The models are so, so good these days. You can just ask them to fix a huge performance edge case that you never encounter but hear about in bug reports. They spend 90 minutes profiling, sampling, and making a demo app to validate the issue — ultimately fixing it in your app when they're certain.