AWS AI Updates: October 29, 2026

1. Amazon Bedrock AgentCore CLI Reaches Version 1.0

AWS released version 1.0 of the Amazon Bedrock AgentCore CLI (@aws/agentcore 1.0.0), a rebuild with faster startup, a consistent command structure, and support for more AgentCore resources. Developers can use the project workflow or work directly with harnesses, runtimes, memory, identity, gateways, and payments without first adding them to a project. The CLI can invoke existing resources by ID or ARN, open an interactive shell, and export a harness, and it adds searchable menus, resource pickers, and guided workflows. Global flags cover region selection, JSON output, debugging, and version information across all commands. Existing projects can be moved to version 1 with agentcore migrate, and the CLI is installed from npm with source on GitHub. Source

2. Amazon Quick Adds Classifier-Based Smart Approvals for Agent Actions

AWS added smart approvals to Amazon Quick, a review layer aimed at reducing approval fatigue when agents act on a user’s behalf. A separate classifier model evaluates each eligible action against the user’s request, approving routine actions automatically and referring consequential ones to the user. Actions such as sending content outside the organization, changing permissions, deleting content, making payments, or going beyond what the user asked for are sent for human review. Smart approvals only applies where the user’s setting is “ask” each time, never grants new access, covers only the exact action reviewed, and routes the action to the user if a review fails or times out. It is on by default, admins can restrict it with custom permissions, users can turn it off per conversation or scheduled run, and AWS CloudTrail logs every review and setting change. Source

3. Lambda Durable Functions Gain a Telemetry Exporter and an MCP Server for Agent Queries

AWS released the Durable Execution Telemetry Exporter plugin for AWS Lambda durable functions, which are used for multi-step applications and AI workflows that can run for up to one year. The plugin captures step-by-step execution history (status, operation timing, step input and output, errors) and sends it to any of 13 built-in destinations, including CloudWatch Logs, S3 for Athena, DynamoDB, Aurora, and OpenSearch Service, with field redaction and a configurable sampling rate. AWS also released a companion Model Context Protocol (MCP) server so AI agents such as Kiro can query the exported telemetry in natural language, for example to list recent failed executions and find the cause. The exporter supports TypeScript, Python, Java, and Go and is available at no additional charge in all Regions that support durable functions. Source