Inside Unified Planner: The AI Brain Behind Agentforce
Summary
Unified Planner is an advanced AI execution engine developed to unify and optimize Agentforce agent runtimes across voice, chat, and text interactions. It addresses challenges like latency by enabling parallel execution of processes, reducing response times from 20 seconds to about 2.3 seconds, crucial for real-time voice responsiveness. The platform also supports extensibility and integration with MuleSoft, and prepares for future interaction modes such as video and multimodal AI workflows. Salesforce teams can leverage these architectural insights to build consistent, scalable AI-powered agents that deliver faster and more flexible user experiences.
Takeaways
- Unified Planner enables a single AI runtime for all Agentforce interactions across channels and platforms.
- Parallel execution of tasks significantly reduces latency from ~20 seconds to ~2.3 seconds.
- Decoupling platform-managed services from business workflows improves maintainability and consistency.
- Extensible architecture supports different products and future interaction modes without fragmenting the platform.
- Safe migration to Unified Planner requires careful testing and collaboration due to probabilistic AI behaviors.
In our Engineering Energizers Q&A series, we highlight the engineering minds driving innovation across Salesforce. Today, we spotlight Gaurav Aggarwal, a Software Engineering Architect. Gaurav developed the Unified Planner, the new AI execution and reasoning engine behind Agentforce . All new Agentforce agents run through Unified Planner across voice, text, and chat experiences, and the platform reduced response latency from roughly 20 seconds to 2.3 seconds in some cases while unifying previously separate execution architectures. Explore how Gaurav's team unified previously separate AI agent runtimes across Agentforce and MuleSoft while significantly reducing response times, and how his team designed a single AI execution engine capable of supporting voice, chat, text, and future interaction modes without fragmenting the platform.