AWS Well-Architected Agent preview: Support eligibility, Regions, and review workflow

What launched on October 1, and what remains in preview

The official AWS release notes record the release of AWS Well-Architected Agent on October 1, 2026. The launch announcement and current product documentation provide verifiable scope information. Compiled on October 2, this article discusses preview capabilities; it does not treat its preparation date as the launch date or invent an exact release time not provided by AWS.

The Agent continuously analyzes AWS environments and provides recommendations at resource, application, and architecture levels across four optimization pillars: cost, security, resilience, and performance. It prioritizes them according to organizational business goals. Application-level recommendations are separately in beta, and AWS explicitly requires review of AI-generated content. This is a different product from Bedrock Managed Agents, which runs model/tool sessions.

Check the Support plan before looking for the console entry point

The current eligibility table lists Business+, Enterprise On-Ramp, Enterprise Support, and Unified Operations; Developer and Business are not eligible. Business and Business+ are different names. Having some form of paid Support does not establish eligibility.

Business+ permits up to 2 profiles and 7 applications per profile. The three higher plans listed above permit up to 10 profiles and 30 applications per profile. Have your organization's AWS administrator first verify the current Support plan and target account, then check for Agent profiles in the Well-Architected console. A missing entry point does not establish a network fault and should not prompt the purchase of an unverified plan.

Assess hosting Regions, scanning scope, and quotas separately

Profiles can currently be hosted in US East N. Virginia (us-east-1), US East Ohio (us-east-2), and US West Oregon (us-west-2). Documentation also permits scanning resources in all commercial Regions. Do not extend “all commercial Regions” to China Regions, GovCloud, or every resource type, or claim that the Agent has a hosting endpoint in any Region.

Each profile can analyze up to 100 AWS accounts and define 10 goals. Each recommendation generation produces up to 30 recommendations; that limit does not apply to the total number stored. Architecture review permits up to 5 runs per profile per day, which may be consecutive while quota remains. The product overview says scheduled recommendations refresh weekly, while the quotas page requires at least 24 hours between scheduled generations for the same profile. The former is a refresh cadence and the latter a minimum cooldown; they do not combine into a guarantee of new recommendations every day.

Prepare a controlled trial using the official setup path

Read Getting started first: enter the AWS Well-Architected console, select Agent profiles, then Add profile. Define target accounts, Regions, optimization pillars, and at least one business goal. Start with a small set of noncritical experimental resources and a clear cost, performance, or resilience question to validate.

Have an authorized administrator configure the execution role, cross-account access roles, and WellArchitectedAgentResourceScanning permissions according to the documentation. Permissions affect resource scanning; do not grant arbitrary administrator access to troubleshoot. This article has not created roles, connected accounts, or executed remediation, and does not establish that your tenant has preview access.

Add at least one application context; without it, scheduled recommendations do not run. After saving, allow up to 48 hours for the first recommendations, and confirm that the profile, accounts, Regions, and application context match the intended scope. Do not repeatedly create duplicate profiles while waiting. Record the actual start time and console status.

Generating recommendations does not authorize deployment

Implementing recommendations distinguishes AWS Trusted Advisor SSM runbooks from AI-generated guided actions. AI suggestions may provide CLI, SDK, console steps, or updated IaC templates, but may contain errors or incomplete information. Review them before production use.

One-click SSM remediation also requires customer consent. For an IaC proposal, first compare the original template with the suggested changes and check permissions, data-deletion risks, network changes, resource replacement, and estimated costs. Validate in a test environment before deploying under your organization's approval process. Retain the original template, execution records, and rollback path. “ready-to-deploy” does not mean the Agent has already modified production on its own.

How to assess the trial and when to stop

At minimum, check the actual resources referenced, target pillar, business goal, effects across pillars, trade-offs, and risk mitigations. For resources not scanned, first check unsupported resource types, then scanning permissions and application context. An absence of recommendations does not establish that all resources are compliant.

Stop execution and ask the responsible team to review if a recommendation references nonexistent resources, uses inapplicable cost assumptions, or proposes unapproved permission changes or deletion. For a profile with no initial results after 48 hours, retain its status, configuration, and actual timing, then consult documentation or the Support service you are entitled to use. Recreating profiles or blindly expanding permissions is not a substitute for diagnosis.

What remains unknown about cost and applicability

The official material reviewed confirms Support eligibility and feature quotas, but no dedicated Well-Architected Agent pricing statement was verified. The traditional Well-Architected Tool's “no additional charge” wording cannot be transferred directly to the new Agent. Confirm Support-plan costs, underlying AWS resource costs, and costs arising from actual remediation separately.

Cost Explorer is optional. When it is disabled, the Agent estimates costs using public AWS pricing. For more specific resource-level cost data, check Cost Explorer and resource-level daily-granularity configuration in Getting started. Official documentation says the main Cost Explorer service cannot be disabled once enabled, although optional granularity settings can be adjusted. The console itself has no usage charge, but API requests and resource-level granular data may incur charges. Check these costs and organizational data settings before deciding to enable it; do not treat it as a free prerequisite that can be reversed at any time.

This article suits organizations evaluating cloud environment optimization and architecture review. If you lack eligible Support, clear business goals, a team able to review IaC, or a safe trial scope, establish those conditions first. For general model-invocation needs, see the Bedrock Managed Agents public preview guide. Do not mix the two Agents' Regions, interfaces, or eligibility rules.