Why AWS scrapped OpenSearch’s architecture to chase agent workloads

AWS has overhauled OpenSearch Serverless to better handle bursty AI agent workloads, with separate storage and compute and a new proprietary storage layer.

The redesigned service can scale to zero when idle, supports search and vector collections at launch, and is aimed at lower-cost, faster-scaling usage.

AWS says the new architecture could cut costs by up to 60% versus provisioned clusters at peak capacity.

It also adds integrations with Vercel and Kiro, plus OpenSearch Agent Skills for tools like Claude Code and Cursor, with more features planned for memory, log analytics, governance, and reasoning.