Neon has introduced an innovative autoscaling feature for serverless Postgres that allows seamless scaling of CPU and memory resources without dropping connections or interrupting queries, thus enabling applications to handle peak demand efficiently while reducing infrastructure costs. Over a year after its launch, nearly a third of Neon's customers utilize this feature, demonstrating its effectiveness and stability, despite challenges such as kernel panics and I/O throttling. The architecture utilizes virtual machines in Kubernetes clusters, with a custom resource definition and controller called NeonVM, and an autoscaler-agent to monitor metrics and make scaling decisions. Recent improvements have focused on smarter autoscaling using metrics from the Local File Cache and accommodating rapid memory allocation through swap, with plans to enhance the scaling algorithm using HyperLogLog estimators. These advancements are part of Neon’s broader efforts to create a cloud-native serverless Postgres platform that separates storage and compute, offering unique features like branching and point-in-time restore. Neon encourages user feedback and participation to further refine its platform and support the development of next-generation applications.