AWS previews an agent for cloud architecture reviews
The preview combines infrastructure recommendations and template reviews with limits on access and execution
AWS announced a preview of its Well Architected Agent on October 1. The service ranks infrastructure recommendations against business goals across cost, security, performance and resilience. It also reviews infrastructure templates and proposes changes before deployment.
Access depends on the support plan
The eligibility table limits access to Business+, Enterprise On Ramp, Enterprise Support and Unified Operations customers. Developer and standard Business plans are excluded. Profiles are hosted in Northern Virginia, Ohio or Oregon, but can scan resources across commercial AWS Regions. A profile can cover up to 100 accounts.
AWS’s setup guide requires a profile, business goals, account access roles and at least one application context for scheduled recommendations. With a properly configured profile, initial generation begins within 48 hours. Creating the profile alone does not finish onboarding.
The refresh schedule needs clarification
The dedicated freshness guide and release notes describe weekly refreshes. The recommendations overview instead says roughly every 24 hours. These pages conflict, so teams should confirm the cadence before relying on it operationally. Resource and application recommendations cannot be requested on demand.
Template reviews are a separate workflow
The architecture review guide describes a manually started check of templates stored in S3, with the bucket in the profile’s hosting Region. Reviews use the Well Architected Framework lens and return updated code in the source language. The limit is five reviews per day per profile.
The announcement names Terraform, CloudFormation and CDK. The detailed input guide explicitly names CDK and Terraform projects, supplied as a ZIP archive or S3 folder. Teams should check their actual project format before treating the preview as a replacement for an existing review process.
Generated fixes still need an operator
The implementation guide draws an important boundary. Recommendations derived from Trusted Advisor checks can use predefined Systems Manager runbooks that the service executes with consent. These can also be scheduled.
For recommendations generated by the AI engine, the service supplies guidance such as commands, console instructions or updated infrastructure code. AWS says it does not execute these generated steps on the customer’s behalf. Customers remain responsible for reviewing, testing and running them.
AWS also labels application recommendations as beta in the service overview and asks customers to review their relevance before acting.
Judge the change rather than the suggestion
Our view is that a useful trial should measure how much engineering review a proposed change still needs. Pick a contained workload, compare the advice with an existing review and inspect both the intended benefit and any new failure mode. Keep approval and recovery checks in the normal deployment process.
Archival March 2019 photograph titled AWS Amazon Web Services Office in Houston, Texas. Copyright 2019 Tony Webster, licensed under Creative Commons Attribution 2.0. Resized and converted to WebP. The photograph shows office signage, not the agent interface.
