Marc Benioff on Why AGI Hype Is Overblown for SaaS

TL;DR

Marc Benioff argues today's large language models are just a finite set of algorithms trained on finite internet data, not conscious intelligence, so AGI claims are hype. Salesforce is instead betting entirely on agentic AI tied to its Data Cloud, which combined with AI now exceeds $1 billion in revenue as its fastest-growing product ever.

Transcript

We have all been sold a lot of hypnosis around what's about to happen with AI. You're talking to somebody who is extremely suspect if anybody uses those initials. Agi. Today we're joined by one of the greatest founders of the last 20 years, Mark Benny off from Salesforce. When you look at other data clouds, Snowflake, Data Brick, Palanteer, Foundry... Read More

Key Insights

• Large language models today are two things combined: a finite set of algorithms that have improved incrementally over five years, and a relatively finite set of data pulled off the internet. Together they feel intelligent but are not conscious or a 'being'.
• AGI is treated with deep suspicion by Benioff, who says people have been sold 'hypnosis' around what AI is about to do, and that this is not the actual current state of technology despite how amazing today's AI is.
• Over-reliance on inaccurate AI is a real danger. Benioff cites articles about doctors so dependent on inaccurate AI that they gave patients bad advice while becoming intellectually lazy, calling it a huge warning sign.
• Salesforce refuses to join the AI talent-buying frenzy. While Zuckerberg pays record sums and offers reach a billion dollars, Benioff says the company is focused instead on defining the next generation of the enterprise.
• Agentic support let Salesforce cut human support agents from about 9,000 to about 5,000, rebalancing that headcount into other growing parts of the company. Its help.salesforce.com uses an omni-channel supervisor coordinating human and digital agents.
• Salesforce had over 100 million leads across 26 years it could never call back due to too few SDRs. Its new agentic sales product now calls those leads back and holds conversations, integrated through the omni-channel supervisor.
• Data Cloud plus AI has passed $1 billion in revenue and is Salesforce's fastest-growing cloud product in 26 years. A federated data cloud harmonizing all company data (a reason for buying Informatica) makes the AI more accurate.
• Benioff believes every piece of software Salesforce sells will be agentic, comparing it to how software was never on-prem again after 2000. He promises humans and agents working together across Sales Cloud, Service Cloud, and Slack.

Questions & Answers

Q: Why is Marc Benioff skeptical about AGI?

Benioff says he is extremely suspect of anyone using the initials AGI because people have been sold a lot of hypnosis around what AI is about to do. He argues today's large language models are just a finite set of algorithms trained on a relatively finite set of internet data. They can feel intelligent, but they are not a person, not conscious, have no childhood, and haven't suffered, so AGI is not the current state of technology despite how amazing today's AI is.

Q: What are large language models made of, according to Benioff?

Benioff describes large language models as two things combined. First, a finite set of algorithms that have gotten a lot better but only incrementally over the last five years or so. Second, a relatively finite set of data that has come off the internet. Together these two things provide the current state-of-the-art of LLMs. He compares the feeling of intelligence to using Eliza on a TRS-80 as a teenager: it seemed like a person and felt accurate, but it was not actually intelligent, conscious, or a being.

Q: What is the danger of over-relying on AI in medicine?

Benioff points to two articles he read about doctors who were using AI and became so over-reliant on an AI that was inaccurate that they started giving their patients bad advice. At the same time, those doctors became intellectually lazy. Benioff calls this a huge warning sign for everyone around AI, illustrating that while today's AI is amazing, its inaccuracy combined with human over-dependence can cause real harm and erode people's own judgment and diligence.

Q: Is Salesforce buying AI talent like other big companies?

No. Benioff says Salesforce is not entering the talent-buying frenzy seen at other large incumbents, even as figures like Zuckerberg pay up for talent like never before and offers reach a billion dollars with, in his view, little to show for it beyond the team. Instead, Salesforce is focused on defining the next generation of the enterprise, looking at what is working and what is not, and letting tactics dictate strategy over time in enterprise software.

Q: How has agentic AI changed Salesforce's support operations?

Salesforce built an agentic layer at help.salesforce.com where an omni-channel supervisor balances attention between human support agents and digital agents. This let Benioff reduce the number of human support agents from about 9,000 to about 5,000. That freed headcount was rebalanced into other parts of the company that needed more help and support, since Salesforce is still growing. Benioff frames this as a major change in how the company is structured and how its technology is built and delivered, with Salesforce acting as customer zero.

Q: How is Salesforce using agentic AI for sales leads?

Benioff shares that over 26 years Salesforce had more than 100 million people contact it that it was never able to call back, simply because it did not have enough sales development representatives. Although the company has around 15,000 salespeople, it never had that many SDRs. Now its new agentic sales capability is calling everyone back and having conversations with them, then deeply integrating those interactions through the omni-channel supervisor into the new agentic sales product being shown at Dreamforce.

Q: How big is Salesforce's Data Cloud plus AI business?

Benioff says Data Cloud and AI together now exceed a billion dollars in revenue, a figure disclosed on the last earnings call, and it is the company's fastest-growing cloud product in 26 years. The Data Cloud is federated to all data sources in the company so everything is harmonized in one place, which is part of why Salesforce bought Informatica. Having all data together makes the AI more accurate. Agentforce, which was not shipped until November, now handles as many customer interactions as support agents.

Q: How does Benioff view competitors like Snowflake, Databricks, and Palantir?

Benioff notes that other data clouds such as Snowflake, Databricks, and Palantir Foundry are all around the three to four billion dollar revenue level, and says they are in his sights. He finds Palantir's Foundry, a data cloud integrated with analytics, exciting and inspiring. He positions Salesforce's Data Cloud plus agentic Tableau, Informatica, and MuleSoft as a data foundation, while acknowledging Palantir sells into markets with government certifications that Salesforce does not currently serve, even though the US federal government is already Salesforce's largest customer.

Summary & Key Takeaways

• Benioff is openly suspect of anyone using the initials AGI, arguing today's large language models are a finite set of incrementally improved algorithms trained on finite internet data. They can feel like a person but are not conscious, have no childhood, and haven't suffered, so much of the AI narrative is hypnosis rather than reality.

• Rather than buying AI talent like other incumbents, Salesforce focuses on agentic AI. Its help.salesforce.com agentic support layer, run by an omni-channel supervisor balancing human and digital agents, let it cut support headcount from about 9,000 to 5,000 and redeploy those people into other growing parts of the company.

• Salesforce's Data Cloud combined with AI has surpassed $1 billion in revenue as its fastest-growing product in 26 years. Agentforce, unannounced a year ago and unshipped until November, handles as many interactions as support agents. Benioff eyes Snowflake, Databricks, and Palantir Foundry, all at three to four billion in revenue.