AI prepares. Lawyers judge.
AI compresses the production of legal work. What remains is smaller, faster and concentrated in judgment, and a refusal is completed work, not a failed one.
For most of the history of legal services, the lawyer performed the whole workflow: gathering the facts, researching, drafting, formatting, communicating and reviewing. The price of legal work was the price of all of that, bundled, by the hour.
AI pulls that bundle apart. It can compress most of the mechanical work. What remains is a smaller unit, faster to deliver, and concentrated where it always mattered: professional judgment.
Production and judgment are different units
The old unit is hours of bundled production and judgment: a lawyer prepares the agreement from scratch, then decides whether it is right. The new unit is minutes of focused judgment: software prepares the work, and a licensed professional reviews the decisions that are actually theirs to make, with software handling the coordination around them.
That changes who can afford to ask. As legal artifacts get cheaper to make, far more people reach the moment where they ask the one question software can't answer for them: is this actually okay? Someone who would never pay for bespoke drafting may well pay for a qualified person to check the part that matters.
"Add a lawyer" can become a button
Picture the end of any flow that produces legal work: an operating agreement your incorporation product just generated, a contract your procurement agent negotiated, a filing your tax software prepared. Next to "download" sits one more choice: have a licensed professional review it before you sign.
That interaction can live anywhere legal work is created: AI assistants, incorporation products, contract tools, tax software, real-estate systems, HR platforms, procurement agents, and products nobody has built yet. Every product capable of producing legal work can become a way to reach legal judgment.
The four answers a reviewer can give
A review is not a yes. Asked to decide whether an agreement can safely be executed, a professional can:
- Approve it.
- Approve it with changes.
- Not approve it.
- Ask for more information.
Every one of those is professional judgment, and every one is completed work. A product that can only handle the first answer isn't buying judgment. It is buying a stamp.
api.lawyer's contract encodes this. When a Signer finishes a Gig, the result is one of three outcomes, and the API refuses anything else:
{ "error": "outcome must be 'attested', 'send_back', or 'declined'"}
Never pay for a predetermined rubber stamp
This is the line we hold hardest. If a network only pays when the professional says yes, it is paying for approval, and the signature stops meaning anything. So in api.lawyer a Declination is:
- Paid. The declining Signer's review is completed, paid work. The flat Legal Fee was fixed when the Gig was posted, so saying no costs the Signer nothing, and the signature never sits under economic pressure.
- Final. No process, meter or operator can move the Matter past it. Only the Client can, by giving a direction.
- Neutral. A refusal is never scored against the professional who made it.
You can watch "final" hold in the sandbox. Decline the Gig, then try to advance the Matter:
{ "error": "Matter m_5 carries a Declination — the Signer's no is final against every process, meter, and operator; journal a Client Direction before any re-advance"}
Why this is the trust architecture
For a founder, the takeaway is practical: your product buys qualified judgment, not approval, so your interface has to handle "no", "not yet" and "change this" as first-class outcomes. That is what the typed states are for.
For a bar-journal editor, it is the point: the independence of the professional's judgment isn't a promise in our terms of service. It is structural. The economics don't reward a yes, the record can't route around a no, and the professional's determination stays theirs.
Play both branches, attest and decline, in the sandbox walk.