where they are.

Having it has never been the same as using it.

Each of those arrived long before it was general, and the gap is the part firms actually live through. The first two are also a different kind of thing: the Twelve Tables and Justinian’s compilation changed who was permitted to reach the law, and institutional changes land over generations. The Tables were public for a century and a half before the procedure for invoking them was. Everything after them changed what reaching the law cost, and those land in decades. The lag has been collapsing ever since, and is now shorter than a single matter.

Twice it stopped collapsing, and the second time is now. Predictive coding has been approved and defensible since 2012 and is still named in 22 percent of responses on how electronic evidence gets reviewed; three quarters of litigators say the reason is that they do not know the technology, which is not the same as doubting it. Machine reasoning moves faster: 30 percent report their offices using AI tools, 45 percent expect it mainstream within three years. Roughly a third of the profession has this. The other two thirds include the lawyers on the far side of the table.

The experts are already using it.

Not as a forecast. In August 2026 a Harris County jury returned $61.5M against 3M over the Watson Grinding explosion, and the defence engineering expert, billing $475 an hour, had produced a report that both sides at trial put at 85 to 90 percent ChatGPT. Discovery turned up roughly 350 pages of his prompts, one of which asked the model to show how 3M is 0% at fault. That account comes from trial reporting rather than a written ruling, so take the verdict as the documented part and the prompts as reported.

The written rulings say the same thing more carefully. A Stanford expert on AI and misinformation had his declaration struck for two fabricated citations he had not written himself. A creditor’s valuation expert filed a 172-page report produced in 72 hours that the court found was “written by artificial intelligence at the instruction of” the expert, and excluded it. A damages expert was excluded for relying on an analytics tool whose algorithm he could not explain. By mid-2026 one academic tracker counted over 1,600 AI-hallucination cases across all filings.

The line courts are drawing is judgment against substitution, not AI against no AI. The same opinion that struck the Stanford declaration said plainly that it did not fault the expert for using AI to research. Four months later a court declined to exclude an engineer who wrote his report from decades of experience and then used a model to check it. That is the standard, and it is being set now, by courts, while most of the profession has not noticed there is one.

This is what the seventh revolution actually looks like arriving in this market: not carefully, and not from the vendors. The work is being done by people typing into a chat window at $475 an hour with nothing checking the output. Every design decision on this site, the verified quotes, the dropped citations, the refusal to produce anything that gets filed, is an answer to that paragraph rather than to a competitor.

Where this goes.

The direction of travel is not subtle. Analytical capability that was scarce and expensive is becoming abundant and cheap, and it is not stopping at the level it reached in 2025. Three curves carry it: whether a machine can do expert-grade analysis at all, how long it can work unsupervised, and what a piece of that work costs.

The middle one is the one to watch. This product’s first layer is a ninety-second run, well inside what a model already finishes unsupervised. The second, a thousand agent-to-agent simulations arguing a matter end to end, is a far longer task that becomes ordinary in the next few doublings. One layer on capability that exists, one on capability arriving to a schedule other people are measuring.

None of this is new. It has been forecast for three quarters of a century, by the people who built the machine and by the legal profession’s own adviser on it, and the useful thing about their forecasts is that they carry dates and can therefore be marked.

None of that replaces lawyers. It moves the constraint. When analysis stops being rationed the binding constraint is judgment: which argument to run, which expert to retain, which matter to take. Those decisions get better when the analysis under them is complete rather than affordable, and the advantage sits with the litigator rather than whoever bills the most hours for the reading.

Most matters buy no analysis at all.

The obvious objection is that a market cannot survive having its price cut by three orders of magnitude. The name for the counter-argument is the Jevons paradox: a falling unit cost expands what is worth doing with a thing faster than the saving shrinks the bill. The nearest clean measurement agrees. The IRS counted tax software’s share tripling to a quarter of returns by 2003 while paid-preparer use rose to 62 percent. Both grew. What shrank was the number of people getting no help at all.

It happened above the atmosphere too. A kilogram to orbit cost about $54,500 on the Space Shuttle and about $2,720 on a Falcon 9. The launch market did not shrink by a factor of twenty. It grew, because satellite constellations that were impossible at the old price became businesses: Starlink alone now approaches $12 billion a year. The demand that appeared was demand that could not have existed before.

That is not a law. Total spend falls where demand is capped outside the price, which is why cheap booking shrank travel agency rather than growing it. So the question is precise. The number of matters is capped. The number of analyses per matter is not, and today it is usually zero. At $25,000 a read most matters get none, and the analysis that would have changed the strategy is never bought. At $50 it stops being a decision: it runs at intake before the matter is taken, against every archetype the other side might retain rather than the one you guessed, and again when the served report moves the facts. None of that comes out of an existing budget, because none of it happens today.

Market sizes here are usually asserted rather than built. Ours has its derivation attached, and the figures that did not survive checking left out.

The first is the market Supreme Mind was built for; the second is larger and is where the same method goes next. What matters most is in neither: the reads nobody buys at $25,000, which no survey counts because they never happen.

The longer answer to where this ends is a statement of intent rather than a description of a product, so it lives on its own page: the fourth institution.

Who is building it.

Richard John Van Der Aa, founder. Supreme Mind is a solo-founded company in New York. Today I am the whole team. The first hires the company makes will be engineering and a New York litigation-side commercial lead.

I grew up in Veldhoven, farmland two generations ago and now where the machines every advanced chip is printed with are built. Three of the seven revolutions above fall inside one working life; I watched a region compress about as much change into two.

I studied applied mathematics and later data science and artificial intelligence at Eindhoven University of Technology, and left three quarters through, on track for cum laude, because the models could already do most of what the coursework asked. Staying to be graded on it seemed a poor use of the window. Three years applying it commercially since, on top of four building alongside the degree.

How this company started.

2024, a freelance job that became the thesis. I built a language-model pipeline that researched and wrote biographies of roughly 19,500 Dutch lawyers across 5,500 firms and published them as juristi.nl, now the largest AI-searchable lawyer and law-firm database in the Netherlands. By hand that corpus is a team of researchers working for years; it took one person a few weeks. Language models have changed the economics of text-based professional work by orders of magnitude, and law is the most text-heavy industry there is.

2025, six months of asking instead of building. I spent half a year talking to lawyers about where the hours go and watched more than a hundred legal-technology demonstrations. Almost all of it automates drafting, research or review: work that was already cheap, made cheaper. Very little touches the analysis, and not for want of trying. It is the expensive part because it is hard, and PhD-class reasoning is a capability models only reached in 2025.

Early 2026, the United States and the expert-witness market. I moved the conversations to the largest legal market in the world, and the same answer kept coming back from partners. Expert work is the line item nobody can compress. SEAK’s 2024 fee survey of about 1,600 experts puts the median rate for file review and case preparation, the work a first read is made of, at $450 an hour, up 12.5 percent in three years. The first strategic read runs $25,000 to $45,000 and takes weeks, and the decision to buy it is made before anyone knows whether the matter can carry it.

Why this is hard to copy.

A fair question, and it deserves a direct answer. Harvey, Legora and the other horizontal legal-AI companies build for the whole matter lifecycle and are very good at it. None of them builds the opposing expert. Why that gap stays open, and where it is fragile.

An archetype is a content artifact, not a prompt. The unsolved problem is not plausible expert-sounding prose. It is methodology fidelity under pressure: defending a specification choice, holding a position on an estimation window, conceding the right thing when a good question lands. That is weeks of engineering per archetype against real deposition and Daubert material, so a library is a multi-year catch-up rather than a quarter of work.

The library compounds and the market does not reward half of it. Testimony is concentrated: fifty archetypes reach about half of all civil expert testimony, five hundred reach ninety percent. But a firm’s matters cluster inside one practice, so what matters is depth in a vertical, not share of the market. Fifty skimmed off the top of every practice area are worth little to a securities firm; ten covering the securities bench are worth a great deal. Hence one practice at a time.

Being scored is the part that cannot be copied later. Replaying resolved matters blind and publishing the misses produces calibration data that only accrues by running, so a competitor starting in two years starts two years behind. It is a moat that only exists if you show it.

The honest weakness, since we have listed the strengths: the depositions and rulings underneath are public, and anyone willing to do the work can reach them. What is defensible is the work, done first and in the open, not the material.

How we build.

Selling analytical software into litigation means the standard is absolute: one fabricated citation ends a vendor relationship permanently. These four rules are what make the output something a partner can rely on.