Salesforce to Spend $300 Million on Anthropic Tokens — While Freezing Engineer Hiring
CEO Marc Benioff says AI coding agents are "awesome" and have made human engineers 30–50% more productive, prompting a radical shift in how the company spends its capital.
Salesforce CEO Marc Benioff has revealed that his company could spend nearly $300 million on Anthropic AI tokens in 2026 — an extraordinary figure that signals a fundamental reorientation of how enterprise technology companies allocate capital. Speaking on the All-In podcast, Benioff said most of the spending is tied to coding work powered by AI agents and Claude, Anthropic's flagship model.
"These coding agents are awesome. Anthropic is awesome. I am going to probably use $300 million of Anthropic this year at Salesforce. Coding. Everything's going to be cheaper to make."
— Marc Benioff, CEO, Salesforce · All-In Podcast, May 2026The announcement comes on the heels of a bold decision Benioff made in early 2025: Salesforce would not be hiring any additional software engineers that year. The rationale was straightforward — AI tools, including Salesforce's own Agentforce platform, had boosted engineering productivity by more than 30%, making new headcount unnecessary. That freeze has now extended into 2026, while AI spending balloons in its place.
From headcount to token count
The shift represents more than a cost substitution. Benioff described it as a "digital labour revolution" — one in which AI agents are not simply assisting engineers but actively taking on large portions of their workload. By mid-2025, he reported that AI was handling between 30% and 50% of Salesforce's total global output, across engineering, service, support, and marketing.
The financial logic is stark. Salesforce cut its support staff from 9,000 to 5,000 by deploying AI agents. The savings from reduced headcount are now being redirected toward API token consumption — specifically, toward Anthropic's Claude models, which Salesforce uses across engineering pipelines and increasingly within Slack, the workplace communication platform it owns.
Key developments at a glance
Engineers aren't gone — they're managing AI
Benioff was careful to clarify that AI has not reached a point where it can fully replace human engineers. Salesforce's 15,000-strong engineering workforce is evolving rather than disappearing. Engineers are increasingly moving into supervisory roles — overseeing AI-generated code, reviewing outputs, and directing agents rather than writing all code manually. Tools in use include Anthropic's Claude models, OpenAI Codex, and Cursor AI.
The CEO noted that human oversight remains essential because AI models cannot yet function fully autonomously. "Engineers are not simply working with AI," he said, "but with agents that actively assist them in coding tasks." The result, in his view, is a team that produces far more than before — with a dramatically smaller marginal cost per unit of output.
Managing the cost of intelligence
At $300 million in projected annual token spending, even incremental efficiency gains matter enormously. Benioff said Salesforce is developing an internal routing layer that will direct tasks intelligently — sending complex, high-stakes reasoning to Claude while routing simpler, repetitive tasks to smaller and cheaper models. The logic: not every request needs a frontier model. Routine work can be handled by lightweight AI, reserving Anthropic's most powerful models for problems that genuinely require them.
This kind of tiered model orchestration is becoming a standard feature of enterprise AI strategy, as companies realize that indiscriminate use of large language models can generate enormous and unnecessary costs. Benioff's framing suggests Salesforce is thinking carefully about ROI — not just capability.
A blueprint for the tech industry
Salesforce's trajectory is being watched closely across Silicon Valley. The combination of an engineering hiring freeze, massive AI tool investment, and a pivot toward token-based spending over headcount growth represents a template that other enterprise tech companies may follow. The question for the broader industry is whether the productivity gains are real and durable — or whether the efficiencies plateau as AI tools mature and competition for the same capabilities intensifies.
For now, Benioff's enthusiasm is unambiguous. He described the pace of product iteration as having "accelerated sharply" and said he can now ship and sell software in parallel — something he described as impossible before AI agents entered the picture. Whether $300 million in Anthropic tokens delivers commensurate returns will be one of the defining business stories of 2026.
References
