Architecting Multi-Agent Systems with Salesforce Agentforce

In this session, I'll share how we architected and coded a multi-agent ecosystem where Salesforce Agentforce seamlessly collaborates with an external Google Gemini agent to automate complex enterprise workflows that span CRM, data, and AI decisioning. The core problem was cross-platform orchestration—how to make Salesforce-native AI agents talk to external reasoning models like Gemini in a secure, scalable, and context-aware way. We solved that by leveraging the Agent to Agent protocol protocol to broker communication between the Salesforce trust boundary and external AI reasoning services. Attendees will see how we: • Designed the Agent Interaction Layer that allows Salesforce agents to call, query, and interpret responses from Gemini. • Used Salesforce Flow Orchestration and MuleSoft APIs to keep context synchronized between the two agent systems. • Implemented guardrails for governance and safety, ensuring agents act within Salesforce's security model while still leveraging Gemini's generative power. • Built a working demo of Agentforce and Gemini co-owning a workflow, where Gemini analyzes unstructured customer data and Agentforce executes Salesforce actions based on the insights.