Public transcript

A Conversation with Dario Amodei and Marc Benioff | Dreamforce 2025

By: Salesforce
Length: 38 min
Language: English
Text from: the video's captions
Transcribed: 12 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

At Dreamforce 2025, Marc Benioff interviews Anthropic CEO Dario Amodei about his background, co-founding Anthropic with his sister Daniela, and the company's enterprise focus. Amodei discusses AI agents, Claude writing 90% of code on many Anthropic teams, the role of models in healthcare, and his vision for Anthropic's future. He also touches on data center buildouts, competition with OpenAI and Google, and Anthropic's revenue approaching a $7 billion run rate.

Key points

Dario Amodei grew up in San Francisco, studied neuroscience and computational biology, and worked at Google and OpenAI before co-founding Anthropic with seven co-founders including his sister Daniela.
Amodei says Anthropic has at least a plurality, perhaps the majority, of enterprise customers and enterprise revenue among AI model providers.
Anthropic's revenue is approaching a $7 billion run rate and has grown roughly 10x per year for the last three years.
Amodei predicted six months ago that 90% of code would be written by AI models, and says that within Anthropic and partner companies this is true on many teams, with humans in editing and supervisory roles.
He argues by comparative advantage this means companies may need as many or more engineers, not fewer, calling it a rebalancing rather than a replacement.
He worries about labor disruption across the economy in the next two to five years.
Amodei shares a story of Daniela using Claude during her pregnancy, where Claude listed diagnostic possibilities doctors had missed, illustrating models as a 'second opinion' rather than a sole authority.
Claude's benchmark performance on writing GitHub PRs from a description has gone from about 5% to roughly 77% over the last year and a half.
Anthropic is securing gigawatts of compute, and Amodei says some data center deals in the press seem 'fishy' and possibly involve double counting.
His vision for Anthropic is to put AI at the center of every enterprise and be known as a trustworthy, responsible company, with healthcare and curing diseases as a top upside.

Questions it answers

00:00Introduction and background

Who is Dario Amodei and how did he come to AI?

A San Francisco native and former neuroscience researcher, Amodei moved into AI after seeing early deep learning results, worked at Google and OpenAI, and helped document the scaling laws before co-founding Anthropic.

Studied neuroscience and computational biology at Stanford.
Left OpenAI with seven co-founders around 2020 over scaling and safety convictions.

04:58Founding with his sister

How do Dario and Daniela Amodei run Anthropic together?

Daniela handles day-to-day operations and structure while Dario focuses on strategy, research direction, safety, and vision — a split he says makes both more effective.

Daniela joined Dario at OpenAI in 2018.
Anthropic is 4.5 years old and headquartered in the former Slack building.

09:30Surprises and competition

What surprised Amodei most since leaving OpenAI?

Although scaling laws predicted the growth, seeing it play out was wild, and the biggest surprise is how much the two companies' paths diverged — OpenAI going consumer, Anthropic going enterprise. He describes Google as both partner and competitor, a normal 'co-opetition' dynamic.

Anthropic claims a plurality or majority of enterprise AI revenue.
He was happy Demis Hassabis won the Nobel Prize and dedicated his essay to AlphaFold.

13:00Agents and AI-written code

Is Claude really writing 90% of Anthropic's code?

On many teams, yes — humans play editing and supervisory roles, and Claude even found an obscure cluster bug engineers had missed. By comparative advantage this means as many or more engineers, not fewer.

Amodei calls it a rebalancing, not a replacement.
He expects labor disruption across the economy in two to five years.
GitHub PR benchmark performance went from ~5% to ~77% in 1.5 years.

22:00Healthcare and radiology

How reliable are AI models in healthcare today?

Models work best as a second opinion. Daniela used Claude during pregnancy and it listed diagnostic possibilities doctors missed, but blindly following its advice would not have gone well. Radiology struggles reflect data gaps and a human–model collaboration model still being worked out slowly in high-stakes industries.

Code adoption cycles take months; healthcare takes much longer.
Specialized image tasks may need specialized models and data.

27:00The coding ecosystem

What has surprised Amodei about tools built on Claude?

Products like Lovable that let non-coders write code end to end surprised him most, though he declined to pick a favorite product. He says it's too early to name a winner among the many approaches.

Ecosystem spans IDEs, consumer tools, command-line tools like Claude Code, and more.
Different tools may suit different people.

30:05Five-year vision

Where does Amodei want Anthropic to be by 2030?

Drawing on his essay 'Machines of Loving Grace,' he wants AI to help cure diseases like cancer and Alzheimer's and to put AI at the center of every enterprise, while being known as a trustworthy, responsible company.

Revenue is approaching a $7 billion run rate, up 10x yearly for three years.
He sees enterprises as a 'super intelligence' with a very high skill ceiling.

34:59Data centers and compute

How does Anthropic think about data center buildouts?

Scaling laws imply a need for many gigawatts of power, and Anthropic is securing compute wherever it can. Amodei worries some press-covered data center deals seem fishy and possibly double-counted, and stresses that making money matters more than spending it.

Amazon has been an especially important partner for growth.
If customer demand is there, compute will follow.

Notes

Background and Founding Anthropic

Dario Amodei grew up in San Francisco, attended Lowell High School, and did graduate work in neuroscience plus a postdoc in computational biology at Stanford.
He found biology's complexity beyond human scale, then saw early deep neural network work from Google and the University of Toronto and moved into AI, working at Google and OpenAI.
He and the eventual Anthropic co-founders were among the first to document the exponential scaling of AI technology.
Around 2019–2020 he concluded AI would reshape the economy and that, because models are grown rather than explicitly coded, safety and responsibility were essential.

Working with His Sister

Daniela Amodei, four years younger, is one of seven co-founders; she joined Dario at OpenAI in 2018.
Daniela runs day-to-day operations while Dario focuses on strategy, research direction, safety, and vision; he says he couldn't run the company without her.
Anthropic has existed for 4.5 years and occupies the former Slack building in San Francisco; Salesforce was a very early investor (around a $4 billion valuation).

AI Agents and Code

The biggest current shift is agentic abilities: models doing tasks end to end, seen first in code because developers are early adopters.
Amodei predicted 90% of code would be AI-written within six months; on many Anthropic teams this is now true, with humans editing and supervising.
By comparative advantage, this is a rebalancing, not a replacement — companies may need as many or more engineers, who become 10x more productive.
He worries about faster, broader labor disruption across the economy in the next two to five years.
Claude's benchmark on writing GitHub PRs from a description rose from about 5% to roughly 77% in a year and a half.

Healthcare

Models work best as a 'second opinion'; Daniela used Claude during pregnancy when doctors missed a bacterial infection diagnosis.
Radiology struggles reflect missing specialized data and unresolved human–model collaboration models, which work out slowly in high-stakes regulated industries.

Vision and Compute

His essay 'Machines of Loving Grace' outlines AI's upside, especially curing diseases like cancer and Alzheimer's.
Anthropic's revenue is approaching a $7 billion run rate, up 10x per year for three years.
He sees enormous compute needs, is securing power wherever possible, and says some data center deals seem 'fishy' with possible double counting; he stresses making money matters more than spending it.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.