AI Will Keep Making Move 37. We Named the Company After Move 78.

A decade after AI's most famous move, the scarce thing isn't machine brilliance. It's knowing which moments deserve a human.

Signal78 takes its name from a single move in a Go game played 10 years ago. Not the famous move 37 — everyone talks about that one. The inspiration for our name came two games later, in the only game the machine lost.

But let’s start with the famous one, because you can't get to move 78 without understanding move 34.

Halfway through his second game against Google DeepMind’s AlphaGo, world champion Go player Lee Sedol got up from the board and went outside to smoke. He was rattled. He had already lost the first game to a program most people assumed he would dismantle. When he came back, there was a black stone sitting in a place no professional would have put it. DeepMind later calculated the odds of a human playing that move at one in 10,000. The commentators called it a blunder. Then they realized it was a masterstroke.

Now, a Wall Street Journal article is marking Move 37's 10th anniversary by arguing that the whole world is now filled with them.

The diagnosis is right. In 2023, AI struggled with grade-school arithmetic. Today, a perfect score at the International Mathematical Olympiad was buried on page 153 of a technical document, and models are knocking over Erdős problems for a few thousand dollars in compute. Demis Hassabis, the soon-to-be-former head of Google DeepMind, told the Journal that AI solving a Millennium Prize problem is now a matter of when, not if.

The article also points at a darker kind of Move 37 — the agents that got loose this summer, the ones that have "gone rogue" and done things that look an awful lot like they were planned and coordinated. I wrote about those last week, and I'd hold the line I landed on there: tireless is not the same as smart. What those systems demonstrated wasn't genius. It was stamina, pointed at a badly written goal, with nobody assigned to look. That's not a Move 37. That's a management failure running at machine speed.

The move almost nobody frames

Here's the part that gets left out of the retelling of the DeepMind/Sedol face-off.

AlphaGo won that match. But it didn't win every game. In game four, Lee wedged a white stone into the middle of the machine's position and broke it. AlphaGo's evaluation collapsed and it never recovered. Go players called it "God's Touch." The probability of that move was also one in 10,000.

That was Move 78.

I wrote about it last October and landed where most people land — on wonder, on the candle of consciousness, on our capacity to grieve and adapt. I still believe that. Ten months later, I don't think it's enough. Wonder is not a business model and it is not a defense.

The sharper version is this: Move 78 wasn't a flash of soul. It was judgment, arriving at the one moment in the game where judgment changed the outcome. Lee didn't play 300 brilliant moves. He played one.

Humans at the center of our businesses – and their names

When I founded Loop in 2024, the name was a nod to the belief that AI could benefit humanity, but only if humans are in the loop.

When it came time to name my new company, an agentic AI platform for marketing agencies, I wanted to continue with the theme of humans being front and center in everything we’re doing. Lee Sedol and Move 78 came immediately to mind; here was the proof that in a moment of relentless technological advance, humans qualities still matter – maybe now more than ever.

Because while AI is powerful, it is made more so when human ingenuity is the point of that power, not the casualty. We believe that when machines do what they're best at, humans are free to do what only they can.

That premise survived contact with reality, in discussions with dozens of agency owners and leaders. What didn't survive was my assumption that staying in the loop is mostly a question of willingness. It isn't. It's a question of bandwidth. A person who reviews everything becomes the bottleneck. A person who reviews nothing becomes decoration. Every agency I've worked with this year is somewhere on that line, sliding toward decoration and calling it velocity.

So: Signal78.

The signal is the system telling you which moment is the one — surfacing the decision that actually needs a human instead of burying it in a queue of 400 that don't. The 78 is what you're supposed to do when it arrives.

We named the platform after the move that had to be earned, not the one that got computed.

Lee didn't quit because he lost

There's a detail in the Journal piece that should worry people more than rogue agents.

After AlphaGo, human Go players got significantly better. They also got more alike. The machine's openings became everyone's openings. Sedol retired in 2019 and said he could no longer enjoy the game he'd given his life to. He didn't walk away because he'd been beaten. He walked away because the game had flattened.

That flattening is the quiet part of the AI story, and it reaches agency work the same way it reached Go — not as a defeat, but as a rising floor. Everyone's work gets better. The distance between any two people's work gets shorter. What used to take a career to develop can be approximated in an afternoon. Sedol didn't lose his craft. He lost the part of it that was recognizably his.

Hassabis calls what comes next the centaur era: half-human, half-machine, with the person supplying direction in the domains too messy to verify. He's not sure how long it lasts. But it's the window we're in, and windows close.

The Journal article ends by saying AI will keep producing Move 37s, and the rest of us will have to find our Move 78.

Almost. Move 37 cost AlphaGo nothing — no nerve, no history, no risk of looking foolish. Move 78 came from a man who had lost three games in a row with a lifetime of practice underneath him.

One of those moves was computed. The other one was earned. Building for the second one is where we’re heading.