AWS AI Updates: September 11, 2026

1. Lambda Durable Functions Now Checkpoint Every Pydantic AI Model and Tool Call

AWS. Lambda durable functions integrates with Pydantic AI so that each model call and each tool call inside an agent run becomes its own durable execution step. An interrupted run resumes from the last completed step rather than replaying the whole agent, which avoids paying twice for completed inference calls and avoids duplicate side effects from tools that write to external systems. The integration works with any Python-based Lambda durable function and is available in every region where durable functions run, at standard Lambda pricing. Source

2. OpenSearch Serverless Became a v0 Provider for Vector and Full-Text Search

AWS. Amazon OpenSearch Serverless is now provisionable directly from v0 by Vercel: describing an app in natural language causes v0 to create the collection, index the data, wire up the endpoint, and set the environment variables. v0 loads provider-specific agent skills so generated code follows the recommended OpenSearch patterns, and the integration covers vector search for RAG workloads as well as full-text search. It is live in 17 regions across North America, South America, Europe, and Asia Pacific. Source

3. AWS Elemental Inference Extracts Scene-Level Metadata From Live Video in Real Time

AWS. Elemental Inference now analyzes live streams in parallel with encoding and emits structured metadata: IAB taxonomy categories, GARM brand suitability signals, detected objects and actions, and scene descriptions. The metadata feed ID is embedded into SCTE-35 ad markers through MediaLive so MediaTailor can pull the signals at ad break time for context-aware ad decisioning. The pitch is scene-level understanding of a live feed without standing up custom ML infrastructure. Source

4. AWS Transform for .NET Generates Unit Tests for the Code It Modernizes

AWS. AWS Transform for .NET now assesses an application for testability in parallel with the standard .NET assessment, plans which classes and methods to cover, and generates the unit test code. It targets testable units such as business logic and controllers, and is opt-in either at job start or after a transformation completes. The feature is exposed through the AWS Toolkit for Visual Studio extension in all regions where Transform for .NET is supported. Source