When should you think about rate limiting?

So, I've been giving a lot of thought to system resilience recently. We all know about load balancers for traffic spikes and queues for async tasks, but what about controlling the *rate* of requests? When in your system's lifecycle does implementing rate limiting start to make sense? Is it just about preventing abuse, or are there other scaling or performance benefits to consider early on? I'm really curious to hear about your experiences and thought processes on this.