Section 6 · Governance and Observability · 10% of exam
6.1 Managing and Monitoring Agents
Exam objective
Explain how administrators manage an agent through its lifecycle and monitor its behavior, health, trust, and access after deployment.
Governance continues after activation
Managing an agent is an ongoing lifecycle, not a one-time build task. A responsible owner should know the agent’s purpose, audience, supported channels, active version, data sources, actions, permissions, and escalation path. Define what successful operation means before launch: for example, the agent should resolve approved requests, select the expected actions, respect identity controls, and escalate issues it cannot handle. Assign people to review results and decide who can change instructions, actions, model settings, and access. Keep an approved change process so production edits are reviewed, tested, and attributable to a version.
Turn on the right observability
Salesforce observability tools collect different kinds of evidence. Agentforce Session Tracing records sessions and their processing steps when enabled. Agent Analytics aggregates performance signals across sessions. Agent Optimization lets administrators inspect unresolved or lower-quality sessions and analyze intents. Builder Interaction Summary and Trace help investigate a single interaction, while Agent Health can alert when configured thresholds are crossed. These tools serve different needs: aggregate monitoring highlights trends, and traces explain what happened in one session. Check that the necessary features are enabled and available for the org and agent type before relying on a dashboard.
Protect collected data
Session telemetry can contain user messages, generated responses, retrieved context, and other sensitive values. Treat it as governed business data. Restrict access to tracing data and the Data 360 data space to people with a legitimate need. Review retention and masking requirements, especially when data may include personal, health, or payment information. Monitoring access should be part of the security design, not an afterthought. Salesforce notes that Session Tracing data can contain sensitive information and recommends permissions and data masking controls. A monitoring tool should not become a broader access path than the agent itself.
Use a consistent operating rhythm
Review high-level health and outcome trends on a regular schedule, then investigate exceptions, material changes, or user feedback. Check whether usage, errors, task resolution, escalation, and trust signals remain within the expected range. Confirm that the active version and dependencies match the approved release. When a problem appears, preserve the relevant session evidence, classify the issue, and assign an owner. After a fix, validate it in a controlled environment and document the result. Good management combines technical monitoring with clear ownership, user feedback, privacy controls, and a repeatable process for changing the agent.
Separate aggregate monitoring from session investigation. Use Analytics to find trends, then Optimization or Trace to see what happened in a specific interaction.
Session tracing may capture sensitive prompts, responses, and other user data. Limit access to the trace data and apply the organization’s data governance requirements.
Extra concept: Observability is useful only when signals have owners and thresholds that trigger a review or action.
Official references:
- Salesforce Certified Agentforce Specialist Exam Guide https://help.salesforce.com/s/articleView?id=005298924&language=en_US&type=1
- Understand Agentforce Observability KPIs and Tools https://help.salesforce.com/s/articleView?id=ai.agentforce_understand_observability_kpis_and_tools.htm&language=en_US&type =5
- Set Up Agentforce Session Tracing https://help.salesforce.com/s/articleView?id=ai.generative_ai_session_trace_setup.htm&language=en_US&type=5
Key takeaways
- Assign owners for the agent, its active version, permissions, actions, and monitoring signals.
- Use Analytics for trends and session-level traces to investigate individual interactions.
- Protect observability data because it can include sensitive user and agent content.
