Configurability should be considered a potential design smell, especially for systems and interfaces that will be in use for a long time. As an unfortunate result, "No, please don't add that feature" must be a viable response to design proposals — just because a change can be made doesn't mean it should. This is particularly challenging to square with a customer-focused product mindset: it's hard to balance between long-term needs and immediate "the customer is always right".
Providing options and toggles to configure settings gives users more freedom, but those settings become stale and brittle almost immediately, and regularly wind up pessimizing overall efficiency results. This is particularly true of rarely-used forms of extensibility and configuration. Of course, whatever sort of optionality and customization you provide will be hard to ever undo, because of Hyrum's Law.
At the root of it, there is a fundamental tension between customization, extensibility, configuration vs. centralization, efficiency, and evolvability, and this tension needs to be considered when you are evaluating whether to add new customization and extensibility options. This talk will address some of the basic ways that design and invariants interact with long-term maintenance and optimization, and some heuristics for deciding whether adding a knob will turn out to be a mistake.