Salesforce is Already Close to Crushing Last Year’s 12.3 Tn Tokens Record

Enterprises globally are moving from AI pilots to production. Salesforce’s Chief Digital Evangelist, Vala Afshar, believes companies must now learn to operate alongside autonomous AI agents and digital labour.

Speaking at a roundtable discussion during Salesforce’s Mumbai World Tour, Afshar painted a sweeping picture of what he described as the rise of the agentic enterprise, where humans and AI agents work side by side, not as tools and operators, but as colleagues.

“It’s not a strategic imperative anymore,” Afshar said. “You’re either an agentic business or you’re a declining business.”

Speaking about Salesforce, he revealed that the company consumed 12.3 trillion AI tokens last year, but may surpass that number within just a few months this year.

“We’re spending trillions, tens of trillions of tokens this calendar year,” he said. Notably, Salesforce CEO Marc Benioff recently said the company expects to spend nearly $300 million on Anthropic tokens in 2026.

Afshar argued that the economics of AI will not simply revolve around token usage. Instead, enterprises will increasingly measure work units, or tokens tied directly to completed business tasks.

“If a token is used to close a service case, score a marketing lead, or shift an opportunity in sales, those aren’t just any tokens,” he said. “Those are work units.”

However, he added that enterprises need to rethink efficiency. “Not every process needs to be agentified,” he said.

Afshar described six stages of enterprise AI transformation—redesign, reskill, redeploy, restructure, recalibrate, and reclaim.

He said Salesforce itself moved around 3,000 employees into sales roles without making equivalent new hires because AI agents had taken over repetitive operational work in other parts of the company. “We were able to reskill people who were doing repetitive low-impact work that agents are now doing.”

Notably, last year, Salesforce cut its customer support workforce from roughly 9,000 to 5,000 employees, with AI agents now reportedly handling about half of support interactions.

Benioff recently said Salesforce is not planning to hire more software engineers because AI tools have significantly improved productivity internally, including in engineering, customer support, marketing, and analytics functions.

Salesforce Goes ‘Headless’ 

Afshar said that enterprise software itself will evolve away from human-centric interfaces. “Humans log into websites. Humans use dashboards. Agents don’t,” he commented.

Salesforce recently announced Headless AgentForce 360, allowing AI agents to directly interact with enterprise systems via APIs and Model Context Protocol (MCP) servers.

“We removed all the human user interfaces,” Afshar said. “Those agents can get the information they need, the rules they need, and execute tasks for companies.”

According to him, enterprises must now design products and services assuming that future consumers may increasingly be AI agents rather than people.

Software is No Longer Just a Tool

Afshar said that software is now turning into active digital labour.

“For about 24 years, we built software, we gave it to humans, and humans used the software as a tool to do their jobs,” he said. “Now, we are presenting software that you can onboard as a digital colleague.”

He described this as the emergence of a hybrid work environment with collaboration between humans and intelligent agents.

Afshar said 60,000 employees at Salesforce are now daily active users of Slackbot, which has evolved into an agentic AI assistant. Employees use it to prepare for customer meetings and conduct research.

He further explained that software can now execute tasks for the users. “Humans and AIs are co-existing, co-collaborating, and co-creating value at a greater speed, scale, and intelligence unlike anything we’ve experienced before.”

Afshar said that enterprises deploying AI agents are fundamentally reallocating work. Repetitive and high-volume tasks are increasingly handled by software systems, while human workers shift toward creative, strategic, and higher-complexity responsibilities.

“These companies are going through what I call cognitive transfer,” he said. “They’re transferring highly repetitive and often low-impact work to software.”

No Company for the Average

Despite the optimism around productivity and scale, Afshar’s broader message was ultimately one of urgency. He argued that autonomous enterprises will identify inefficiencies and average performance much faster than traditional companies.

“There’s no demand for average anymore,” he said. “If you’re an average company producing average products and average service, you’re not going to compete.”

For boards and CEOs, he suggested AI adoption is rapidly becoming existential rather than experimental. “You no longer have an excuse to ignore poor customer experiences or employee attrition,” he said. “Without discipline and implementation, your company is on a path of irrelevance.”