What Really Won the Trillion-Dollar Supreme Court Case | Neal Kumar Katyal

ELI5/TLDR

A top Supreme Court lawyer was hired to kill a president’s four-trillion-dollar tariff scheme, something no lawyer had managed in 237 years. To prepare, he leaned on four “teachers”: a mindset coach, an improv coach, a meditation coach, and a custom AI he’d trained on 25 years of every justice’s words. The AI predicted the justices’ questions almost word for word, even the chief justice’s likely escape route to a yes. But his real point is that the AI only got him to the door, and the thing that actually won the case was the one thing a machine can’t do: connect with another person in real time.

The Full Story

The setup

Neal Katyal argues cases at the Supreme Court, standing ten feet from nine justices who throw roughly fifty questions at him in thirty minutes, no script allowed. Five months before this talk he took on a case to strike down the president’s signature initiative: tariffs imposed on nearly every country on earth in April 2025, using a 1977 law, with no vote in Congress.

If the president can command the global economy by yelling emergency, what can’t he do? Checks and balances don’t just bend, they break.

Everyone told him it was hopeless. Six of the nine justices were Republican-appointed, three by the president himself. And no Supreme Court in 237 years had ever struck down a president’s signature program. He took it anyway. Then, three weeks out, a teammate tried to oust him from the case and got a newspaper editorial calling Katyal’s selection a “strategic mistake.” He wasn’t replaced. He argued it. He won, six to three.

Four teachers

The body of the talk is structured as a slow reveal: four coaches, each setting up a “was this the secret?” question, each answered “no.”

Ben, the mindset coach. Ben works with athletes on “game day” nerves. His first question was simply “What are you afraid of?” Katyal admitted something he’d buried: walking past the portraits of past justices, all of whom looked nothing like him, the son of immigrants, he always felt he didn’t belong. Imposter syndrome, in other words, doesn’t keep score of your wins, only your doubts. The night before, terrified, Katyal listed everything he had to do. Ben told him to swap one vowel: change “got to” into “get to.”

“I get to defend the Constitution of the United States. I get to, the son of immigrants, remind the country of what it’s about.”

One letter turned terror into something closer to joy.

Harvey, the relentless reader. Katyal describes Harvey as a tireless researcher who reads the 200th tariff case as carefully as the first, spotting patterns and predicting “angles of attack.” A month out, Harvey told him Justice Barrett would ask about license fees, almost the exact words she later used.

Liz, the improv coach. Her one instruction: actually listen. She taught him the core of improv, “yes, and”, which means absorbing what the other person says and building on it rather than waiting to talk. In the courtroom this turned interrogation into dialogue. When a justice attacked, he validated the worry, then bridged back to his point. (The talk plays real argument audio here, where Katyal name-checks each justice’s own stated concerns and answers them on their terms.)

Bob, the meditation coach. The self-described last person on earth to meditate (“I do not own crystals”) spent twenty minutes a day focused on a single word. Bob even rented an apartment a block from the court so they could work daily. The result: when Katyal walked in, “the static was cleared.”

The twist

Three of the four teachers are human. Harvey is not. Harvey is a custom AI built with a legal-AI company, trained on every question every justice asked in 25 years plus every opinion, concurrence, and dissent they wrote. From that, patterns emerged. It predicted Gorsuch would press the taxing power, Kavanaugh would push tariffs-versus-embargoes, Barrett would worry about refunds. Most strikingly, it predicted not just the chief justice’s question but a possible escape route, a way he could rule against the tariffs while still protecting the institution he’d spent his career defending. Katyal says he held that narrow door open and the chief justice walked through it.

He’s careful to defend why this isn’t a cheap trick. Predictability in a judge isn’t a weakness to exploit; it’s integrity. A justice who returns to the same principles year after year is showing character. The AI was reading consistency, not finding a crack.

The shadow, and the real point

Then the warning. AI has a “shadow side”: when a tool is powerful, people stop thinking. “The computer says so” — four words and human judgment switches off, people fold “like a cheap lawn chair.” If Katyal had simply parroted Harvey’s output, he says, he’d have lost. Harvey was a sparring partner, “brilliant, tireless, occasionally insufferable,” never a god. It asked the questions; the humans found the answers.

The bigger claim: for centuries, expertise was accumulated knowledge — the doctor or lawyer who’d read and seen the most. AI makes that particular edge nearly worthless, because pattern-recognition across vast data is now available to anyone. What’s left, the one thing AI can’t do, is connect: persuade one specific person by reading what’s beneath the surface and adjusting the tone, the pause, the look. His proof is a moment Harvey didn’t predict — a Barrett question he answered by simply looking at her and trying to understand her worry. The closing question he leaves the audience: not “will you be replaced?” but “what is the irreducibly human thing that you do?” Go deeper into that.

Key Takeaways

A custom AI trained on 25 years of every Supreme Court justice’s questions and written opinions predicted the justices’ actual questions in a major case, sometimes near-verbatim, including a likely “escape route” reasoning path for the chief justice.
The “get to” vs “got to” reframe: swapping obligation language for privilege language reframes performance anxiety into purpose. A one-word mental edit, not a personality change.
Imposter syndrome ignores your track record and feeds only on your doubts — winning more does not dissolve it.
The improv principle “yes, and” applied to high-stakes argument: validate the questioner’s concern first, then bridge to your point. Turns interrogation into dialogue.
A judge’s predictability is best read as integrity (consistent principles), not as a weakness to be gamed.
The shadow side of powerful tools is deference: “the computer says so” shuts off human judgment. Treating AI as a sparring partner rather than an oracle is the guard against this.
AI is commoditizing the historical expert advantage — breadth of accumulated knowledge and pattern recognition. The remaining human edge is real-time connection: reading one person beneath the surface and adjusting delivery to them.

Claude’s Take

This is a polished, well-built TED talk and the central structural trick — four “teachers,” three human, one secretly an AI — is genuinely effective. The “get to / got to” reframe and the “yes, and” point are small but real, the kind of thing you can actually use.

Two things to flag. First, the case itself: as of this talk’s framing, Katyal presents a clean six-to-three win striking down the tariffs as settled history. Treat the legal specifics as his account, told by the winning advocate, not a neutral record. Second, and more important, Harvey is a product. Katyal built it “with a legal AI company,” and the real-world company named Harvey is a well-funded legal-AI startup. So this is partly a very good product demo wearing a humanist message. The humanist message (“AI predicts, humans connect”) is comforting and probably half-true, but notice it’s being delivered by someone whose whole pitch is that the AI nailed the predictions. The “connection won it” claim rests on a single anecdote (the one Barrett question Harvey missed) and is essentially unfalsifiable — we can’t rerun the case without the AI.

Scored a 7: strong craft, a couple of portable ideas, honest enough to name AI’s “fold like a lawn chair” failure mode. Docked for being thinner than it looks once you separate the genuine insight from the implicit advertisement.