Amazon OpenSearch Serverless Next Generation: Rebuilt for Agentic AI
AWS announced the next generation of Amazon OpenSearch Serverless on May 28, 2026, describing it as a ground-up rebuild designed for agentic AI and dynamic workloads rather than an incremental update to the existing service. The headline capability change is instant autoscaling: the new generation scales from zero to thousands of requests per second and back to zero during idle periods, doing so up to 20 times faster than the previous generation. Collections are ready in seconds rather than minutes.
For developers, the most immediate difference is cost. AWS reports up to 60% cost savings compared to provisioning OpenSearch Service clusters for peak capacity, achieved through the scale-to-zero architecture. The billing model separates compute (OpenSearch Compute Units for indexing, search, and GPU-accelerated operations) from storage (GB-month), which means read-heavy or bursty workloads no longer pay for idle indexing capacity. Full-text search and vector search collection types are both supported, making it suitable for the hybrid retrieval patterns that agentic applications typically need.
The integrations with developer tooling are equally notable. The service ships with native connectors for Vercel and Kiro, meaning teams can attach an OpenSearch Serverless collection to a Next.js or Kiro-based application from the project dashboard without writing custom provisioning code. An OpenSearch Agent Skills repository provides pre-built skill packages that encapsulate domain-specific search logic, and OpenSearch Launchpad inside Kiro offers guided architecture planning for common patterns. AWS SDK and CLI support for programmatic collection creation is available at launch, and the service is generally available across all commercial AWS regions where OpenSearch Serverless currently operates.
Read more — AWS News Blog
Google Cloud Looker MCP Server Now in Preview
Google Cloud released a Looker-managed Model Context Protocol server in preview, allowing AI agents and MCP-compatible tools to connect directly to a Looker instance and query its semantic models. The integration is enabled from the Admin panel in Looker, where administrators can turn on MCP support and configure which AI tools are permitted to connect.
For teams already using Looker as their semantic layer, the MCP server means that the metrics, dimensions, and business logic encoded in LookML becomes directly accessible to any agent that speaks MCP — without requiring the agent developer to rebuild those definitions in a separate prompt or retrieval pipeline. An agent asking "what was last quarter's revenue by region?" can resolve that through the Looker semantic model rather than requiring the question to be mapped to a SQL query from scratch each time.
The server also integrates with BigQuery and Snowflake connections in preview: Looker can now expose in-database analytic models (BigQuery Graph and Snowflake semantic views) directly, keeping semantic definitions consistent across Looker and other BI tools. This addresses a common pain point where teams maintaining both a Looker instance and AI applications had to keep two separate sets of metric definitions in sync. The Looker MCP server is available for workspaces running Looker 24.0 and later.
Read more — Google Cloud Blog
Links & Sources
You May Also Enjoy
