AI-Driven Test Execution — quiz

A tester gives an AI agent a charter: 'explore the new gift-card redemption flow for 30 minutes and report risk areas.' Which framing produces the most useful output?

• Tell the agent to click randomly with no goal
• Provide explicit personas, key risks (currency rounding, expired cards, partial redemption), success/failure heuristics, and a structured report format (steps, observations, evidence, hypothesis) — so the AI returns a triable bug list, not a stream of clicks
• Restrict the agent to a single page
• Ask the agent to summarise without producing evidence