The Killer App Is a Lead List

Zuckerberg wants 1.3 million GPUs by end of year and is spending $65 billion to get them. That number — $65 billion — is not a budget, it's a statement. It says: we are not slowing down, we are not hedging, we are not waiting to see how this plays out.

Everyone is doing this. OpenAI, Google, Amazon, Microsoft, Meta — all of them, simultaneously, all in. The natural question is whether they all win, and the honest answer is probably that some of them do, in the way that railroads won — a few survivors, a graveyard, and the world permanently reorganized around whatever they built. The ones who lose won't lose because they were timid. They'll lose having spent fortunes, at full speed, believing completely.

Meanwhile, at the application layer, browser agents are having a moment. The idea is exactly what it sounds like: an AI that operates a web browser the way a person does, clicking and scrolling and typing and waiting for page loads. Browserbase — who built Stagehand, which is genuinely well-designed, one of the few tools in this space that feels thought through — just dropped Operator. Hyperbrowser just launched. There are more coming. You can feel the category forming in real time, the way you could feel crypto wallets or Kubernetes dashboards forming, that specific energy of ten companies solving the same problem and none of them knowing yet which one people will actually use.

What are people building with these agents?

The demo making the rounds today is a browser agent constructing a B2B lead list.

Sixty-five billion dollars. One-point-three million GPUs. Several simultaneous bets, by some of the wealthiest institutions in human history, that they are building something close to artificial general intelligence. The use case that circulated — the thing that got people excited enough to share — was a robot filling a spreadsheet with company names and contact emails.

Lead generation. The task that used to fall to the newest person on the sales team, the one still figuring out where the coffee is, the one who will definitely leave in eight months.

I'm not saying this is wrong. Lead lists are real work, someone has to do them, automating them is fine, sales teams will actually use this. But there is something almost meditative about the distance between the scale of the infrastructure and the specificity of what emerges — all that compute, all that capital, all those cooling systems pulling power from grids that are now visibly struggling, and here is what surfaces first at the edges: a very fast, very expensive assistant who can do one thing pretty well.

Maybe that's how it always goes. The first cars just replaced horses on the same roads.