Empiric antimicrobial therapy is typically broad-spectrum, in that it treats both a multitude of either Gram-positive and/or Gram-negative bacteria, diverse fungi or parasites respectively.
When more information is known (as from a blood culture), treatment may be changed to a narrow-spectrum antimicrobial which more specifically targets the bacterium or fungus known to be causing disease.
However, the delay and expense that would be required to perform definitive species identification in every single clinical case are not affordable, so some degree of trade-off is accepted on the principle of the benefits outweighing the risk.
The fact that "acting on practical experience in the absence of theory or complete knowledge" can have both legitimate and illegitimate forms stretches back to long before science existed.
In the example of bloodletting to correct excess water, the fact that fluid balance is a legitimate physiologic concern didn't mean that the then-state-of-the-art "understanding" of causation was well founded overall.
Thus, in evidence-based medicine, the goal is that every clinician will make decisions for every patient with total mastery and critical analysis of the entire scientific literature at their fingertips.
This is a formidably vast goal to implement operationally (because it is not even possible for one person to master all extant biomedical knowledge on the basis of individual education[5]), but development of health information technology such as expert systems and other artificial intelligence in medicine is underway in pursuit of it.