Marc Benioff does not hedge. On a recent episode of the All-In podcast the Salesforce chief executive laid out plans that would have seemed audacious even a year ago. His company expects to spend roughly $300 million this year on tokens from Anthropic. Nearly all of it will go toward coding agents. “These coding agents are awesome. Anthropic is awesome,” he said. “I am going to probably use $300 million of Anthropic (tokens) this year at Salesforce. Coding. Everything’s going to be cheaper to make.”
The statement landed with force. It marks a shift from equity investor to one of the AI startup’s biggest commercial customers. Salesforce first put more than $300 million into Anthropic starting with its Series C round in early 2023. That stake, now roughly 1 percent in a company valued near $380 billion, has delivered paper returns exceeding ten times the original outlay. Salesforce Ventures continued backing every round through the latest Series G in February 2026. But the new annual token commitment signals something different. Consumption at scale has begun.
Tokens represent the basic currency of large language models. They are fragments of text that models process to generate output. Providers bill enterprises according to volume consumed. At current pricing for Claude Opus 4.7, which runs $5 per million input tokens and $25 per million output tokens, $300 million buys enormous computational work. And Benioff insists the return already shows. AI agents have produced “unprecedented” efficiency gains across service, support, distribution and marketing. Last August he announced the company had cut its support workforce from 9,000 to 5,000 thanks to those agents.
Coding now stands as the next frontier. Faster product iteration. Lower development costs. Output at a pace once unimaginable. “I can do things that I just could not do before. I can go faster than ever before. I can implement my software and sell it at the same time,” Benioff explained. “I’ve never been able to do that before. Today, I have humans, agents, and headless platforms all interoperating, never before.” The implications stretch beyond one software maker. If engineering teams can build and ship at multiplied speed, entire business models face pressure to adapt or risk obsolescence.
Yet not every task requires the most powerful model. Benioff made this point explicit. He does not believe every token generated by employees must flow to a frontier system like Claude. Instead he called for an intermediary routing layer. Complex reasoning tasks go to Anthropic’s best models. Simpler queries route to smaller, far cheaper alternatives from Anthropic’s Haiku line, Meta’s Llama or DeepSeek. Even modest optimization at this spending level could save tens of millions. And Benioff signaled Salesforce intends to build that layer rather than wait for others to provide it. The endorsement from the leader of the world’s largest enterprise software provider carries weight. It frames model routing as table stakes for cost control in the agent era.
Slack sits at the center of these ambitions. Salesforce acquired the workplace messaging platform for $27.7 billion in 2021. Benioff has long called it “the interface to AI.” In March the company rolled out more than 30 new AI capabilities that transformed Slackbot from conversational helper into an agentic system. It transcribes meetings, monitors desktop activity, executes tasks through third-party tools via the Model Context Protocol and even functions as a lightweight CRM. All of it runs on Claude. Slack revenue is projected to reach $3 billion this year. Starting this summer every new Salesforce customer will receive the platform automatically provisioned and AI-enabled from day one.
Now Benioff wants coding inside Slack itself. “We’re even working on technology inside Slack to make it easier for everybody to code,” he told the podcast. “You’re going to see some cool stuff with Slack and code I’m not ready to talk about yet. But there’s no question that we are in a new moment in coding.” The remarks build on a deepening partnership. In October 2025 the two companies expanded their collaboration to deliver trusted AI for regulated industries. Claude became a foundational model for Salesforce’s Agentforce 360 platform, available securely within the company’s trust boundary for sectors such as financial services, healthcare, cybersecurity and life sciences.
“Salesforce and Anthropic share a vision for a trusted AI ecosystem that puts customers at the center,” Benioff said in the official announcement. “By bringing Salesforce directly into Claude — and Claude’s intelligence into Salesforce and Slack — we’re giving every company the power to work in entirely new ways.” Dario Amodei, Anthropic’s CEO, echoed the focus on safeguards. “Regulated industries need frontier AI capabilities, but they also need the appropriate safeguards before they can deploy in sensitive systems. We’ve built Claude to deliver both: the performance and the safeguards.”
Early customers already demonstrate the potential. CrowdStrike and RBC Wealth Management use Claude via Amazon Bedrock in Agentforce. At RBC the integration helps advisors prepare for client meetings far faster, freeing them to focus on relationships rather than paperwork. In engineering Salesforce itself deploys Claude Code for internal workflows. Anthropic, in turn, runs its own operations on Slack and Agentforce, using the models to summarize threads and trigger actions.
The broader market context makes Benioff’s comments even more striking. Anthropic’s annualized revenue run rate climbed from about $9 billion at the end of 2025 to approximately $30 billion by the end of March 2026. Enterprise adoption of Claude for coding, legal work, financial services and general reasoning drives much of that growth. The company also formed a $1.5 billion joint venture with Blackstone, Hellman & Friedman and Goldman Sachs to embed Claude inside portfolio companies of the world’s largest private equity firms. Token consumption is moving from experimental line item to structural operating cost.
Wall Street has taken notice, sometimes nervously. Software stocks swung wildly earlier this year amid fears that AI agents might displace traditional SaaS offerings. Yet partnerships like the one between Salesforce and Anthropic helped stabilize sentiment. When Anthropic showcased enterprise agent capabilities tied to Slack and other tools, shares of Salesforce and several peers rebounded. Benioff himself pushed back against bears who saw his company with its back against the wall. Customers, he argued, were not replacing Salesforce products with pure AI. They were augmenting them.
Still the $300 million figure raises questions about sustainability and margins. If one large enterprise commits at this level, others may follow. The economics could shift how vendors price AI features and how customers budget for them. Routing intelligence becomes critical. Without it token bills could balloon uncontrollably. With it companies gain the ability to match task complexity to model power and cost. Salesforce appears determined to control that intelligence layer.
Benioff’s vision extends further. Humans, agents and headless platforms now interoperate in ways that let his teams build and sell simultaneously. The pace feels new. The risks feel real. Data governance, model reliability and regulatory compliance remain front and center, especially in the industries the expanded partnership targets. Yet the trajectory looks set. Coding agents will not remain confined to back-end engineering. They are headed straight into the communication and collaboration tools where most knowledge workers spend their days.
Recent coverage reinforces the momentum. Business Insider reported the podcast remarks in detail, highlighting how the spend underscores AI’s transformation into a core operational expense. Analysts note that consumption-based pricing tied to token usage could reshape software economics more broadly. If Benioff’s bet pays off, cheaper and faster development cycles may flow through to customers in the form of more capable products delivered at lower cost.
The partnership did not emerge overnight. Salesforce provided Anthropic with early access to enterprise feedback, customer introductions and rigorous testing in real-world scenarios including data residency requirements in Europe. Anthropic offered models optimized for safety, interpretability and tool use. Constitutional AI, the company’s method for guiding models toward helpful, harmless and honest behavior, aligned well with enterprise demands. That foundation supports the current scale-up.
What comes next will test the thesis. Can routing layers tame the expense? Will coding inside Slack prove as transformative as Benioff predicts? How quickly do other large organizations adopt similar token budgets? For now the numbers speak. A company that once bought 1 percent of Anthropic for around $300 million now stands ready to spend that same amount annually on its output. The transaction has evolved from ownership to usage. And usage, at this volume, changes everything.
WebProNews is an iEntry Publication