Business Leaders at Dreamforce: Older AI Models Are Sufficient

At Salesforce’s annual Dreamforce mega-conference in San Francisco this week, the keynote conversations between CEO Marc Benioff and the leaders of Anthropic, OpenAI, and Nvidia delved into the intense AI safety debate. The executives weighed in on whether the pace of AI model development is accelerating too rapidly.

Nvidia CEO Jensen Huang, sharing the stage with Benioff, notably urged “frontier labs to run as fast as you can.”

However, on the ground at the massive 50,000-attendee event, a venue described over a decade ago as the “Super Bowl of software,” the prevailing sentiment was different. Far from obsessing over the existential fears of artificial intelligence and the potential threats posed by rapid advancements, many attendees expressed that they are struggling to effectively leverage the technology as it currently exists.

“It’s already hard enough to keep up,” said Alec Bronston, senior Salesforce director at Chicago-based retail data company Spins. He suggested that a potential slowdown would offer “a lot of opportunity to just even catch up and get our feet wet.”

This year’s Dreamforce carries a distinct tone, unfolding amidst a critical juncture in the AI boom. Just days before the conference began, an Anthropic researcher resigned, stating that top AI labs were “gambling with our lives.” This prompted Anthropic’s Dario Amodei and OpenAI’s Sam Altman to propose safety initiatives and advocate for a more measured pace of AI model development.

In the months leading up to the event, Benioff has been actively positioning Salesforce within the burgeoning AI landscape and working to counter the “SaaSpocalypse” narrative. His objective has been to assure investors, customers, and employees that AI serves as an accelerant to Salesforce’s business, rather than a threat. Despite a significant surge in August, Salesforce shares have declined by 8% this year, lagging behind other software companies like Adobe and Autodesk, which have experienced more substantial drops.

During Salesforce’s most recent earnings call in August, the company introduced “Claudeforce,” a new feature designed to enable salespeople to access critical data directly within Anthropic’s Claude chatbot.

On the sprawling exposition floor, where companies within the Salesforce ecosystem invest heavily to showcase their software and services, Anthropic’s booth emerged as the most popular on Tuesday. Young employees of the AI lab, distinguished by oversized white sweaters, engaged passersby with live demonstrations.

Salesforce customers and partners attending the conference indicated to CNBC that existing, more mature, and cost-effective AI models are already sufficiently powerful for routine sales and customer service operations. Many organizations are still in the process of defining their AI budgets and determining whether to utilize models from Anthropic or OpenAI, or to opt for more affordable open-source alternatives.

“The frontier models are way ahead already,” observed Jaya Rohit Vuyyuru, a vice president at consulting firm SummitX. “A lot of the customer base is still getting their feet wet. There’s still that gap where the clients can still fulfill and start getting a sense of what agents can do.”

Meanwhile, as attendees navigated the expansive Dreamforce “Campground,” they encountered whimsical corporate mascots, which generated viral memes across social media platforms. These mascots offered a brief diversion from the pervasive AI discussions that permeated the event.

Some attendees reported that their employers have adopted Salesforce’s Agentforce tools, which are capable of handling customer service inquiries and sales-related queries. It’s noteworthy that for bots built with Agentforce, Salesforce is not relying on Anthropic’s latest Claude Fable 5.1 or OpenAI’s new GPT-6 Astra, according to a support page on its website. This suggests a strategic decision to leverage more established AI capabilities for core functionalities.

“The majority of agentic outcomes aren’t driven by frontier capabilities,” stated Tim Sanders, chief innovation officer at G2, a software review company. “They’re driven by last year’s AI. It’s not that relevant to agentic providers, and certainly not that relevant to SaaS,” or software as a service.

Kevin Lee, chief technology officer at Nice, a vendor of cloud contact center software, indicated that his company does not depend on high-end models like Fable for the majority of its workloads.

“In large part, with the models that are out there already today, and even one generation behind, they are highly performant and effective at doing the things that our customers need,” Lee commented. “It’s almost like everything beyond this point is icing on the cake.”

This perspective does not diminish the transformative power of AI across the software sector. Numerous executives acknowledge its profound impact on their businesses.

Docusign, an electronic signature software developer and Dreamforce exhibitor, employs “all the big frontier models as well as some of the open-weight models,” according to CEO Allan Thygesen.

According to Docusign’s website, “larger frontier models are reserved for judgment-intensive work — complex clause analysis, multi-document reasoning, summarization — where general reasoning capability is worth the higher per-call cost.”

Frontier models are typically proprietary, limiting user access to their underlying datasets or the ability to fine-tune the technology for specific applications. In contrast, open-weight models can be downloaded and deployed on any infrastructure.

During its three-day presence at Dreamforce, Docusign operated from a rented space at the Canopy by Hilton, located across the street from San Francisco’s Moscone Center, providing ample quiet meeting areas. In a recent interview, Thygesen revealed that Docusign utilizes a technique called model routing to direct each request to the most cost-effective AI system.

This approach is a key strategy for Nice and many other companies seeking to manage costs effectively while capitalizing on the benefits of AI.

Tim Sanders of G2 also highlighted that cloud software companies must navigate the evolving token-based economy, where pricing is determined by AI usage and output demand, rather than traditional subscription models.

Software companies need to adapt both in their technological implementation and in their service delivery strategies as customers transition to token-based consumption. A single token, for context, represents approximately three-quarters of a word.

“SaaS up to now has had no variable cost to deliver services,” Sanders explained. “The profit margin structure changed.”

Sanders further elaborated that the shift from a conventional SaaS model to an agentic model could potentially reduce gross margins—the profit remaining after accounting for the cost of goods sold—from upwards of 85% to approximately 45%.

Certain companies are actively pursuing access to the most advanced AI capabilities. Databricks, a data analytics software firm specializing in AI agent development, recently deployed GPT-6 Astra to its entire workforce of 3,500 software developers this week.

“Astra unambiguously out performs our previous highest-end models (Opus 5, Sol 5.6) on highly complex tasks,” Patrick Wendell, vice president of engineering at Databricks, stated on X, referring to models from Anthropic and OpenAI, respectively. This indicates a significant leap in performance for complex computational tasks.

Nagarro, a systems integrator, finds itself in a different strategic position.

Ram Reddy, chief technology officer for consumer industries at Nagarro, spent time at his company’s booth at Dreamforce. Some visitors engaged him in discussions about regulation, following Amodei’s dialogue with Benioff.

However, Reddy noted that the most prominent topic of conversation was Claudeforce, rather than the future trajectory of frontier AI. He mentioned that at Nagarro, engineers typically wait approximately three months before integrating the latest models released by AI labs.

“We are not one of those first early adopters jumping at it,” Reddy concluded.