# AWS Adds Infrastructure and Governance Tools for AI Agents

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/28496
Published: 2026-10-08T20:14:48.000Z
Updated: 2026-10-08T20:14:48.000Z
Section: Technology

> The releases span runtime performance, managed agents, architecture reviews and sandbox controls, with features at different stages from general availability to developer preview.

Amazon Web Services has introduced a series of infrastructure and governance tools for AI agents, targeting persistent state, startup speed, permissions and oversight as software takes on longer multistep tasks.

The releases center on Amazon Bedrock AgentCore, which AWS positions as an infrastructure layer for workloads in which agents retain context, call external tools, execute code and perform multiple actions without a new human instruction at every step.

The next-generation AgentCore Runtime became generally available Sept. 18 with elastic memory management, session isolation and snapshot-based startup. Testing produced 75th-percentile cold-start times of 1.9 to 2.0 seconds for container images ranging from 200 MB to 2 GB. The previous version recorded times ranging from 5.4 to 30 seconds in the same comparison.

The runtime launched across five regions: US East (N. Virginia), US East (Ohio), US West (Oregon), Europe (Ireland) and Asia Pacific (Tokyo).

On Sept. 29, AWS announced Bedrock Managed Agents, developed jointly with OpenAI. The preview allows agents to run inside AWS, preserve session state, use tools, execute code and coordinate multistep tasks.

Each agent receives its own AWS Identity and Access Management (IAM) role. Customers can require human approval before consequential actions, and supported application programming interface activity is recorded through AWS CloudTrail. The preview is available in US East (N. Virginia), US West (Oregon) and US East (Ohio), with no additional Bedrock Managed Agents charge during the preview beyond the cost of underlying AWS resources.

AWS Well-Architected Agent entered public preview Oct. 1. It can evaluate infrastructure across more than 65 AWS services and produce recommendations covering cost, security, performance and resilience.

The tool generates resource-level, application-level and architecture-level recommendations, including infrastructure-as-code changes. Scheduled recommendations are refreshed on a weekly cadence, and one agent profile can analyze up to 100 AWS accounts.

On Oct. 7, AWS launched Strands Box in developer preview as an open-source sandbox under the Apache 2.0 license. The system combines operating-system isolation with Dogwood policies that govern agent actions.

Those policies can assess an agent’s previous actions, their sequence and accumulated limits instead of treating every action as an isolated event. AgentCore Runtime is generally available, while Bedrock Managed Agents, AWS Well-Architected Agent and Strands Box remain in preview stages.