One of the tech industry’s biggest hype men, Salesforce CEO Marc Benioff, wants you to know that he is psyched about generative AI, but even he doesn’t think it can do what its biggest proponents say it will do. And he (not surprisingly) blames Microsoft.

A month after Salesforce’s enormous tech conference, Dreamforce, during which Benioff endlessly pitched, pumped, and lauded AI — at least as it’s used in Salesforce’s own products – he’s now on a setting-expectations tour. And he’s trash talking his biggest competitor and arch rival, Microsoft. He recently appeared on the podcast Rapid Response, hosted by former Fast Company editor-in-chief Bob Safian. 

Talking about AI’s potential, he said, “I’ve never been more excited about anything at Salesforce, maybe in my career.”

But he also warned that “customers have been told things about enterprise AI, maybe AI overall, that are not true,” he said. “I think Microsoft has done a tremendous disservice to not only our whole industry but all of the AI research that has been done.”

Benioff had negative things to say in particular about Microsoft Copilot’s accuracy and usefulness. He even compared Copilot to Clippy, Microsoft’s widely panned 1990’s talking paperclip cartoon that was supposed to be an assistant to Microsoft Office users.

“We may have heard from these AI priests and priestesses of these LLM model companies and Microsoft and others about AI is now curing cancer, and AI is curing climate change, and we all have to plug into these nuclear power plants to get these data centers. None of this is true,” Benoff said. 

He was doubling down on a comment he made on X, where he said, “LLMs (Large Language Models) are not the direct bridge to AGI, and much of AI’s current potential is simply oversold. AI isn’t yet curing cancer or solving climate change as pundits claim.”

That was a dig at OpenAI CEO Sam Altman. Earlier this summer he postulated that with AI-enhanced health tech, “Maybe a future version will help discover cures for cancer,” Altman said at the Aspen Ideas Festival, reported Newsweek.

Benioff also cited some research by Gartner about Microsoft Copilot. A Gartner report released in April called “The Top 10 ‘Gotchas’ of Copilot for Microsoft 365” found that only a quarter of the organizations doing Copilot pilot programs are currently planning a large-scale rollout. That’s actually a pretty good number, given how young this tech is, and how slow enterprises can be to adopt it. That said, Gartner also concluded that as Copilot improves, so will enterprise adoption.

Microsoft might point to market research conducted by Forrester that cited a slew of bottom-line benefits for small businesses who are using Copilot. Forrester found that Copilot led to a tiny uptick in revenue, while reducing operating costs and speeding new-hire onboarding, based on a survey of 266 small companies. However, we must note that this research was paid for by Microsoft. Make of that what you will.

Still, Benioff has a point that the GenAI we have today, while sometimes mind-blowing, is not ready to replace human workers in most cases. True, the podcast-making capabilities of Google’s NotebookLM – which can create bantering AI-generated hosts that explain material –  is quite the party trick. Yet, it’s hard to see how such a thing will directly reduce the kind of soul-sucking menial labor tasks that bog down most corporate jobs. 

Benioff is also spot-on that the one area where GenAI does seem to be doing a screaming good job with enterprises is AI agents. That ties in nicely with the Salesforce product he’s been hyping lately, Agentforce. A slew of other tech companies and startups are working on AI agent technologies, too, from building use-case specific ones, to offering platforms where businesses can build their own. Just a few examples include OutRival, Atlassian’s Rovo AI, and Sierra, the startup founded by former Salesforce co-CEO Bret Taylor and Google vet Clay Bavor.

AI agents are especially finding their way into customer service, ranging from much improved website chat bots to field service guides. Examples include Zingtree, Talla or Neuron7. Salesforce has offerings here, too.

AI agents are also finding success with sales, too, especially prospecting, the high-rejection rate cold-calling and emailing that is the bottom tier of every sales organization. Examples include AI Regie.ai, AiSDR, Artisan and 11x.ai.

“I think we’ll have more than a billion agents running from Salesforce within the next 12 months,” Benioff speculated, based, he said, on getting about 10,000 customers at his tech conference to try it.

However, it’s also worth pointing out that there are other areas where LLMs are already valued and where Microsoft, specifically, does have game. Software programmers and engineers increasingly use them to help them test and debug or generate code examples, including with Microsoft’s GitHub Copilot. Countless startups are offering AI coding assistants as well, like JetBrains and Continue, to name just a couple. 

Microsoft’s close relationship with OpenAI means that its cloud, Azure, is a popular choice for enterprises who are using LLM models to build their own GenAI apps, too. So, Microsoft’s tentacles into GenAI extend far beyond having Word write documents, having Excel whip up charts or having Teams transcribe meetings. 

At the same time, when the people who are literally talking up AI to sell their own AI products warn that the AI is overhyped, it’s safe to say that AI is overhyped. 

“It’s about managing expectations while harnessing AI’s capabilities,” Benioff explained.