AWS just turned its Well-Architected best-practice stack into something that acts more like a staff architect than a static checklist. On October 1, 2026 the company put AWS Well-Architected Agent into public preview: an AI-powered Support service that reads your infrastructure, metrics, and topology, then ranks fixes by the business goals you declare.
If you already run Trusted Advisor and the classic Well-Architected Tool, this is the next layer. The agent is meant to stop at ready-to-apply remediations (console walk-throughs, CLI, updated Terraform/CloudFormation/CDK), not another PDF of findings you never finish.
What ships in preview
Per the AWS What's New post and Channy Yun's longer AWS News Blog write-up, the agent correlates utilization metrics, resource configs, and application topology against Well-Architected practices across more than 65 AWS services. You set goals across cost, security, performance, and resilience (AWS also says "reliability" in the What's New blurb). Recommendations come back prioritized by impact and effort at resource, application, and architecture levels.
A concrete example from AWS: if reliability is your top goal, you might get a multi-AZ failover suggestion for a critical database, with an SSM runbook to apply the change and a cross-pillar note on how that move hits cost and performance before you commit. Where it can, the agent also reviews IaC (Terraform, CloudFormation, CDK), flags gaps, and returns the code edits to align with the framework.
| Layer | What the agent returns | How you act |
|---|---|---|
| Resource findings | Specific issues, dollar impact when AWS can estimate it, step-by-step remediations | Console walk-through, CLI, or SSM runbook |
| Application findings | Consolidated issues across multiple resources in one app context | Same remediation menu, scoped to the app |
| Architecture patterns | Broader design gaps with IaC code changes | Copy updated Terraform / CloudFormation / CDK into your repo |
Onboarding, from the News Blog: create an agent profile in the Well-Architected console, pick accounts/Regions and pillars, attach customer-managed IAM roles so the agent can read configs and metrics, then wait up to about 24 hours for the first resource and application recommendations. You can also upload a zip of pre-production IaC for an architecture review before deploy. API access and an AWS MCP Server path are called out for wiring recommendations into existing ops tooling.
Where it runs, and the Support catch
Agent endpoints are in US East (N. Virginia), US East (Ohio), and US West (Oregon). You can onboard workloads from any AWS commercial Region. The service is delivered by AWS Support and needs an AWS Support plan. AWS does not publish a separate consumption price in these launch posts, so budget conversations still start with whatever Support tier you already pay.
The Register frames the launch as AIOps with an escape hatch: AWS is pushing recommendations with manual apply paths (including CLI) because platform teams still want to verify machine advice before it touches production. That matches the News Blog's Responsible AI note: generative recommendations can be wrong or incomplete, and you own evaluation and oversight in your environment. The classic Well-Architected Tool with custom lenses remains available for fully manual reviews.
Geeknewz practical take
Geeknewz's view: if you already pay for AWS Support and your FinOps or platform team still runs Well-Architected reviews as a quarterly scramble, this preview is worth a scoped pilot on one non-critical account. Start with cost or resilience as a single declared goal so you can judge whether the ranked list and IaC diffs beat Trusted Advisor noise. If you are on basic support only, this is not a free bolt-on. And do not let "automation-ready" turn into unattended production changes on day one. Treat the first month as a second pair of eyes that still needs a human merge.
Source: AWS What's New; AWS News Blog; The Register.