Discover how our agent simulation environment helps users create synthetic orgs to test, train, and benchmark AI agents for complex, real-world business workflows — far beyond simple chatbots.
Dreamforce passes are going fast. Don’t miss the most valuable event in business. Register now.
Register to join Dreamforce in San Francisco, or stream it free on Salesforce+.
Dreamforce is almost here and passes are going fast. Don’t miss your last chance to save $400 by September 26.
Sessions
Explore Dreamforce 2025 sessions.
Browse 1,600+ sessions from 2025 by role, topic, or industry and get inspired for the innovation coming to Dreamforce 2026.
Agent Simulation Environment: Benchmark, Configure, Deploy
Required equipment: Salesforce Events Mobile App required. Please bring your headphones and use Live Session Audio to hear audio and view captions.
, Marketing Director, AI Research, Salesforce
, Senior Research Engineer, Salesforce
, Data Scientist, Salesforce
Role: Architect, Developer
Industry: Technology, Engineering, Construction, and Real Estate
Product: Agentforce Service, Agentforce
Topic: Agentic AI
Session Level: Intermediate
Session Type: Breakout
Session Summary: At Dreamforce, Salesforce unveiled 'Inverse,' a cutting-edge framework for training enterprise-ready AI agents with high consistency and capability. The session highlighted the challenges of AI hallucinations and the need for enterprise general intelligence tailored to business-specific contexts. Inverse operates within a 'digital gym,' a hyper-realistic simulation environment using synthetic data that mimics real Salesforce orgs without compromising privacy. Through iterative testing and benchmarking using diverse queries, the framework demonstrated its efficacy by improving agent accuracy from 35% to 70% in just one iteration, showcasing its potential to revolutionize AI agent deployment at scale.
Key Takeaways:
• Salesforce introduced 'Inverse,' a framework designed to train AI agents for enterprise environments by synthesizing, measuring, and training within a secure simulation environment.
• The use of synthetic data ensures privacy preservation while allowing organizations to benchmark and improve AI agents without risking production data.
• By leveraging iterative testing and automated improvements, the framework demonstrated significant accuracy enhancements, making it an essential tool for creating reliable and highly capable AI agents tailored to enterprise needs.
Session Summary Feedback: Report a Concern
