Salesforce Inc. keeps putting all of its eggs in the agent basket — but the strategy appears to be paying off so far.

At Dreamforce 2026, Marc Benioff, chief executive officer of Salesforce, bantered with Nvidia Corp. CEO Jensen Huang and led the announcement of a new artificial intelligence interface layer, AIforce. The event also unveiled more details about Headless 360, which enables users to access Salesforce software without its user interface.

“This whole headless thing points to the fact that AI agents are not another feature cycle, meaning agents aren’t a feature of the enterprise,” said John Furrier (pictured, left), executive analyst at theCUBE Research. “They’re forcing a complete reconstruction of the enterprise stack … from the applications, the end interfaces, down through security, infrastructure and capital allocation.”

On the latest episode of theCUBE Pod, Furrier and Dave Vellante (right), chief analyst at theCUBE Research, discussed takeaways from Dreamforce 2026. They also analyzed the development of AI agents in the enterprise and Nscale Global Holdings Ltd.’s initial public offering filing.

Salesforce ‘system of intelligence’ takes shape

Dreamforce reflected Salesforce’s ambition to create what it terms an Enterprise AI Harness, which combines data, business knowledge, workflows and control into a common, composable architecture for agents. This announcement signifies a shift toward a “system of intelligence” that acts as a map for agents to understand a business’s processes.

“Headless becomes that system of intelligence,” Vellante explained. “That’s where the data is harmonized, the process knowledge gets injected. You see harnesses within the LLMs, but they’re not deterministic. AIforce … allows any client surface to basically create the interface. The reality is this stuff is not mature — Agentforce is still in its early days — but there’s clear business value.” 

Salesforce has fully committed to agents, perhaps in an attempt to avoid the potential “SaaSpocalypse” that is still spooking investors. Much of its new stack is formed from composable agents, which give users greater flexibility and security — at least, in theory.

“Their advantage might be the remaining system of data, the workflow, the identity data and the business context underneath it all,” Furrier said. “That’s why I think Slack becomes a potential agentic front door. Unstructured data works well with graphs, works well with the infrastructure that AI likes. It’d be very interesting to see how they develop that.”

The on-premises renaissance has continued, becoming part of the bedrock of AI revenue. Salesforce is still in the process of figuring out its pricing, Vellante noted, while research from Qualitate suggests that 75% of respondents who had modeled or experienced the financial impact of headless access expect their Salesforce spending to increase, not decline.

“You’re seeing Claude create the interface,” Vellante said. “You’re seeing Slack create the interface. Virtually, any harness is going to be able to create the interface. That’s going to drive token demand. Where’s that token demand going to come from? Core infrastructure in the data center. A lot of that is going to be on-prem. There’s no question about it.”

Nscale files for an IPO

Breaking news came in during theCUBE Pod: Nscale, a U.K.-based company that builds AI-optimized data centers, filed to go public on the New York Stock Exchange. The Nvidia-backed company, which has yet to become profitable, is betting on continued growth in demand for its data center capacity.

“[Nscale] they’re an example of the kind of playbook CoreWeave built, which is fully integrated business,” Furrier said. “They went full-stack. They bought Anyscale. So, huge cloud provider opportunity for this AI infrastructure serving enterprises. Nebius and CoreWeave have a new competitor. This IPO is expected to be pretty smashing and we’ll see how it goes.”

Furrier also took some time to analyze the development of agents within the enterprise world. AI is starting to make more autonomous decisions based on enterprise operational data, and with the new headless offering from Salesforce, agents could fundamentally reshape how businesses operate.

“What happens when agents start to learn about how enterprises actually work over time?” Furrier asked. “You have a new training inference dataset. Spirit Airlines bankruptcy — there’s a bid for the data, 20 years of data. The buyers of that data were for training purposes. That brings up the commoditization of the system of record. At what point does that become disposable?”

Here’s the full episode of this week’s theCUBE Pod:

You can also listen to us on Apple Podcasts or on Spotify. And for those who prefer to watch, check out our YouTube playlist. Tune in now, and be part of the ongoing conversation.

Photo: SiliconANGLE