Question 5
Section 3Which Google-recommended practice helps catch QUALITY regressions over time after a gen AI app is in production?
Correct answer: A
Explanation
Google recommends ongoing production monitoring to detect when model behavior changes over time. Tracking KPIs, drift, and periodic evaluations helps surface "quality regressions" after launch by comparing current performance against baseline behavior and flagging degradation early.
Why each option is right or wrong
A. Continuous monitoring with KPIs, drift tracking, and periodic evaluation
Google’s production guidance for gen AI systems emphasizes post-launch evaluation rather than one-time testing: you should continuously monitor business and model KPIs, track data/model drift, and run periodic re-evaluations against a baseline to detect degradation after deployment. In this question’s context, that is the only option that directly addresses quality regressions over time in production, because it creates an ongoing feedback loop to catch performance drops as user inputs and model behavior change.
B. Disabling versioning to keep one model live forever
Keeping one unversioned model reduces traceability and makes regression comparison or rollback much harder.
C. Removing all logs to save cost
Logs are essential for diagnosing failures, auditing outputs, and spotting quality changes over time.
D. Manually upgrading models without tests
Model upgrades should be validated with tests and evaluation, not pushed manually without safeguards.