A New Labour Model: Salesforce to Spend $300M on Anthropic AI Tokens

Moving past traditional engineering headcount, CEO Marc Benioff embraces a 'digital labour revolution' where human developers supervise autonomous coding agents.

The structural blueprint of tech enterprise labor is experiencing a seismic realignment. Speaking on a recent episode of the All-In podcast, Salesforce CEO Marc Benioff revealed that the customer relationship management (CRM) giant is on track to spend approximately US$300 million on Anthropic AI tokens, with the overwhelming majority of that computing budget directed toward software development and coding applications.

The massive financial commitment highlights an operational pivot. Rather than pouring capital into expanding its software engineering headcount, Salesforce is treating raw AI model usage as a core utility line item, scaling up computational power rather than human desk space.

Benioff described the emerging paradigm as a triple threat: "A new labour model, new productivity model, and a new economic model."

From the Engineering Freeze to the Supervisory Era

This development marks the culmination of a strategy Salesforce initiated over a year ago. In an effort to optimize margins, the company froze the hiring of new software engineers, pointing to immediate, massive efficiency returns from its internal AI integrations.

"We're not adding any more software engineers... because we have increased the productivity... with Agentforce and with other AI technology that we're using for engineering teams by more than 30%," Benioff previously noted.

However, Benioff clarified that this $300 million token spend does not imply the immediate extinction of human programmers. Salesforce continues to maintain its backbone of roughly 15,000 engineers, but their day-to-day functions have been fundamentally altered.

Instead of writing boilerplate code or manually refactoring legacy systems from scratch, human developers utilize external frontier intelligence—such as Anthropic’s Claude models, OpenAI Codex, and Cursor—to execute tasks.

"When they start to use these models, they're now working not only with the AI, but agents to help them code—and they can even become somewhat supervisory over these agents," Benioff explained. "But still, those engineers are needed. The model still cannot operate autonomously. We're not at that level yet of AI."

Why Tokens Over Talent?

In the mechanics of large language models (LLMs), tokens are the baseline units of data processed to comprehend prompts and render output. By allocating $300 million directly to metered token consumption, Salesforce is treating code generation like a variable utility expense—akin to cloud hosting or electricity—rather than a fixed labor asset.

The economic rationale is hard to ignore. AI coding agents drastically lower software development costs, compress prototyping timelines, and eliminate human bottlenecks in QA testing and debugging. To keep these staggering token bills manageable, Salesforce is actively developing internal routing architectures. These systems automatically gauge the difficulty of an engineering request and seamlessly toggle the task between massive flagship models and smaller, cost-effective LLMs based entirely on structural complexity.

Reallocating the Corporate Budget

While technical hiring remains stagnant, Salesforce is reallocating those saved capital reserves back into human-facing roles. The company has aggressively ramped up its sales divisions, onboarding thousands of specialized sales professionals tasked with communicating the complex enterprise value of its booming AI ecosystems.

The strategy appears to be paying off handsomely. Salesforce’s standalone AI venture, Agentforce, has rocketed to an estimated $800 million in annual recurring revenue (ARR). Furthermore, the company reports that artificial intelligence now autonomously handles between 30% and 50% of Salesforce’s global corporate workload.

Salesforce's move sets a clear template for the broader tech industry. The race is no longer about which corporation can hoard the largest roster of engineering talent, but rather which organization can build the most secure, deeply integrated workflow framework to guide autonomous software agents