It would be wonderful if the current per-Assistant Custom Instructions, and Custom Assistants themselves, were orthogonal, i.e., you could maintain them independently of one another and mix-and-match a given Custom Instruction (or "Personality") with a given Custom Assistant.
This would allow creating a number of personalities (or domain experts) via different Custom Instructions across various domains and re-using them on different models.
The primary affects on workflows / user-experiences would be:
Settings would show "Personalities" as its own section with the ability to add/remove/edit various system prompts.
The assistant prompt would have two select widgets: one for model, and one for "Personality" (the particular custom instruction selected).
A user would:
Go into Assistant, go to Personalities, and create one or more new Personalities (gourmet chef, software engineer, financial advisor, creative writer) via appropriate system prompts.
In the Assistant search interface, the user would select one of these Personalities independently of the model they select.
