Salesforce Decode
Salesforcedecode
Back to questions
AgentforceIntermediatetestingqaregressionuat

Design testing strategy for Agentforce before go-live

Real World Scenario

UAT begins in two weeks for a Agentforce deployment handling password resets, order tracking, and subscription changes. QA team asks what test artifacts you need beyond manual chat scripts.

Expected Answer

• Build golden conversation sets per topic covering happy path, edge cases, and escalation triggers • Automate regression runs comparing expected action invocations and response keywords • Test adversarial inputs: injection, profanity, off-topic, and ambiguous multi-intent messages • Validate action authorization with users at different permission levels and unauthenticated guests • Load test action backends—agent UX fails if order lookup API throttles at 50 concurrent sessions • Include accessibility and multilingual utterance variants in test corpus • Define exit criteria: max misroute rate, action error rate, and p95 response latency

Follow-Up Questions & Answers

Click to expand — each follow-up includes a direct, interview-ready answer

Direct answer: Build golden conversation sets per topic covering happy path, edge cases, and escalation triggers Also consider: Automate regression runs comparing expected action invocations and response keywords In practice: Test adversarial inputs: injection, profanity, off-topic, and ambiguous multi-intent messages Balance speed of delivery with maintainability.

Architect Perspective

Non-deterministic outputs require testing contracts, not exact strings—assert on actions taken, citations present, and guardrail triggers. Invest in continuous golden-set evaluation; manual chat scripts do not scale past launch week.