Different testing tools answer different questions. Builder preview accelerates focused iteration, traces explain a particular execution, and batch testing measures behavior across a defined suite.

A mature workflow uses all three instead of treating one successful conversation as proof that the agent is ready.

Agent development lifecycle from ideation through observation on a trust, governance, and security foundation

Match the tool to the question

  • Use preview for fast exploration of instructions, routing, variables, and individual actions.
  • Use tracing when the visible answer does not explain where a session went wrong.
  • Use batch suites to measure regressions and compare pass rates across many cases.
  • Use explicit evaluation criteria so success is based on business expectations, not reviewer intuition.

Create an efficient test workflow

  1. Debug a small representative case in preview before expanding the batch suite.
  2. Record the expected subagent, actions, and response properties for each case.
  3. Review failures by severity and cluster them by responsible layer.
  4. Promote fixed failures into a versioned regression suite that runs before release.
Further reading

Resources

This article is based on the following Salesforce learning and product documentation.

Continue the architecture path
ObservabilitySet Up Testing Criteria

Define expected subagents, actions, and response-level evaluation criteria.

ObservabilityView Testing Center Results

Read subagent, action, and response evaluation pass rates.

ObservabilityAgentforce Session Tracing: Debug Actions and Errors

Read Agentforce session traces, understand the session data model, and diagnose routing, grounding, and action failures with a worked example.