The Rise of ‘Agentic AI’: How Salesforce is Building Voice Agents That Can Handle Anything You Throw At Them
70% of all customer interactions at businesses will happen via voice. That’s a staggering statistic, and it highlights a critical vulnerability: even the most sophisticated voice AI can crumble under the pressures of real-world conversations – spotty connections, distracted callers, and complex requests. But what if we could proactively train AI to handle those very scenarios? Salesforce is betting on a new approach, and it could redefine the future of customer service and beyond.
From Handwriting Recognition Failures to Voice AI Triumph
The story begins with a simple observation. Years ago, the dream of accurate handwriting recognition stalled because human handwriting is simply too variable. As John, a chief scientist at one company, pointed out, you’d have to “reteach people how to write” to make it work. Voice, however, proved more malleable. Tools like Dragon now boast up to 99% accuracy with clear audio. But that accuracy plummets when conditions aren’t ideal. This is where Salesforce’s eVerse comes in.
eVerse: Simulating the Real World for AI Agents
Unveiled at a recent Salesforce AI Research roundtable, eVerse is a simulation framework designed to train voice and text agents using synthetic data generation and reinforcement learning. Essentially, it allows Salesforce to create a digital proving ground where AI agents can encounter – and learn to overcome – the challenges of real-world interactions. Think spotty connections, frustrated customers, and rambling explanations. This isn’t about perfect speech recognition; it’s about agentic AI – AI that can proactively solve problems and adapt to unpredictable situations.
Human-AI Augmentation: The Power of Collaboration
A particularly compelling demonstration involved healthcare billing, led by Dr. Sara Murray, chief health AI officer at UCSF. Simple inquiries, like co-pay status, can be handled entirely by AI. But complex questions – “What will my out-of-pocket cost be with multiple insurance plans?” – still require human expertise. eVerse facilitates a seamless handoff, allowing human agents to assist and, crucially, to use no-code tools to train the AI, improving its performance over time. This aligns with the growing consensus, championed by authors like David De Cremer in “The AI Savvy Leader,” that the future lies in human-AI augmentation, with domain experts leading the training process.
The Flywheel of Continuous Improvement
Salesforce envisions a continuous cycle of improvement, a “flywheel” encompassing design, build, deployment, and refinement. Madhav Thattai, Salesforce’s COO, emphasizes the need for real-time feedback and workflow optimization. eVerse enables this by creating realistic “personae” – simulated customers with specific problems – and an enterprise simulation environment for rigorous stress testing. The framework focuses on three key pillars: synthesizing data, measuring performance, and optimizing agent behavior. This isn’t a one-and-done solution; it’s a commitment to ongoing learning and adaptation.
Building Trust Through Reliability
According to Jayesh Govindarajan, Salesforce’s executive vice president of AI engineering, eVerse allowed them to “test every nuance of human conversation” before launching Agentforce Voice, a component of Agentforce 360. This rigorous testing is crucial for building trust and delivering the responsiveness and consistency enterprises demand. The goal isn’t just to create AI that *can* work, but AI that customers can *rely* on.
Beyond Today: The Path to Enterprise General Intelligence
Salesforce views this progress as a stepping stone toward “enterprise general intelligence” – a point where AI equals or surpasses human capability. As Bob Muglia, former CEO of Snowflake, astutely observed, such AI won’t be subservient; it will be a peer. The development of agentic AI raises profound questions about the future of work and the relationship between humans and machines. Are we, through these advancements, inching closer to that world?
The evolution of voice AI, from the failed promise of handwriting recognition to the sophisticated capabilities of eVerse, demonstrates the power of focusing on adaptability and continuous learning. The companies that prioritize building reliable, resilient, and human-augmented AI agents will be best positioned to thrive in the increasingly voice-driven future. What are your predictions for the future of agentic AI? Share your thoughts in the comments below!
