A major one for me is preventing duplicate records.

If the client POSTs a new object to insert it into the database; if there is a connection failure and the client does not receive a success response from the server, the client cannot know whether the record was inserted or not without making an expensive and cumbersome additional read call to check... The client cannot simply assume that the insertion did not happen purely on the basis that they did not receive a success response. It could very well be that the insertion succeeded but the connection failed shortly after so response was not received. If the IDs are auto-incremented on the server and the client posts the same object again without any ID on it, the server will create a duplicate record in the database table (same object with a different ID).

On the other hand, if the client generates a UUID for the object it wants to create on the front-end, then it can safely resend that exact object any number of times and there is no risk of double-insertion; the object will be rejected the second time and you can show the user a meaningful error "Record was already created" instead of creating two of the same resource; leading to potential bugs and confusion.