Testing Center limits and considerations
Key Takeaway
Run Testing Center only in a sandbox, because testing an agent can modify CRM data.
Agentforce Testing Center runs large batches of test cases against an AI agent. It checks the agent's responses, its subagent and action handling, and its response quality, so you can find issues early. To use it well, you need to know where to run it, how its usage is metered, and which features conversation-level testing doesn't support yet.
What Testing Center evaluates
Testing Center runs batch tests across many scenarios at once. AI agents are non-deterministic, so large-scale testing surfaces patterns and edge cases that smaller-scale testing often misses.
Testing Center evaluations include these checks:
- Response accuracy
- Subagent and action validation
- Quality metrics, such as:
- Completeness
- Coherence
- Conciseness
- Latency
- Instruction adherence
Test cases can come from three places:
- You can generate them from the agent's core subagents and actions.
- You can create them as Q&A style tests from knowledge content.
- You can upload them as custom scenarios.