Salesforce AI Research is set to launch eVerse, an enterprise simulation environment designed for training voice and text agents through synthetic data generation, stress-testing, and reinforcement learning.
eVerse is designed to address "jagged intelligence": the situation where AI systems perform exceptionally well on complex tasks but struggle with simpler ones, posing significant business risks. Tackling this reliability issue will enhance steps towards Agentic Enterprise—a workplace where human employees and AI agents can effectively collaborate in a cohesive ecosystem. The platform builds trust in AI outputs through an ongoing process of synthesizing data, evaluating performance, and refining agent behavior.
eVerse has been used to power the development of Agentforce Voice, enabling it to handle unpredictable customer interaction scenarios: spotty phone connections, background noise and diverse accents. With help lines still a go-to channel for many, enterprise-grade reliability is key.
The AI agents are trained and stress-tested in eVerse’s virtual environments that mirror the noise, accents, crosstalk, and complexity of real-world operations. Before launching, Agentforce Voice was put through thousands of conversations simulated through eVerse, allowing teams to identify and fix errors early and deliver enterprise-grade reliability to best serve customers.
Madhav Thattai, COO of Salesforce AI, said: “With eVerse, we were able to explore many nuances of human conversation before Agentforce Voice reached production. This type of rigor is what turns breakthrough research into scalable products and dependable customer experiences. It’s how we’re expanding that same level of responsiveness and consistency across the full observability stack to solve our customers’ most complex needs.”
eVerse is currently in pilot testing with customers such as UCSF Health, where the team is partnering with clinical experts to train and refine AI agents that simplify healthcare billing.
Industry data shows that only 60–70% of healthcare contact center inquiries are routine and can be fully handled by AI agents, as much of the knowledge isn’t formally documented. To increase coverage, eVerse optimizes accurate AI agent behavior and adapts to complex scenarios over time with reinforcement learning from human feedback (RLHF), achieving up to 88% coverage across routine and complex tasks according to preliminary evaluations.
Sara Murray, MD, MAS, VP & Chief Health AI Officer at UCSF Health, said: “When used responsibly, we believe AI can help our teams simplify one of the most complex parts of healthcare, creating a billing experience that feels seamless and truly patient-centered.”
Silvio Savarese, Chief AI Scientist at Salesforce, said: “Our partnership with UCSF Health demonstrates how applied science translates directly to customer value, proving that when you train agents in environments that mirror real-world complexity, they perform reliably when it matters most.”