/
Published on 10.02.2026
TLDR: If you are still manually spot-checking whether your AI application works, you are living on borrowed time. This article from the Decoding AI newsletter lays out a three-pillar framework for AI evals: optimization during development, regression testing in CI, and production monitoring via sampling. It is the first in a seven-part series and focuses on the where, when, and why rather than the how.