Salesforce CEO Marc Benioff Thinks AI Has Hit a Roadblock

It’s hard to turn on a computer and not see evidence of AI’s advances into our online lives. It’s in the Microsoft or Google tools you use on your work PC, and the social media apps you use to escape the stresses of reality, and it seems that some kind of buzzy new AI advance gets announced almost daily. But are all these AI chatbots, with ChatGPT in the lead, actually as smart as we think they are? One tech leader, Salesforce CEO Marc Benioff, is beginning to doubt the hype. In fact Benioff thinks we may have hit a ceiling in the development of “large language model” (LLM) AIs, and suggests they actually won’t get much smarter, despite the news of new models or new capabilities. Where the real next-gen AI action is at, Benioff thinks, is actually in AI agents, not chatbots, and he’s betting big on that prediction within his own company. In an interview on the Wall Street Journal’s Future of Everything podcast, Benioff explained his thinking. Essentially even though AI companies are desperately trying to push for “next generation” LLM chatbots, like the much-rumored GPT-5 from OpenAI, Benioff thinks we’re “hitting the upper limits of the LLMs right now.” Benioff admits we have “incredible tools to augment our productivity, to augment our employees, to prove our margins, to prove our revenues, to make our companies fundamentally better,” and even “have higher fidelity relationships with our customers.” But he also says we’re nowhere near the level of AI seen in in “these crazy movies”—meaning the kind of super-smart AI seen in popular sci-fi. In particular Benioff worries that there are some players in the AI game who are evangelizing the tech by suggesting it can solve some of the world’s biggest problems, but it really can’t, and that’s actually a distraction from the actual benefits AI can provide. What’s really coming up in AI tech, Benioff thinks, isn’t super-smart AIs like in the Terminator movies (and we’re all glad the apocalyptic vision of the franchise hasn’t come to pass. Yet.) but powerful “agentic” AI. While chatbots work in a call-and-response style, answering queries when users ask for help, AI agents are chunks of code that can actually perform “actions” in an online environment, like finding appropriate data and then using it for filling in forms, or pressing “buy” on a shopping cart in an online store. In an X posting yesterday, Benioff argued that government “regulatory, compliance, and political demands” are “consuming up to 40% of budgets,” and they’re growing fast. So it’s time for a “transformation” via AI agents, which can “revolutionize operations-automating reporting, audits, case management,” and more. He suggested it was time to “replace bureaucracy with an agentic layer that serves people, not politics.” He added a personal spin on the idea, by saying “Welcome to the future—welcome Agentforce!” in a blatant advert for his company’s recently unveiled agent-based AI system called Agentforce that can, at launch, act like a digital sales rep. Why should we care, though? Benioff is a billionaire, and though he’s certainly got his finger on the pulse of tech, and he’s got skin in the game, his company isn’t developing cutting-edge AI in the same manner as OpenAI or Google. Though his posting on X was aimed at a certain sector—the paperwork load from various government offices—Benioff is essentially predicting the near future of AI-assisted work, where many menial or frustrating “bureaucratic” office tasks are dramatically sped up by agent-based tools. AI critics will worry Benioff is predicting AI will replace some, perhaps more menial, office roles—but arguably he’s saying agents will free up employees’ time to be more effective at the tasks that actually comprise their jobs: for example, if filing a travel expense takes up a few hours of a worker’s day, they’re not going to be contributing to the company’s bottom line…but if an AI can do that job for them, then they’ve gained two useful work hours. And Benioff’s words about slow-paced AI development may ring true in other ways: recently it emerged that AI giant OpenAI was struggling to develop its next-gen ChatGPT engine, and was being forced to try wholly new tactics. If we’re all expecting “smart” AIs to transform our workplace, we may have to wait a while. BY KIT EATON @KITEATON