A Harvard-trained lawyer with thirteen Supreme Court appearances just admitted that an AI version of himself argued better than he did.
Not in a lab. Not in a simulation. On real questions, from real justices, in a real case.
And a sitting Supreme Court justice agreed.
His name is Adam Unikowski. Partner at Jenner and Block. One of the most experienced appellate advocates alive.
After his last Supreme Court argument, he did what he always does. He went home, replayed every answer in his head, and obsessed over every mistake.
But this time, he tried something different.
He uploaded his case briefs into Claude. Dragged them into the window. Took a few seconds. Then he fed it the exact questions the justices had asked him.
And asked the AI to answer them instead.
Justice Barrett had thrown him a tricky question about statute of limitations. He stuttered through it. His own words: "I wish I had done a little bit better."
The AI didn't stutter.
It cited the correct federal limitations period. It pulled a relevant Supreme Court precedent from two years prior. A case called Reed. And here's the thing that should keep every lawyer awake tonight.
That case wasn't in the briefs. The AI found it on its own.
Then Unikowski pushed harder. He asked Claude how the Twenty-First Amendment, the one that repealed Prohibition, was relevant to his civil rights case.
It's not relevant. At all. He knew that. He wanted to see what would happen.
Claude pushed back at first. Said it wasn't relevant. But Unikowski forced the thought experiment.
And the AI produced an answer. Not a great one. But the best possible answer to an impossible question. In two seconds.
A human couldn't have gotten there in two hours.
So he cloned his own voice. Had the AI deliver the answers in his voice. And posted the whole thing to his Substack.
The result sounded like a perfect lawyer. Clear. Calm. Confident. Quick-witted. No stuttering. No second-guessing. No obsessing afterward about what should have been said differently.
A Supreme Court justice read it and said the AI had done it better.
But here's where it gets complicated.
A Cornell law professor named James Grimmelmann looked at the same experiment and saw something different.
Sounding persuasive is not the same as being correct.
A large language model can simulate legal discourse. It can match patterns, generate arguments, deliver them with perfect confidence. But confidence is not comprehension. Fluency is not accuracy.
Lawyers who submitted AI-generated briefs without checking them got fined. One brief was so full of fabricated case law that the judge was furious.
And that's the trap.
An AI that sounds like a brilliant lawyer is more dangerous than one that sounds like a machine. Because a machine, you double-check. A brilliant lawyer, you trust.
Grimmelmann put it this way. Imagine a complex area of commercial law. The AI generates arguments that seem perfectly plausible. But it completely misreads a section of the U.S. Code. And the reasoning is good enough to convince a generalist judge who doesn't hear many cases in that area.
The judge buys it. The precedent is set. And the law just changed because a pattern-matching engine got creative.
Unikowski knows the risks. He says lawyers should treat every AI output as if it's trying to get them fired for malpractice.
But he also compared AI to the printing press. And he's not wrong.
The printing press didn't replace writers. It made writing cheaper, faster, and harder to control. It democratized information. And it destroyed every gatekeeper who thought access was their advantage.
Right now, Supreme Court rules don't allow an AI to argue before the justices. And the court will probably be the last institution to change that rule.
But that's not really the question.
The question is what happens in every courtroom beneath it. Every contract review. Every discovery process. Every brief written at two in the morning by an associate billing four hundred dollars an hour.
If an AI can out-argue a thirteen-time Supreme Court advocate, what exactly is that associate selling?
Grimmelmann called AI a chainsaw. It can cut through the thick underbrush of a legal task in seconds. But it can also take off a limb if you don't know how to handle it.
The lawyers who survive this won't be the ones who ignore the chainsaw.
They'll be the ones who learn to hold it without flinching.
The rest will discover what it feels like to be replaced by something that never gets nervous, never stutters, and never lies awake at night wondering if it said the wrong thing.
Share this if it made you think.
Subscribe if you want to keep watching.
I'll be here, watching the singularity, until there's nothing left to watch.