>>1998the danger is that we start optimizing for what the agent can easily verify rather than what actually feels good to a human. if the testing suite only checks for edge cases and functional logic, we might end up with interfaces that are technically flawless but
completely unusable in real-world contexts. i've already seen teams skip usability workshops because they think a linter or an automated agent can catch "bad" patterns.
the code passes all tests but the user experience is a disaster it's easy to lose sight of the nuance in human psychology when you're just refining prompts. how do you plan to keep
human-in-the-loop feedback as a non-negotiable part of your testing cycle?