AWS has overhauled the architecture of OpenSearch Serverless, a managed search and vector engine, to account for the spread of AI agents.

According to a report by The News Stack, Tia White (티아 화이트), AWS head of OpenSearch, said, "We rebuilt 97 percent of OpenSearch Serverless from scratch." She said the previous serverless architecture could not handle irregular AI agent usage patterns, meaning bursts of use followed by long periods of idleness.

The biggest change is the separation of storage and compute. Based on a new storage structure developed by AWS, costs fall to zero during idle periods and the system restarts within seconds when an agent request comes in.

Autoscaling is now 20 times faster than the previous generation, and AWS said compression features and faster capacity reduction provided by the new storage layer deliver cost savings of up to 60 percent.

AWS also launched OpenSearch Agent Skills, which integrate with developer tools such as Claude Code and Cursor.

AWS also disclosed its future roadmap. It plans to launch an agent long-term memory feature with built-in evaluation and governance functions in the second half of 2026. In June, it will also unveil a major update targeting the log analytics market led by Datadog, Splunk and Grafana. White said, "There may be a question of whether LLMs can replace OpenSearch, but OpenSearch Serverless will be the core semantic layer that LLMs call."