Salesforce CEO Marc Benioff Says Company Will Spend $300 Million on Anthropic Tokens
Salesforce CEO Marc Benioff Says Company Will Spend $300 Million on Anthropic Tokens Salesforce is betting heavily that AI “coding agents” can reshape how software is built, committing hundreds of millions of dollars to Anthropic’s technology even as it reassures workers that humans will still steer the process.
In 2024, Salesforce began reporting major productivity gains from internal AI tools like Agentforce. CEO Marc Benioff told investors that engineering productivity had risen by “more than 30%,” leading the company to halt new software-engineering hires for the following year while expanding roles in sales to explain AI products to customers.
By 2025, Salesforce had deepened its partnership with Anthropic, offering customers access to Claude models to boost “efficiency, insight and personalisation across entire company operations.” Internally, Benioff framed AI as transforming — not replacing — engineering work, saying developers increasingly act in a supervisory capacity over AI systems such as Anthropic models, OpenAI Codex, and Cursor.
In May 2026, Benioff put a headline number on that shift. Speaking on the All-In podcast, he said Salesforce expects to spend about $300 million this year on Anthropic tokens, “almost entirely on coding,” a scale of usage that would make the company one of Anthropic’s largest commercial customers. He described AI coding agents and Anthropic as “awesome,” and argued that the spend will make “everything at Salesforce cheaper to build” by accelerating product development and reducing costs.
Benioff also revealed that Salesforce is developing technology to make coding easier directly inside Slack, its $27.7 billion workplace messaging platform, hinting at “some cool stuff with Slack and code” as Slack evolves into an AI interface.
Across these announcements, Benioff has tried to balance enthusiasm for “unprecedented” efficiency gains with assurances that AI cannot yet operate autonomously, casting the $300 million token spend as part of a new labour model where human engineers oversee increasingly powerful digital co-workers.
Continue reading https://foxvector.com
Write a comment