On Salesforce’s latest earnings call, CEO Marc Benioff addressed the strategic question hanging over every enterprise AI roadmap: what happens if foundation models stop being infrastructure and start becoming platforms?
Just as Windows, macOS, iOS, and HTML became foundations for entire application ecosystems, large language models (LLMs) could eventually host applications directly, taking value away from the software layers above them.
Benioff acknowledged that possibility:
“Could those models themselves become platforms? Could OpenAI then also be a platform? Could Anthropic be a platform? Absolutely, those could be new platforms.”
Salesforce’s response, however, is not to compete at the model layer. Instead, the company is doubling down on what those platforms do not yet deliver at enterprise scale: trusted context, governed workflows, compliance, security, reliability, and the ability to turn intelligence into real customer work. As Benioff put it:
“Our job as a software company is to help our customers to create success, and to take that and help them connect with their customers in a whole new way.”
That framing set the tone for how Salesforce is positioning Agentforce inside customer experience, sales, and service operations.
Why Salesforce Believes It Still Owns the Enterprise Layer
Benioff described foundation models as a new layer that sits firmly at the bottom of Salesforce’s stack.
“These models are new parts of our infrastructure that we really did not have in place a few years ago.”
Salesforce has long run its own models, but today’s environment is defined by scale and plurality. Intelligence now flows in from multiple partners, including OpenAI and Anthropic.
“We’ve always had models at the bottom of our infrastructure, but now we really are able to say, ‘Look at this. We’ve done 19 trillion tokens with these models.’”
Those tokens represent consumption of intelligence, but they do not, in Salesforce’s view, represent value by themselves. The value appears higher up the stack.
Even if models become platforms, Benioff argued that they are not yet equipped to run regulated, customer-facing enterprise operations.
“There is a lot to do to not only automate… those contact centers, the Salesforces, the employees with Slack, to also then unleash the agents in a way that is compliant, that is secure, that is available, that is scalable, that is reliable.”
The current reality is “humans and agents working together,” Benioff added, with Salesforce positioned as the system that makes that coexistence operational.
What Agentforce Adoption Reveals About Enterprise AI Readiness
Enterprise AI adoption is still a way from large-scale replacement of software as a service (SaaS). And within Salesforce, use of its Agentforce AI agent platform remains a small part of its overall customer base. The company reported close to 50 percent growth in Agentforce customers during the fourth quarter of last year to approximately 22,000-23,000, with 29,000 Agentforce transactions. But that sits within Salesforce’s much broader footprint of more than 150,000 customers globally and over 1 million users on Slack, a reminder that agentic adoption is still early relative to its total base.
One of the clearest signals on the call was Salesforce’s attempt to shift how enterprise AI value is measured.
Patrick Stokes, Salesforce’s President and Chief Marketing Officer, explained why token counts fall short:
“You can ask it a question, it can write you a poem, but that’s not really all that valuable in the enterprise world. What’s valuable is creating a document for you or updating a record.”
That thinking led Salesforce to introduce Agentic Work Units (AWUs), a metric designed to track completed actions rather than raw intelligence usage, Stokes said.
