"The software doomsday theory is nonsense. You can't run corporations with AI models alone."
Marc Benioff, Salesforce, Inc. chief executive officer, said this in his keynote at Dreamforce 2026, the annual flagship tech conference held on the 15th (local time) in San Francisco, California. Because Generative AI is probability-based, using only AI models leads to many errors that make it impossible to handle work reliably, he said, adding that true "enterprise AI" that boosts productivity can be realized by running AI on top of existing software that has accurate corporations' data and business rules.
Salesforce, Inc., the world's No. 1 customer relationship management (CRM) software company, dismissed fears of a so-called "SaaSpocalypse," the idea that AI will cannibalize existing software as a service. Instead, it declared "the end of software where people had to do everything themselves."
Benioff said, "There have been all sorts of absurd worries about a SaaSpocalypse over the past six months, but we realized in the end that it only applies to software that doesn't take work off people's hands." He stressed that Salesforce, Inc.'s major software product lines, where AI agents handle tasks in place of people, are growing rapidly.
This trend is showing up in actual results. Benioff said, "Salesforce, Inc. has grown steadily for 27 years, entered the $40 billion in annual revenue range over the past two years, and is expected to surpass $50 billion next year," adding, "Customer attrition in the second quarter this year was the lowest ever, and customer contract duration hit a record high."
◇ Embedding corporations' 'context' into AI models… unveiling AI Force
That day, Salesforce, Inc. rolled out a slew of new AI-based products ready to be put straight to work. Representative examples include the next-generation interface "AI Force," which helps corporations use the data and business functions they have accumulated in Salesforce, Inc. across diverse AI environments, and "Koa," a CRM reasoning model based on Nvidia's "Nemotron."
Benioff said, "AI is triggering an interface revolution," explaining that AI Force serves to smoothly link "the intelligence of AI models" with the business context customers have accumulated in Salesforce, Inc. Until now, people had to directly log into multiple enterprise software tools to do their jobs; now AI will traverse multiple software and work tools to do the work in place of people.
Salesforce, Inc. will offer AI Force in various forms tailored to the environments where users work—such as Co-Work, Slack, and Lightning—and plans to support a software development kit (SDK) so corporations can build their own AI applications. Users of Anthropic's coding tool "Claude Co-Work" will get "Claude Force," while Slack users will get "Slack Force," where people and AI agents execute tasks together based on conversational and work context. For example, if an employee asks Claude for a list of customers to follow up on later, AI will find and analyze the customer information stored in Salesforce, Inc.
AI Force aims to connect four layers—Salesforce, Inc.'s data, applications, AI agents, and real-time interfaces—into one, cutting information search and analysis work that used to take days down to seconds.
It also emphasized that it has secured security and data control by applying a zero data retention (ZDR) principle that prevents corporations' data from being stored in external AI models or used for model training. Benioff said, "Salesforce, Inc. products are built to retain no data at all," adding, "Customers' data doesn't leak outside and remains only customers' data."
◇ Amodei says "we need AI safety standards"… Jensen Huang says "let the market decide"
At the Dreamforce keynote that day, Dario Amodei, Anthropic CEO, and Jensen Huang, Nvidia CEO, also shared views on the industry's hottest topic: calls to "slow the pace of AI development." Earlier, Amodei had argued that the pace should be slowed for AI safety.
Amodei likened AI safety to safety management in the auto industry. "Suppose a competing car company has a brake problem or a defect arises in manufacturing," he said. "You might feel tempted to blame your competitor, but the responsible way to respond to such a situation is to start by reviewing your own record."
He argued that, rather than criticizing accidents at competitors and touting one's own safety, a joint, industrywide response to strengthen safety is needed. Amodei emphasized a three-step approach: first review one's safety record and vulnerabilities to improve practices; then increase transparency and expand investments in safety; and finally work with AI corporations and the international community to establish safety standards that apply across the industry.
He said the pace of AI progress over the past 10 years has been faster than expected. Amodei said, "We predicted roughly how much AI models would advance under the 'scaling laws,' where performance improves as you add computing resources, but we didn't expect it to be this fast," adding, "We're surprised that such speed is possible in the business world."
He went further to assess that "even if the development of AI technology were to stop at the current level, the value of the AI currently used by corporations and society is only 5–10% of AI's potential." He stressed that while developing new models is important, it is also a key task to apply and spread already existing AI to corporations' data and real-world operations.
Jensen Huang, Nvidia CEO, who took the stage next, emphasized that AI safety is an engineering problem. Responding to calls for regulations for AI safety, Huang said, "Market controls are already at work, so no new regulations are needed." He added, "If we can't be confident about performance and safety after sufficiently testing advanced technologies, it's right to delay release, but innovation speed and product safety aren't an either-or choice; we must achieve both." He said innovation should proceed as quickly as possible, but if an AI system is deemed uncontrollable or unsafe, it is necessary to pause and verify it is being developed properly.
