Public transcript

Live: Anthropic CEO Dario Amodei talks AI at Dreamforce summit after calls for development slowdown

By: USA TODAY
Length: 53 min
Language: English
Text from: the video's captions
Transcribed: 13 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

This is a live Dreamforce keynote in which Salesforce leaders Marc Benioff and Patrick Stokes unveil AI Force, a set of AI interfaces built on Claude, including Cloud Force, Slack Force, and Agent Force Co-worker in Lightning. Benioff frames AI Force as a way to unlock 'trapped value' in enterprise data by combining probabilistic AI with Salesforce's deterministic data and metadata layers. Anthropic CEO Dario Amodei joins briefly to discuss his recent essay on AI pacing, saying the industry should improve its own safety practices and set shared standards, and that only five to ten percent of the technology's potential value is being realized today.

Key points

Marc Benioff introduces AI Force as a new interface layer combining AI with Salesforce's data, apps and semantics, and agents layers.
Benioff says Salesforce generated $500 million in pipeline last quarter using Hunter and Piper agents, and Casey has resolved 5 million service conversations.
Benioff states Salesforce products are built with zero data retention, so customer data is not used to train models.
Patrick Stokes demos Cloud Force inside Claude Co-work, including a custom command center, a blocked security review on a $4.88 million Northstar deal, and a Salesforce plugin for Claude available in open beta.
Dario Amodei responds to his essay on AI pacing with a car-industry safety analogy: review your own practices, organize industry-wide standards, and add an international component.
Amodei says the biggest surprise of the last decade is the pace of economic adoption, and that even with technology frozen, only five to ten percent of its possible value is being used.
Agent Force now has over 30,000 customers and is doing 7 billion AWUs, with new out-of-the-box agents like Hunter, Piper, Casey, Paige, and Finn.
Benioff announces Salesforce revenue is entering the $50 billions next year and highlights 23 million trailblazers across 90 countries.

Questions it answers

02:58Dreamforce opening and community

What is Salesforce celebrating and announcing at the start of Dreamforce?

Benioff opens the biggest Dreamforce ever, celebrates 23 million trailblazers in 90 countries, and highlights community giving including $130 million raised for UCSF Children's Hospitals. He also notes Salesforce is entering the $50 billions in revenue next year and welcomes Finn as part of Agent Force.

1,600 sessions are offered at the conference.
Salesforce runs 15 of 15 cabinet agencies and 50 of 50 US states.
Usher and Gwen Stefani perform, with another $10 million going to UCSF Children's Hospitals.

15:43Trapped value and the four layers

Why does Benioff say models alone cannot run the enterprise?

Models are probabilistic and not grounded in a single source of truth, so they can produce wrong numbers. The solution is unifying probabilistic AI with Salesforce's deterministic data through four layers: data, apps and semantics, agents, and a new live interface.

Data 360 integrates, federates and harmonizes enterprise data.
Apps were always built metadata-first, enabling recomposition into AI interfaces.
Zero data retention means customer data does not train any model.

28:53AI Force product family

What is AI Force and how is it delivered?

AI Force is the new live interface layer unlocking trapped value across Salesforce. It ships as Cloud Force for Co-work users, Slack Force for Slack users, Co-worker for Lightning users, plus an SDK, and an AI Force Max edition bundles everything together.

7,000 users already run Cloud Force internally at Salesforce.
Hunter generated $500 million in pipeline last quarter.
Casey resolved 5 million service conversations in about 18 months.

34:04Cloud Force demo

What did Patrick Stokes demonstrate with Cloud Force?

Stokes showed a personal command center built by asking Claude, surfacing pipeline, service hotspots and marketing data, and flagging a blocked security review on a $4.88 million Northstar deal. He sent a Slack message through Claude to mobilize his team and showed the Salesforce for Claude plugin, available in open beta.

The plugin bundles MCP servers, zero data retention, and Claude skills.
Users can get running on these interfaces in six to eight minutes.
The demo included a sad trombone and air horn coded into the account record.

43:23Dario Amodei on pacing

How does Amodei respond to calls for an AI development slowdown?

Using a car-safety analogy, he says the responsible response is to examine your own record, improve practices, recommit to transparency, invest in safety, organize industry-wide standards, and add an international component. He notes this is the three-step plan from his essay, with the first step already committed.

He says pacing the frontier does not mean freezing the technology.
His biggest surprise is how fast AI companies grew and became central to everything.
Even frozen in place, only five to ten percent of possible value is being used.

50:32Agent Force and out-of-the-box agents

What is new with Agent Force?

Agent Force has over 30,000 customers and does 7 billion AWUs, with agent script and voice now GA. Because building agents with probabilistic systems is tricky, Salesforce built out-of-the-box agents anyone can deploy: Hunter, Piper, Casey, Paige, and Finn.

Finn is described as the number one customer service agent.
Building agents is a new muscle and skill organizations must learn.
Business users, not just IT, can implement and deploy these agents.

Notes

Opening and Community (02:58–11:26)

Marc Benioff opens Dreamforce, describing a customer meeting in Lison, Switzerland where a Claude-built interface unlocked "trapped value" locked in metadata and systems.
He celebrates 23 million trailblazers across 90 countries and 700 community groups, and highlights giving: $130 million raised for UCSF Children's Hospitals, about $1 billion in grants, and $13 million to the local police force.
Salesforce now runs 15 of 15 cabinet agencies and 50 of 50 US states.
Revenue has moved from the $30 billions to the $40 billions in two years, entering the $50 billions next year.
Finn, called the highest performing customer agent in the world, joins Agent Force.

The Vision: Four Layers (13:47–26:57)

Benioff argues models alone cannot run the enterprise because they are probabilistic and not grounded in a single source of truth.
The answer is four layers: data (Data 360), apps and semantics (now headless), agents, and a new live interface.
Salesforce apps were always metadata-first, so they can be recomposed into AI interfaces like Co-work, Slack, and Lightning.
Zero data retention (ZDR) is built in: customer data does not go into models.
AI Force ships as Cloud Force for Co-work users, Slack Force for Slack, Co-worker for Lightning, plus an SDK.
Salesforce itself has 7,000 users on Cloud Force; Hunter generated $500 million in pipeline last quarter and Casey resolved 5 million service conversations.

The Demo (34:16–42:56)

Patrick Stokes shows his personal Cloud Force interface inside Claude Co-work, built by asking Claude, including pipeline, service hotspots, and marketing data.
A blocked security review threatens $4.88 million of Northstar pipeline; he sends a Slack message via Claude to mobilize his team.
The Salesforce for Claude plugin bundles MCP servers, ZDR, and skills; it is in open beta on the AppExchange.

Dario Amodei Interview (43:44–50:32)

Amodei addresses his essay on pacing with a car-safety analogy: examine your own record, recommit to transparency and safety, organize industry standards, and add an international component.
His biggest surprise is the pace of economic adoption, not just technical progress from scaling laws.
Even frozen in place, only five to ten percent of the technology's possible value is being used; diffusion is the opportunity.
His CCO uses Salesforce with Claude daily to analyze top deals and at-risk deals.

Agent Force (50:32–end)

Agent Force has over 30,000 customers and does 7 billion AWUs; agent script and voice are now GA.
New out-of-the-box agents include Hunter (outbound sales), Piper (website), Casey (customer help), Paige (IT/HR), and Finn.

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