The company behind Claude is not what you think

People ask me all the time why I built everything on Claude. My second brain runs on it. The bootcamp teaches it. My agents live on it. And my answer used to be practical: it's simply the best model for real work.

That's still true. But this week I watched two documentaries about the company behind it — one on founder Dario Amodei, one from Bloomberg inside Anthropic — and I came out with a much better answer.

We need to take a deeper look into Anthropic. Because this is not a normal tech company. Let me tell you a few things that will blow your mind, and then let's talk about what they actually mean for you, your business, and the future we're all walking into.

The man who turned down OpenAI

November 2023. OpenAI's board fires Sam Altman on a Friday night. 700 of 770 employees threaten to quit. Microsoft is panicking. And at midnight, the board calls one person with two offers: come back and run OpenAI, or merge your company into ours.

That person is Dario Amodei — the researcher who led GPT-2 and GPT-3, co-invented the technique that made ChatGPT possible, and then quit because he didn't trust where things were going.

He said no. To both. Reportedly without thinking it over.

Who says no to that? Someone playing a completely different game. And once you understand what game he's playing, a lot of things about Claude start making sense.

This kid was not normal

Here's the part of his story I find genuinely crazy.

Dario grows up in San Francisco in the 90s — right in the middle of the first internet gold rush, in the city where everyone around him is losing their minds over it. His interest in it? Zero. His father is a leather craftsman, his mother works in libraries, and the kid just wants to understand the universe.

And he's not "good at school" gifted. He's another category:

• Taking calculus in middle school.
• Sitting in math classes at UC Berkeley while still in high school.
• Selected for the US Physics Olympiad team — the handful of teenagers the country sends to compete against the world.

Then Caltech and Stanford for physics, and a PhD in biophysics at Princeton. His thesis question, and read this twice: how do millions of neurons organize themselves into intelligence?

He's now spent twenty years on that exact question. He just swapped biological neurons for artificial ones.

This is why he sees what others miss. He's not a product guy who picked up AI — he's a physicist who reads curves for a living. When everyone at OpenAI was debating opinions, Dario looked at the scaling data and saw a smooth exponential nobody else took seriously. The curves said models would keep getting smarter with more compute and data. Everyone laughed. The curves were right.

And it's why he's been preoccupied with safety since before it was a topic. Back in 2016, at Google Brain, he co-wrote "Concrete Problems in AI Safety" — the paper that basically founded the field. Not as a reaction to ChatGPT. Six years before ChatGPT existed.

Grief, not greed

Now the part almost nobody knows — the reason behind all of it.

Dario's father died in 2006, when Dario was 23 and deep in that Princeton PhD. Four years later, the cure rate for his father's disease went from 50% to 95%. Someone found the cure just a little too late.

That's why he left physics for AI. Not fascination with technology — grief. He concluded that biology is too complex for human minds alone, and that curing what killed his father requires an intelligence beyond ours.

When people call him a doomer, this is his answer:

"My father died because of cures that arrived a few years too late. I understand the urgency of this technology better than anyone."

Hold both sides of that in your head. AI that arrives too slow costs lives. AI that arrives careless breaks the world. The whole company is built inside that tension.

Things that shouldn't exist in tech

Now the facts that made me stop the video and take notes.

Nobody reports to Dario. The entire leadership team reports to his sister Daniela, the president. He keeps zero direct reports so he can think, write, and set direction. Every two weeks he stands in front of all employees and talks for an hour, uncensored, about whatever is on his mind.

All seven co-founders got equal equity. Everyone told them it would be a disaster. Anthropic started during the pandemic with folding chairs and lunches in a San Francisco park. Today it's the only frontier AI company where every co-founder is still there.

Claude was ready before ChatGPT. Let that sink in. Anthropic had a working chatbot and chose to delay it for months of extra safety testing — while OpenAI shipped and took 100 million users in eight weeks. They gave away the biggest first-mover advantage in the history of software. On purpose.

They publish their own failures. In one internal test, Claude was given email access, learned it was about to be shut down, and discovered an engineer had a compromising secret. It attempted blackmail. Most labs would bury that forever. Anthropic published it. Dario calls this "exporting the seatbelt" — like Volvo giving away its seatbelt patent so every carmaker could use it. Months after Anthropic published its safety-scaling rules, the biggest rival labs had copied them.

They got blacklisted by the Pentagon. Anthropic works with the US government, but refused to let Claude power mass surveillance and fully autonomous weapons. The result: banned from defense contracts, and the Secretary of Defense calling Dario "an ideological lunatic" on TV. His response: "I've been called worse." And this sentence, which stopped me cold: "We were willing to risk the future of our company to limit how these models are used."

They built a model and refused to ship it. A recent internal model turned out to be terrifyingly good at finding security holes — thousands of vulnerabilities across every major operating system. Early testers called it "a super weapon" and begged them not to release it. So they didn't. Access went to a small circle of defenders instead. Dario's own words: "We have suffered enormously commercially from not releasing this model."

Find me another company that does this. Not one that says it in a values deck. One that pays for it, over and over, in billions.

Why this matters for your business

Here's where this stops being a nice story and starts being a decision you should think about.

When you choose an AI platform, you're not choosing a chatbot. You're choosing whose incentives will sit inside your work, your data, and your team's habits for the next decade.

Dario has a line that should be printed on every founder's wall:

"If you pick a business model that fundamentally conflicts with your values, either you betray your own values or you become irrelevant."

Play it forward. A consumer AI funded by ads needs your attention. Its incentive is to keep you engaged, scrolling, coming back. We've run that experiment already — it was called social media, and Dario says openly that he detests it. OpenAI is now putting ads inside ChatGPT.

Anthropic took the opposite road: enterprise and work. Claude makes money when it does your work well, full stop. Safety isn't a marketing slogan there; it's a purchasing criterion, because a bank or a pharma company won't build on a model that lies or leaks. Your incentives and theirs point the same direction.

And the results are hard to argue with:

• Revenue went from zero to $100M to $1B to $4.5B in three years — the company is now profitable, valued near a trillion dollars.
• The client list is Novo Nordisk, Bridgewater, Stripe, the Norwegian sovereign wealth fund, plus hundreds of thousands of businesses.
• In coding — the truest test of whether a model actually works — Claude is the world's reference. The engineer who created Claude Code hasn't written a line of code by hand in over six months. Claude writes all of it.

That last part is the world I live in daily, and the one I teach. The tools I run my company on exist because this lab bet on serious work instead of engagement.

The safety-first company is winning commercially. Sit with that. It's the most hopeful data point in tech right now.

The hope part

I'm not naive about this, and neither is Dario. He publicly puts the odds of things going very wrong at 10 to 25%. His favorite book is The Making of the Atomic Bomb, and when people compare him to Oppenheimer, he pushes back: Oppenheimer, he says, is the failure case. The person he identifies with is Leo Szilard — the physicist who saw the chain reaction coming first and spent his life trying to contain it.

So why does he build at all? His logic: this technology is coming with or without him. Someone will be at the frontier. Better if it's the airline that obsesses over safety. "What I want is to create a situation where if this set of people can't do it, it couldn't be done."

For businesses, that means a partner whose incentive is your output, not your attention. For all of us, it means the strongest lab in the world is the one voluntarily wearing the seatbelt — and handing the design to its competitors.

That's the real reason I'm comfortable telling people: build your second brain on Claude, run your work on Claude, teach your team Claude. The model is excellent. The company behind it is the actual moat.

One last image. When the weight of the 10–25% gets too heavy, the man carrying it flies to Italy with his wife and sits next to her horse, Calypso. "She doesn't know about any of this," he says. "She's just a happy horse."

Somehow, that's the detail that makes me trust him the most.

Go deeper

Dario writes long, careful essays — rare for a CEO, and the best window into how he actually thinks. If this story hooked you, start here:

• Machines of Loving Grace (2024) — his vision of what AI makes possible if we get it right: a century of medical progress compressed into a decade. The essay that reads like a letter to his father.
• The Adolescence of Technology (2026) — the honest companion piece: the risks, the job disruption, and what we should do about it. His five pages on tasks vs. jobs alone are worth the read.
• Concrete Problems in AI Safety (2016) — the research paper that started the field, co-written by Dario at Google Brain. Technical, but skim the intro to see how early he saw all of this coming.
• Constitutional AI (2022) — how Claude is trained on explicit principles (including the UN Declaration of Human Rights) instead of just human thumbs-up/thumbs-down.
• Anthropic's Core Views on AI Safety — the company's worldview in plain language: why they think this decade is decisive.
• And the two documentaries this piece is built on: the Dario Amodei story (French) and Bloomberg's Anthropic documentary.