What do you prototype with, and how do you decide the fidelity?

Match fidelity to the question. Paper or rough frames answer whether the concept makes sense. A clickable Figma prototype answers whether a flow is understandable. Coded prototypes are worth it for timing, gestures, performance, or data-heavy interactions where a fake will mislead you. Prototype the smallest slice that answers the question, and throw it away without sentiment.

Fidelity follows the question. If I am still deciding what the thing is, I want the ugliest possible artifact, because polish shuts people up. If I put a beautifully finished screen in front of a stakeholder I get compliments; if I show a grey box sketch I get the real objection, which is what I actually need. Once the concept is settled, a clickable Figma prototype of the main path is usually enough to test comprehension, and I only wire the routes the tasks require rather than every button. Where I go further is anything where feel is the point. Drag and drop, an animation that has to communicate what moved where, or a list that has to stay responsive with ten thousand rows. Those lie in Figma. On one project I prototyped a filtering interaction in code with real data, and it was immediately obvious that our design assumed instant results while the query took nearly two seconds, which changed the design entirely. And I delete them afterwards. A prototype is an argument, not an asset.