Wink Pings

Anthropic Co-Founders in Their Own Words: A Cheap Domain, “Dragon Mode,” and the Origin of Claude Code

Seven co-founders sit down to tell early Anthropic stories. The domain was a bargain; early AI would scream in all caps; when Claude launched, they admitted they were behind. A user on X broke the key milestones into four steps.

Anthropic co-founder video cover

Anthropic is the company behind Claude. Its seven co-founders (Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, Chris Olah) appear together to tell the story of the company’s early days. The video is shot like an eighties sci-fi film. Some online commenters said it looks like the first episode of a documentary about human extinction. Others noted that seven people told one story and none of them mentioned a pivot.

The video is here.

The video is not short, but some online commenters broke out the key milestones, which are more specific than the official narrative.

First, they divided up the work. Tom Brown went to work on the AWS cluster, Dario Amodei focused on fundraising, another co-founder handled hiring, and Chris Olah and one other person built the earliest company website. On January 26, 2021, the company was incorporated as a public benefit corporation.

The naming process was very casual. The candidate list included squirrel, owl, whale, differentiated, humane. Jared Kaplan proposed anthropic, meaning “human-centered.” They settled on it in the end because the domain was the cheapest. A comment left in the spreadsheet said: “We like this name, it’s good. Is the domain available? Yes.”

Early models were far more of a mess. One co-founder said that the 2021 assistant “was completely dishonest” and would enter “dragon mode,” screaming in all caps. It wasn’t until mid-2022 that they made a model that was actually usable. In March 2023, Claude launched, and at the time it was “far behind the frontier.”

Being behind turned out to be a good thing. One co-founder said it made them “extremely focused,” obsessively scrutinizing compute resources and small algorithmic improvements, and then building things up bit by bit.

After Sonnet 3.5, the model suddenly became good at writing code. They built a tool to go with it, which later became Claude Code. It is an AI coding tool.

Looking back, Anthropic’s story doesn’t look like something that was planned out. The name was a bargain, the early models were chaotic, and the launch was behind. But some online commenters said that the only entity at Anthropic people trust is Claude, and the humans behind it are another matter. That sentence probably also explains why Anthropic is willing to tell this history.

发布时间: 2026-11-21 13:06