Isn't this relatable to how accuracy is a "bad metric" vs sensitivity/specificity?

If one uses a single method that outputs a single value for "good/bad", it will always miss some very important aspect of the problem vs more complex, multi-valued algorithms that can capture its many facets (at the cost of being harder to interpret)?

Putting it in another way, isn't trying to have a silver bullet method that automates the analysis "a step to far" that will most likely lead to bad analysis?