Neon has introduced an autoscaling feature that dynamically adjusts Postgres instances to match the workload's working set size, optimizing performance and cost-efficiency. This innovation leverages techniques like HyperLogLog for estimating the working set size and introduces a time-bounded variant to better handle varying workloads by using sliding windows with timestamps. The approach aims to maintain the working set in memory, significantly boosting throughput and ensuring predictable performance, especially in read-heavy workloads where cache misses can incur latency due to network-based storage. The method seeks to strike a balance between responsiveness and resource efficiency, particularly for workloads that fluctuate over time. Initial results show that this strategy improves performance while being more cost-effective than fixed-size nodes, although it does present some edge cases and potential areas for future refinement. Neon draws inspiration from similar efforts by AWS Aurora, suggesting a potential for standardization in working set estimation techniques across platforms.