Marc Benioff Says the Software Bears Are All Wrong About Salesforce
‘People think we have our back against the wall,’ but customers aren’t replacing its offerings with AI, CEO says
Salesforce plans to unveil a new AI platform, Agent Albert, by year-end, aiming to make its software more valuable amid AI disruption fears.
Its existing AI product, Agentforce, is used by 23,000 customers, but early reviews were tepid due to data preparation issues.
Salesforce invested over $300 million in Anthropic. Shares jumped after the company announced an integration with Claude Cowork in February.
Marc Benioff has some problems.
His enterprise software company, Salesforce CRM 0.51%increase; green up pointing triangle, is the biggest name in a category that Wall Street thinks may get decimated by artificial intelligence.
Its business model has centered on selling software to large companies on a per-employee basis—but many of those firms are expected to downsize as AI agents become increasingly proficient at performing real-world tasks.
It is a daunting double-bind, and it isn’t even the worst-case scenario. Salesforce’s stock is down a mere 28% year to date. The hardest-hit software-as-a-service companies are down about twice that much on similar fears.
But Benioff thinks the bears have it wrong—about the “SaaSpocalypse” thesis generally, and especially about Salesforce. AI, he says, is making Salesforce more valuable to its customers than ever. The leading AI labs couldn’t replace what Salesforce offers even if they wanted to, and they would rather partner with him—for now, anyway. Nor could customers easily “vibe-code” their own sales-management software that could compete with Salesforce on security, compliance and other vital features.
“People think we have our back against the wall when in fact the opportunity has never been greater,” Benioff said in an interview.
An early investor in Anthropic, Salesforce has been developing and pushing its own AI tools for years. By the end of this year, it plans to unveil a new AI platform that automatically studies its users and takes actions on their behalf. Code-named Agent Albert, it is the culmination of an effort that began three years ago, when Benioff, galvanized by the debut of ChatGPT, instituted a standing Saturday meeting to accelerate Salesforce’s AI efforts.
An earlier flagship product of that push, Agentforce, has been somewhat slow to gain traction. Launched in late 2024, it is used by 23,000 customers out of the total base of 150,000 to build custom autonomous agents for use in workflows. Salesforce’s yearly revenue growth, at around 10%, is down somewhat from recent years.
There are signs its most important constituency is beginning to buy what Benioff is selling. Business customers spending big on AI have increased their average median spend on Salesforce in the past three months by 3%, according to data from the venture-capital firm Andreessen Horowitz. That is better than Zoom, Dropbox and Adobe although worse than Canva, Cloudflare, Docusign and others.
Mike Kimbarovsky, a partner at Salesforce investor Chicago Capital, which recently sold shares in the company, said he has been impressed with some of its recent moves, such as deepening its integration with Anthropic’s models and experimenting with new pricing structures.
But what Salesforce really needs, he said, is positive word-of-mouth from clients talking up the value they derive from its AI products. “They need to show revolutionary jumps,” he said.
Benioff has said Salesforce was destined to be an AI-first company as far back as 2014, when it launched its AI research unit. Still, it was caught flat-footed by the arrival of high-functioning chatbots in the form of ChatGPT in late 2022.
Benioff responded quickly. In early 2023, he held a three-day meeting at Salesforce Tower in downtown San Francisco with roughly 40 executives and created an internal chat group called “AI at Salesforce.” The goal: Rewriting the annual plan to make room for AI in the company’s core businesses.
Benioff and other executives spent hours every Saturday for months working on the pivot, said Steve Fisher, president and chief product officer.
The arrival of Agentforce the following year gave Salesforce a way to tap in to rising enterprise interest in AI. It pitched customers on using the product, powered by models from OpenAI, Anthropic and other providers, to create agents that could handle customer-service tickets, qualify leads ahead of sales meetings and other tasks.
But early reviews were tepid. Customers complained of having to spend half their time preparing data so the AI could understand it, limiting the platform’s effectiveness. To help fix the problems, Salesforce built a layer into its tech stack that automatically pulls in customer data from external sources, and it purchased a string of companies that include firms specializing in data management and AI-powered sales.
At education company Pearson, agents now autonomously handle queries about order statuses, refunds and lost access codes for its customers. This has increased the percentage of customer questions that don’t involve human interaction by 40%, said Gabriele Bauman, a vice president at Pearson.
Shree Reddy, PenFed Credit Union executive vice president, said the company deployed an Agentforce agent to help its employees reset passwords and unlock accounts, reducing total IT tickets created by 40%.
Where Agentforce has been lacking is in addressing complex customer problems or those that require a human touch. David Walmsley, chief digital and technology officer at Pandora jewelry, said Agentforce hasn’t been able to reliably recommend products on its own based on the vague context that customers share through its website such as “my wife likes dogs; what should I buy her?”
Another timely action that Benioff took in early 2023: investing in Anthropic’s series C venture-funding round. Beginning with that deal, Benioff said Salesforce has invested more than $300 million into the company, which is considering an initial public offering as soon as this year. Anthropic is currently valued at $380 billion.
Potentially more important to Salesforce than the value of its stake is the relationship. Anthropic makes a practice of partnering with sector-leading companies for product announcements, which can otherwise spook markets into punishing companies seen as standing to be affected. In February, when the two companies announced that the new version of Claude Cowork would integrate with Salesforce apps, Salesforce’s shares jumped 4 %.
Salesforce also uses OpenAI models for Agentforce, and OpenAI, like Anthropic, uses the Salesforce-owned workplace app Slack, which Salesforce is making into a primary interface for its AI tools. Salesforce would have invested in OpenAI, but the startup’s initial contract with Microsoft prohibited such a deal, Benioff said.
Arguing against the idea that Salesforce’s customers could use Claude Code or OpenAI’s alternative, Codex, to vibe code replacements for Salesforce, Benioff said their needs around brand safety, data security and industry-specific compliance make that risky, and Salesforce’s deep experience in those areas is indispensable.
Agents created within Agentforce are constrained by custom business rules that Salesforce calls “scaffolding.” For instance, when the recruiting firm Adecco uses Agentforce agents to screen candidates, they are automatically trained to follow all of its compliance requirements. A human recruiter is still significantly involved in the hiring process, the company said.
In a note to clients, Stifel investment analysts J. Parker Lane and Jack McShane echoed this logic, noting that agent operations “are occurring in high-stakes customer-interaction environments.”
“While the market is noisy with startups, Salesforce is winning deals because CIOs and CTOs prefer a unified platform that integrates agents, actions, data, and workflows,” they added.
However much customers might warm to Agentforce and the new Agent Albert, it doesn’t do Salesforce much good if it is charging by the seat while customers use AI to do more work with fewer employees.
Nearly a year ago, the company launched a hybrid pricing model in which companies can still pay for “seat” licenses for its products but pay more for Agentforce based on consumption. The company says it prices per action taken by its agents, and it also offers a license agreement that enables customers to go “wall to wall.”
Looking for a new way to quantify the value its agents create for customers, it introduced a tracking metric this year it calls an Agentic Work Unit, which it says measures the conversion of generative AI capabilities into tangible results like resolving customer cases or updating records.
In its latest quarterly earnings report, Salesforce said it had processed 2.4 billion AWUs, increasing 57% quarter-over-quarter, the first time it reported the figure.
