A few years ago, when handwriting recognition was all the rage, I asked John, the company’s chief scientist, if handwriting recognition would ever be accurate enough. He immediately answered, “No.” Human handwriting is so diverse that it’s hard to do without retraining how to write (Chinese handwriting is an exception, as it tends to be more uniform).
But he added that eventually the audio will be accurate enough. He was right. While the PalmPilot struggles to reteach us how to write; modern tools Voices like Dragon achieve up to 99% accuracy with clear dictionaries and high-quality voices. But what happens when it is refined? voice technology Is it being applied in less-than-ideal situations: phones with distracted callers, inconsistent voice quality, more complex business processes? This question is important because 70% of agent interactions in the enterprise occur over voice. Issues are often urgent and directly impact the customer experience.
We got a glimpse into the work being done to address the communication gap at the Salesforce AI Research Roundtable. Salesforce AI experts are here to help Itaiseo, Silvio Salvarese and Madhav Thattai We shared information about eVerse, a simulation framework that trains Salesforce voice and text agents through synthetic data generation and reinforcement learning.
EVerse is intended to address the complexities of voice agents, such as unstable connections, distracting behavior, and irritated tone. This situation can significantly degrade the performance of tools such as Dragon.
Dr. Sarah Murray, UCSF Chief Health AI Officer
Empowering agents at work: Medical billing
The most interesting feature demonstrated was the concept of human/virtual agent augmentation. Dr. Sarah Murray, chief medical AI officer at the University of California, San Francisco, demonstrated this concept with a medical billing example. She explained this while asking simple questions like, “Was my copay applied to my bill?”” More complex inquiries that can be fully handled by AI agents; ”“What will my out-of-pocket costs be if I have multiple insurance plans?” still requires human expertise, especially when it requires traversing multiple systems.
With eVerse, people involved in the process can provide agent assistance and no code tools Train agents to improve their natural performance over time. This is consistent with the perspective of authors like David De Cremer, who advocates human-AI augmentation in The AI Savvy Leader and has domain experts train agents rather than IT specialists.
Madhav Thattai, Senior Vice President and COO, Salesforce
Clearly, we can drive improvements to the experiences that matter most. agent AIwhich requires speech and a continuous learning process. Thattai advocated a flywheel that encompasses design and construction, deployment and improvement. Murray emphasized the need to continually improve workflows and get real-time feedback.
To make this work, eVerse enables the creation of personas with any of the above issues and an enterprise simulation environment for training agents through synthetic data generation, stress testing, and reinforcement learning. This framework aims to address these challenges by building trust through reliability and optimizing both agent functionality and consistency through three interconnected steps: data synthesis, performance measurement, and agent behavior optimization.
parting words
This marks another step in the evolution of agent enterprise capabilities.
Jayesh Govindarajan, executive vice president of AI engineering at Salesforce, said: agent force 360. “That rigor is what turns groundbreaking research into trusted customer experiences, and it’s how Agentforce Voice delivers the responsiveness and consistency that businesses expect.”
While this progress is certainly exciting, Salesforce sees it as part of a long journey toward what the company calls enterprise general intelligence, a point where AI can match or exceed human capabilities. Bob Muglia, former CEO of Snowflake, said: “Unlike robots, AGIs are not servants. They are just as smart or smarter than us, and we must show them respect. Perhaps it is best to think of them as our allies.” Are agents leading us into that world? Only time will tell.
