Show Me Where to Put the Fulcrum

Show Me Where to Put the Fulcrum

Elon Musk paid $60 billion for Cursor this week, and he paid in SpaceX stock — days after the biggest IPO in history, the shares already up better than 30%. Sit with the mechanics, because they're the tell. When you pay a fixed price in a currency that keeps climbing, the deal gets cheaper the higher it runs: fewer shares out the door, less dilution, same trophy. It's the oldest move in his book. Spend a currency you've inflated with a story. Tesla did it for years. SpaceX is doing it now.

But the price isn't the story. What did one of the most valuable companies on earth actually buy? Not a model. Cursor is the biggest coding shop that isn't a frontier lab, which means it sees more cutting-edge code, from more real developers, than anyone alive. He didn't buy a tool. He bought a sensor.

That's the Musk pattern, and it's the whole game. Every Tesla mile trains the self-driving. Every launch is telemetry. Every tweet, for him or against him, feeds the machine. Cars, rockets, the timeline — every product is a sensor. (Even, per a Pentagon filing this week, the battlefield; that one's in the Daily 5, handle with care.)

And the Cursor piece has a wrinkle most people miss: Musk doesn't need the contents of your code. The gold is the process — write, debug, deploy, recompile — the largest record of how software actually gets built that exists anywhere. Because Cursor runs on every frontier model plus its own, he now watches how OpenAI, Anthropic, and Google really perform in the wild. He didn't just buy a sensor. He bought an observatory pointed at every rival. Cursor is how Grok gets to recursively improve.

Now the part that matters for you. Archimedes said give me a lever long enough and a place to stand, and I'll move the world. Everybody quotes the lever. The lever is the model, raw capability, and the lever just became free. A year ago we guarded that last 10% of model quality like the crown jewels. Today a Chinese open model ships the same horsepower from Shenzhen at a tenth the price. The lever is a commodity. The place to stand is not.

Anthropic published the proof the same day. 400,000 coding sessions: the human plans, the model executes, and the more expertise you bring, the more it does per instruction. But the number that stops you looks backwards — the gap between expert and intermediate is modest. Here's why. In most fields, nobody has found the fulcrum yet. Everyone's standing it in roughly the same default spot, so everyone gets the same decent result. Coding is the exception, because coding has the tightest feedback loop there is: the test passes or it doesn't, in seconds. So the right placement gets found fast, and whoever finds it pulls away. That's the 1000X coder everyone keeps talking about. There's no 1000X lawyer yet, because law has no unit test.

The lever is free. Knowing where to place the fulcrum is what matters.

At COAI today: the full Signal/Noise — the Grok flywheel, the observatory Cursor hands Musk over every rival model, and the eight questions for finding your own fulcrum — is live at getcoai.com.

Outsider Labs Advisory →

$60 billion — what SpaceX paid for Cursor this week, in stock already up better than 30% since its IPO, which means the deal got cheaper the higher the currency ran. And what did one of the most valuable companies on earth actually buy? Not a model. The index of how the world writes code — the one asset a free, commoditized model can't copy. Musk didn't buy a coding tool. He bought the fulcrum.

Name the one thing you own that a model can't get anywhere else. Cursor was worth $60 billion for a single asset: the index of how the world writes code. Anthropic's value sits in the human's judgment. Yours is a dataset, a process, or a relationship that's only yours. Today: write it on one line. If the line is blank, that's your project.

Find the work in your shop with the fastest feedback loop. That's where AI pays off first — coding got its 1000X because a test passes or fails in seconds; the slow, fuzzy tasks haven't been cracked yet. Today: rank your team's work by how fast you can verify the output, and aim AI at the top of that list before a competitor finds the same fulcrum.

Check whether an agent can actually find you. Bots outnumber humans online now, and search traffic is leaking away by the quarter. Rented reach is evaporating. Today: confirm your product, API, and content are structured so an AI agent can retrieve and cite them — then start building one direct channel to an audience you own outright.

Tuesday, June 16

SpaceX buys Cursor for $60B — and an observatory on every rival. (Full analysis above.) Because Cursor runs on every frontier model, Musk now sees how OpenAI, Anthropic, and Google actually perform in production. He doesn't need your IP — the process data and the cross-model view are the prize. Open question: whether enterprises flinch at "our code routes through Elon."

Anthropic published the receipt on expertise. (Full analysis above.) 400,000 Claude Code sessions: task value up ~25%, debugging down by half. The expert-vs-intermediate gap is modest — because coding's fast feedback loop is the only place the fulcrum's been found yet. No 1000X lawyer until law gets a unit test.

A Pentagon filing says Grok helped fire 2,000 munitions at Iran. One sworn statement — filed to defend xAI in a pollution suit — claims Grok hit "2,000 targets in 96 hours." Single-sourced and unverified, and the strikes reportedly killed hundreds of civilians; no trivializing that. But if conflict is a fact of life, the side whose models learn from every engagement compounds the edge. The flywheel reaches the battlefield too.

OpenAI's losses reportedly exploded nearly 8X — as it files to go public. Leaked docs (Ars Technica) land the same week OpenAI joins SpaceX and Anthropic in the IPO chute. Epoch says the self-funded buildout runs out of cash by year-end. The vendors you rent are about to answer to Wall Street — and Wall Street wants the opposite of cheap tokens.

Robots are filling the shifts nobody else will. US manufacturers face 2.1 million unfilled jobs by 2030; 72% of employers worldwide already can't find talent. Genesis ships humanoids to LG's factories in Q4, and BMW's Figure robots have run 11 months of shifts. This wave is demographics, not cost-cutting — a different buyer with a different urgency.

— Harry and Anthony

Sources: