Salesforce CEO says LLM capabilities are nearing their limit

Marc Benioff, CEO of Salesforce, stated that the future of artificial intelligence (AI) focuses on autonomous agents instead of large language models (LLMs), claiming that the latter have reached their “upper limits.” In a recent episode of The Wall Street Journal‘s “Future of Everything” podcast on November 23, Benioff argued that society has become overly reliant on tools like ChatGPT, leading to inflated expectations regarding AI’s capabilities. He emphasized that while LLMs have played a significant role, the next phase in AI evolution will be characterized by agentic AI that operates independently.

Benioff criticized the notion that LLMs are the centerpiece of AI advancement, expressing concern that this perception has skewed public understanding of technology. He believes that autonomous agents, capable of executing tasks such as sales communications and marketing campaigns, will contribute more to organizational efficiency. Salesforce is actively pursuing this vision through prebuilt AI agents tailored for customer service automation. “We have incredible tools to augment our productivity, to augment our employees… improve our customer relationships,” Benioff stated.

This perspective on agents contrasts sharply with the predominantly chatbot-oriented narrative in the industry. While Benioff acknowledged the current utility of LLMs, he warned against overhyping their potential. He described the media portrayal of AI, influenced by films like “Terminator” and “Minority Report,” as misleading. The fascination with such narratives may lead people to expect solutions to significant issues, such as climate change or cancer, albeit those claims are unfounded at this stage.

Salesforce is gearing up to capitalize on the capabilities of autonomous agents, positioning them as essential tools for enhancing customer engagement and streamlining operations. Meanwhile, OpenAI is set to introduce its own agent, codenamed “Operator,” in January 2025. This agent will be designed to independently carry out tasks such as code writing and travel bookings for users.

Salesforce CEO’s big praise for Google Gemini Live

Nvidia is also joining the movement towards agentic AI. CEO Jensen Huang noted in a recent earnings call that enterprise adoption of AI agents is becoming increasingly prevalent. He acknowledged the potential of these tools to reshape workplace dynamics, suggesting that companies will soon have AI-driven ’employees’ working side by side with human staff.

In emphasizing the limitations of current AI, Benioff addressed what he terms “AI priests and priestesses”—enthusiasts who exaggerate the technology’s capabilities. He asserted that these inflated representations could mislead businesses that stand to gain from practical applications of AI, thus hindering true advancement in productivity and profitability.

Benioff’s skepticism about exaggerated claims serves as a reminder to balance technological enthusiasm with realistic expectations. He reiterated, “Yes, you can do all of these things with AI, but this other part — that we are all living in ‘Minority Report’? No, we’re not there yet.”

In summary, as Salesforce and other industry giants look towards a future populated with autonomous agents, the conversation surrounding AI is shifting. Benioff’s insights highlight the enterprise appetite for tools that can manage real-world tasks effectively while recognizing the current technological boundaries.

Featured image credit: Igor Omilaev/Unsplash