1) Problem Domain Knowledge: This is what people generally mean when they say "domain expertise". This has always been and always will be the moat with/without AI. Simply because this is what understanding and modeling a problem is all about. It abstracts the key concepts/ideas and their relationships in the problem domain and builds a coherent model. This model embodies a set of functionalities with bounded scope and clear assumptions.
2) Solution Domain Knowledge: This is the implementation domain for the above problem. The model arrived at above gives the requirements which must be mapped to concepts/ideas and their relationships in the solution domain. When our implementation domain is a computer system, this takes the form of architecture, algorithms and data structures. The probability of a good solution here is directly proportional to how good a model we were able to construct in the problem domain above.
Albert Einstein;
"The mere formulation of a problem is far more essential than its solution, which may be merely a matter of mathematical or experimental skills."
"If I had an hour to solve a problem, I'd spend fifty-five minutes thinking about the problem and five minutes thinking about solutions."