Hey Jacob, it's a good question

Imagine you want to reduce cold start with Cloud Run, let's say up to 3 concurrent instances. You set the min instance to 3. Same thing with App Engine (you also have a min instance parameter).

Now, what happens at the forth instance if you have more traffic than expected? A new instance is created and the cold start impacts your users.

With the idle instance, you can say to App Engine to keep a number of warm instances, whatever the min instance parameter. Let's say, you set this param set to 2.

When all your min instances are used (3, as before), there are still 2 instances warm ready to serve traffic. If a 4th instance is used, there is no cold start, and the system starts another instance to keep it warm, in the idle instance pool.

It's like a "sliding min instance", you can scale more seamlessly.