Salesforce CEO Marc Benioff published production results from Agentforce deployments on April 20, citing two enterprise customers as evidence that agentic AI validates rather than threatens the SaaS model. At Pearson, Agentforce autonomously handles order statuses, refunds, and lost access codes, delivering a 40% increase in customer queries resolved without human intervention, according to Pearson VP Gabriele Bauman as reported by Benzinga. At PenFed Credit Union, the platform handles password resets and account unlocks, cutting total IT tickets by 40%, according to EVP & CIO Shree Reddy.
The Numbers Behind the Narrative
The production wins are real but narrow. Both use cases, customer query resolution and IT ticket deflection, are the most straightforward automation targets: high-volume, low-complexity, well-defined resolution paths. The Decoder reported that jewelry maker Pandora found Agentforce “can’t deliver reliable recommendations when customer requests are vague,” suggesting the 40% gains don’t extend to ambiguous workflows.
Adoption tells a more complex story. Only 23,000 of Salesforce’s 150,000 customers have deployed Agentforce since its late-2024 launch, a 15% adoption rate after 18 months in market. The company logged 2.4 billion “Agentic Work Units” (AWUs) last quarter, up 57%, but the metric was disclosed for the first time, making external comparison impossible.
The Bear Case Benioff Is Fighting
CRM stock is down 28% year-to-date. The Wall Street thesis, as Benioff described it in a Wall Street Journal interview, is two-pronged: AI agents reduce headcount, which shrinks per-seat revenue, and companies “vibe code” their own solutions instead of paying enterprise SaaS pricing. Benioff called this the “SaaSpocalypse” theory and argued it gets things backwards, claiming AI makes Salesforce more valuable because homegrown solutions lack data security and compliance.
His response is “Agent Albert,” a new AI product codenamed for release by year-end that will analyze users automatically and take autonomous actions. Benioff described Agentforce results as “Agenticware, not Software,” framing the company’s future around agent-delivered outcomes rather than tool access.
Production Metrics vs. Platform Bet
The 40% figures matter because they’re among the first named-customer, specific-metric disclosures in the enterprise agent space. Most vendors have shipped “AI agent” features without publishing resolution rates or ticket deflection numbers from identified customers. Pearson and PenFed provide concrete benchmarks, even if the use cases are routine.
The question is whether these gains scale to the complex, ambiguous workflows where enterprise value actually concentrates. A 40% improvement in password reset deflection is cost savings. A 40% improvement in sales forecasting or supply chain optimization would be transformational. Agentforce hasn’t demonstrated the latter yet, and at 15% adoption after 18 months, most of Salesforce’s own customer base hasn’t bet that it will.