Build and Test Observability Tools
Build and Test tools let you observe agent behavior during development, before production traffic exists. Testing in Agentforce Builder enables you to iterate quickly on individual conversations, while Testing Center automates testing at scale.
Required Editions
| Available in: Lightning Experience |
| Available in: Enterprise, Performance, Unlimited, and Developer Editions. Required add-on licenses vary by agent type. |
Build & Test tools let you observe agent behavior during development, before production traffic exists. Testing in Builder enables you to iterate quickly on individual conversations, while Testing Center automates testing at scale and across a range of inputs.
Agentforce Builder
Agentforce Builder is where you develop and iterate on your agent before it's ready for broader testing or deployment. Preview lets you run test conversations in real time and observe exactly how the agent responds. You can preview conversations in simulate or live test mode:
- Simulate mimics basic agent conversational behavior without writing or modifying data. Use this to validate quick adjustments without touching live records.
- Live Test executes actions against real data end-to-end. Use this when you need to confirm that actions fire correctly and records update as expected. This mode gives you the most complete picture of your agent's actual behavior.
For each preview conversation, Builder gives you several tools to investigate what happened. Results are session-level and immediate including subagent routing, action execution, and response quality for the conversation under test. Start with the Interaction Summary, then go deeper as needed.
Interaction Summary
The starting point for any preview investigation. Available for each conversational turn during the conversation — you don't need to wait until it's complete. It provides a high-level overview of the steps your agent took to generate a response, including the result of each. Hover over any step for more details. It also includes an AI-generated explanation of your agent's process, which helps put the steps in context. If you're not sure what a result means, Agentforce can explain your agent's behavior, suggest next steps, or make agent fixes. When something looks wrong here, the Trace tab goes into more detail.
Trace tab
Shows full trace data for that step including each agent event, time elapsed, and technical details. Drill into any agent event to open a JSON view of the agent's full LLM prompt, including system prompts.
To preview and test your agent in Agentforce Builder, see Agentforce Builder Testing.
Testing Center & Scorers
Testing Center is where the pre-deployment side of the observability loop takes shape. It's where you confirm your agent performs at scale before customers are involved, establish the quality and performance baselines you'll compare production signals against, and define the scorers that carry forward into Analytics after deployment, closing the loop between pre- and post-deployment data signals.
In Testing Center you can test subagent behavior, action execution, knowledge retrieval, prompt templates, and custom criteria specific to your requirements. The test results surface response quality, subagent and action correctness, instruction adherence, and latency, all pre-deployment, per test case.
Scorers
Scorers are the evaluation criteria Testing Center uses to assess agent output. Many map directly to the metrics you'll monitor in Agent Analytics, and custom scorer configurations carry forward as production KPIs after deployment.
| Tier | Scorers | What it evaluates |
|---|---|---|
| Default | Response Scorer, Subagent Assertion, Action Assertion | Whether the right subagent was selected, the right action fired, and the response was acceptable. |
| Quality | Completeness, Coherence, Conciseness, Latency, Instruction Adherence | How good the response was, not just whether it passed. These include coherence, completeness, conciseness, latency, and instruction adherence as distinct dimensions. |
| Custom | Criteria your deployment requires that default scorers can't cover like brand voice, domain accuracy, tone, or use-case-specific compliance. | |
| User Defined Labels | Returns a value from a list you define. Use for classification, enum-style outputs, or compliance labels like brand voice pass/fail, approved response categories, and similar. | |
| User Defined Open | Returns unconstrained output, text, a number, a URL. Use when you need richer or exploratory results that a fixed list can't capture. | |
| Prompt template | Coherence, Conciseness, Completeness, Factuality, Instruction Following, Response Match | Output quality for prompt template test runs including factuality, response matching, and the standard quality dimensions applied to template outputs. |
To set up and run tests in Testing Center, see Set Up and Run Tests in Agentforce Studio (Beta).
