Understanding Sessions & Intents in Salesforce Agentforce

Agentforce Studio’s Sessions & Intents tool provides granular monitoring, auditing, and optimization for AI agent interactions. By leveraging Large Language Models (LLMs) to cluster similar user goals across individual sessions into categorized intents, technical teams can move beyond whole-session summaries.

This intent-level transparency enables Admins, Developers, and Architects to isolate exact points where an agent produced inaccurate, misleading, or off-topic responses—or where configured actions failed to execute as expected.

Step-by-Step: Enabling Agentforce Session Tracing in Salesforce

Step-by-Step Configuration Guide

1. Enable Data Cloud & Assign Permissions
   • Assign the Data Cloud User permission set to yourself and relevant technical users.
   • Navigate to Salesforce Setup and enable Data Cloud if it is not already active.
2. Turn On Einstein Generative AI
   • From Salesforce Setup, enter Einstein Setup in the Quick Find box and select Einstein Setup.
   • Toggle on Turn on Einstein if it is not already active.
3. Enable Agentforce
   • From Salesforce Setup, enter Agentforce Agents in the Quick Find box.
   • Toggle on Agentforce.
4. Enable Agentforce Session Tracing
   • Go to Setup and search for Einstein Audit, Analytics, and Monitoring Setup.
   • Locate and enable Agentforce Session Tracing and Data Model.
   • Ensure Agentforce Optimization is also turned on.
5. Access Agent Optimization
   • Assign the Access Agentforce Optimization Permission Set to yourself and users who need monitoring access.
6. Launch Sessions & Intents
   • Open the Agentforce Studio App to begin auditing and analyzing intent-level interactions.

Technical Best Practices & Recommendations

• Establish a Continuous Audit Cadence: Schedule regular reviews of outlier intents—specifically interactions where fallback responses were triggered or actions failed—to continuously refine prompt instructions.
• Enforce Strict Access Control: Limit the Access Agentforce Optimization permission set exclusively to core technical personas (Architects, Admins, Developers) to safeguard conversational data logs.
• Optimize Action Metadata for Intent Matching: If an intent routinely fails to run an intended action, inspect the action’s description and grounding parameters in Agentforce Builder. LLM routing relies heavily on semantic accuracy in action descriptions.

Note:
Because Agentforce Session Tracing relies on Data 360 pipelines to analyze, transform, and cluster AI conversational logs, the dashboard impacts your credits.
https://help.salesforce.com/s/articleView?id=ai.generative_ai_session_trace_usage_types.htm&type=5

Help Article:
https://help.salesforce.com/s/articleView?id=ai.generative_ai_optimize_setup.htm&type=5

Trailhead:
https://trailhead.salesforce.com/content/learn/modules/agent-analytics-and-monitoring/investigate-agent-sessions-and-intents

Mastering Agentforce Studio: How to Audit & Debug AI Agent Intents