Why AWS scrapped OpenSearch’s architecture to chase agent workloads

AWS has launched a near-complete rebuild of Amazon OpenSearch Serverless, with 97% of the architecture rewritten from scratch. The key change is separation of storage and compute on a new proprietary storage layer, enabling true scale-to-zero when idle and auto-scaling 20x faster than the previous generation. The redesign targets AI agent workloads, which have bursty usage patterns that broke the original serverless architecture's assumptions. Cost savings of up to 60% versus peak-provisioned clusters come from the new compressed storage layer and aggressive auto-scaling. The launch includes support for both search and vector collection types, native integrations with Vercel and Kiro IDE, and OpenSearch Agent Skills compatible with Claude Code and Cursor. A long-term agent memory feature is planned for H2 2026, a major log analytics launch is coming in June, and a TIMESERIES collection type will follow at AWS's New York Summit. AWS positions OpenSearch as a semantic layer for LLMs rather than something to be replaced by them.