Marc Benioff doesn’t think AI is going to replace Salesforce’s engineers. He thinks it’s going to turn them into managers. Speaking on The Future Live with Matthew Berman, the Salesforce CEO described a shift already underway inside the company—one where its 15,000 engineers work alongside AI coding agents from Anthropic, OpenAI Codex, and Cursor, and increasingly oversee those agents rather than doing the work themselves. “They can even become somewhat supervisory over these agents,” Benioff said. “But still those engineers are needed.” His argument for why comes down to a single line: “The model still cannot operate autonomously. We’re not at that level yet.”As evidence, he pointed to the job boards of the biggest AI companies in the world—all of them, he noted, are still hiring engineers at scale.
Why Salesforce froze engineer hiring in 2025 and let AI agents do the work
The productivity gains, though, have had real consequences for headcount. In a February 2026 appearance on TBPN, Benioff confirmed that Salesforce didn’t hire any new engineers throughout fiscal year 2026—using the capacity freed up by coding agents to avoid adding headcount altogether. The company did cut around 1,000 roles earlier in the restructuring. But where it hired aggressively was in sales: nearly 20% more salespeople, to keep pace with what Benioff described as demand coming in from every segment of the market.
Salesforce says AI is now handling up to 50% of company workload
Pull back further, and the engineering story is one part of something much larger. In a June 2025 Bloomberg interview, Benioff said AI now accounts for 30% to 50% of Salesforce’s total workload, at an accuracy rate the company puts at around 93%. He’s framing the whole era as a “digital labor revolution”—with Salesforce as its primary vendor. “No company is better positioned than Salesforce to lead customers through the digital labor revolution,” he said.For engineers, the pitch is that their role isn’t disappearing—it’s moving up a level. Whether that framing survives the next wave of model capability is the part nobody can answer yet.
