Amazon Web Services rolled out a public preview of Amazon Bedrock Managed Agents powered by OpenAI, deploying a customized version of OpenAI’s Agents API built to be native to AWS infrastructure.
Execution Environments and Frontier Model Expansions on Bedrock
Engineers can execute these agents using two distinct runtime paths. They can deploy self-hosted compute on an existing development machine, local container, or standard compute environment.
Alongside the agent architecture, AWS expanded its foundational model catalog on Bedrock by integrating several frontier systems. Meanwhile, OpenAI GPT-6 Astra UltraFast mode introduces a premium speed tier designed for high-throughput API workloads, driving up to six times faster inference and hitting up to 300 tokens per second.
Infrastructure Audits and Native Data Lake Integration
In addition to runtime and model updates, the platform introduced several operational features. The AWS Well-Architected Agent entered public preview, providing an AI-driven service that inspects infrastructure configurations to deliver contextual recommendations targeting cost, security, performance, and resilience. Concurrently, Amazon Aurora PostgreSQL gained direct querying capabilities for Apache Iceberg and Parquet data. This lets database administrators query operational tables and data lakes simultaneously without maintaining separate extract, transform, and load pipelines or duplicating storage.
Complementing the Aurora update, Amazon S3 Tables added native support for Apache Iceberg V3 data types. Developers can now store geospatial coordinates, geometry, geography, unknown parameters, and nanosecond-precision event times directly within table schemas, eliminating the need to encode time-series or coordinate data into basic strings or integers.
Q3 Service Lifecycle Updates and Maintenance Transitions
AWS also detailed its quarterly product lifecycle adjustments, outlining deprecations and migration paths for several developer and enterprise tools. Daily.dev coverage confirms that multiple services are transitioning to maintenance status, ending new customer access on October 29, 2026.
Other offerings are entering sunset status with fixed end-of-support timelines. Amazon Managed Blockchain, Amazon DevOps Guru, and the AWS Backint Agent for SAP ASE all reach absolute end of support on September 29, 2027. Meanwhile, the standalone console for AWS Infrastructure Composer faces discontinuation on December 7, 2026, and Amazon Mechanical Turk reached its formal end of support on September 29, 2026.
Automating Multi-Agent Workflows with New Open-Source Models
Expanding beyond infrastructure infrastructure, AWS introduced Kiro workflows to automate complex, multi-stage tasks with minimal human oversight. Additionally, the platform released Strands Decider, a small open-source 2B decision model optimized for rapid local experimentation and fast system-one routing. To support production deployment of these architectures, AWS established three new Forward Deployed Engineering partner pathways, backed by the organization’s $1 billion investment initiative announced earlier in the year.
