How can I use serverless inference with models generated by SageMaker Autopilot?

At this point there is no direct integration between SageMaker Autopilot and Serverless inferencing. One way to do this take the model.tar.gz from the best training job and deploy it manually as an serverless endpoint. Usually Autopilot uses one of the SageMaker Built in Algorithms and you will be able to us the specific container to deploy it.