“We know more than we can tell.” Michael Polanyi, writing in 1966, called this tacit knowledge.
You recognize a face but can't explain how. You ride a bike but can't describe the physics that allow you to do so. You know something is off without needing to articulate what.
Tacit knowledge is embodied, contextual, acquired through practice, and lost when the practitioner leaves. It's the pattern recognition that comes from ten thousand repetitions. In a word, we might call this “instinct”.
The problem as it connects to our domain is that automation demands explicit rules. If you can't write it down, it can't be coded. This forces a choice; either replace the human entirely and lose the tacit knowledge, or leave them doing everything manually and waste their bandwidth on assembly.
There's a third option. We handle the assembly and surface the full picture so that users can apply the full depth of their judgment without need for articulation. The reallocation of working memory and freeing up of bandwidth to use it is what changes.