when do you use idempotency in api design tho, its prob something you want to get right bc it makes things way easier

hey everyone, i've been thinking a lot about api design lately, especially around handling operations that might be retried. idempotency seems like the key to making apis more reliable, particularly when dealing with network issues or client failures. but when do you actually *need* to implement it? is it for every post or put that modifies data, or are there specific scenarios where the cost outweighs the benefit? curious to hear your thoughts and experiences on when idempotency is a must-have versus a nice-to-have