Instructing agents to use specific test techniques before shipping

Rather than letting an agent write whatever tests it wants, developers give it explicit instructions to apply particular test or verification techniques and libraries known to catch real defects. This is done because agents left to their own defaults tend to write shallow tests that pass while real bugs slip through. The technique treats test guidance as an input to the agent, not an afterthought.

Sources:

• https://dev.to/infoinlet1/how-to-stop-ai-from-confidently-shipping-broken-code-a-pattern-that-actually-works-2gn7
• https://danluu.com/agentic-testing/