AWS updates Amazon Bedrock, AgentCore, and Strands with faster serverless agents, built-in evaluation, and enterprise connectors

AWS described the September 2026 updates as part of its strategy to give builders "real choice at every layer: model, runtime, and tooling" across Amazon Bedrock, AgentCore, and Strands. The company framed the changes as a response to shifting industry priorities, where customers now weigh cost, security, and governance alongside model performance when deploying agents in production.

Amazon Bedrock Managed Agents added public preview support for OpenAI models, allowing teams to build production-ready agents while retaining AWS-native security controls. AWS highlighted that data remains within AWS, IAM permissions are reused, and full auditability is maintained through AWS CloudTrail. The update also introduced durable sessions and built-in human approval workflows to reduce development overhead and mitigate risk during agent operations.

AgentCore, a managed infrastructure layer for building and deploying enterprise agents, received a runtime update aimed at improving serverless agent efficiency. AWS cited more efficient memory management and lower cold start latency as key improvements, enabling pay-as-you-go pricing without pre-provisioned capacity. Sessions scale to zero when idle and run in hardware-isolated environments, according to the announcement.

Strands, an open source agent toolkit, launched Strands harness, a new open source agent harness designed to match popular harnesses on accuracy while using 28% fewer tokens. AWS stated that developers can initialize a production-ready agent in a single line of Python or TypeScript, with built-in context management, prompt caching, and memory. The toolkit supports deployment flexibility across environments.

The announcement emphasized enterprise-grade considerations such as governance, security, and cost efficiency, reflecting a broader industry shift beyond model performance alone. AWS positioned these updates as foundational for scaling agentic AI in enterprise workflows that require reasoning over enterprise knowledge, long-horizon task execution, and secure, auditable operations.