The first department AI can't replace is the one most people thought would go first.
The Summary
- Salesforce CEO Marc Benioff revealed the company's 15,000-person engineering team has been "mostly flat" for two years thanks to AI coding agents, while sales headcount is growing.
- The $145 billion firm already cut 4,000 support roles that AI agents now handle, but admits agents "are not exactly" capable of selling and communicating.
- Salesforce is projected to spend $300 million on Anthropic tokens this year, making it one of the largest enterprise AI deployments on record.
The Signal
Salesforce keeps 15,000 engineers on payroll, but hasn't grown that team in 24 months. During the company's quarterly earnings call, Benioff credited "new coding agents" with "dramatic capabilities" for holding that line. Meanwhile, the sales organization under president Miguel Milano is expanding. The contrast is stark: build with agents, sell with humans.
This isn't theory. Salesforce already eliminated roughly 4,000 support roles last fall, routing those tasks to AI agents. The company is betting $300 million on Anthropic tokens to power internal operations. That's enterprise commitment, not a pilot program.
"The one thing that we are doing here with you, selling and communicating, the agents are not exactly doing that."
Why sales survives:
- Technical execution scales with code. Relationship building doesn't.
- Support tickets have clear inputs and outputs. Sales conversations navigate ambiguity, objections, politics.
- AI can draft the contract. It can't read the room when a VP goes quiet or know when to shut up.
Benioff's framework is simple: agents handle deterministic work, humans handle the legibility problem. Engineering increasingly fits the former category. Complex, yes. But bounded. Sales remains firmly in the latter. The customer's actual need versus stated need, the org chart politics, the timing of the close, these run on human context that doesn't serialize cleanly into prompts.
The spending figure matters. $300 million in Anthropic tokens puts Salesforce in the top tier of enterprise AI deployment. That capital is replacing labor in support, augmenting productivity in engineering, and apparently doing neither in sales. The company is running the experiment at scale and publishing the results in real time through hiring patterns.
The Implication
If you're early in your career, Benioff just told you where to develop skills AI can't commoditize yet. Not in prompt engineering. In reading people, navigating organizational complexity, and closing deals when the relationship is the product. Sales here is shorthand for any role where illegible human dynamics drive outcomes.
For companies, the playbook is visible. Flat engineering headcount with rising output signals working AI leverage. But growth still requires humans who can operate in high-context, high-stakes conversations. The ratio is shifting. The need for the latter isn't.