How AWS automates initial cloud reviews without replacing enterprise judgment

AWS is automating architecture reviews while enterprises retain judgment

AWS’s Well-Architected Agent automates initial cloud architecture analysis while enterprises retain approval decisions. AWS says the AI-based service, now in public preview, examines customers’ AWS environments, recommends improvements, and provides implementation guidance. Enterprises can supply business goals to help rank the findings, while their leaders decide which costs, risks, and operational trade-offs to accept. AWS sells both the infrastructure being reviewed and the Support plans required for access, so it has a commercial interest in adoption.

The Agent builds on more than a decade of architecture-management products: AWS introduced the Well-Architected Framework in 2015 and the Well-Architected Tool in 2018. The Well-Architected name comes from that framework and product history. For architects and platform teams, the question is how much review work direct environment analysis can remove, given the service’s refresh schedule, coverage, and access requirements.

Direct analysis changes the work of a review

To understand that change, start with the existing Well-Architected Tool. Teams describe their workloads by answering structured questions, assess them against the Framework’s best practices, and receive an improvement plan. Their answers provide the evidence, so preparing that description is part of the review.

The Agent takes on the initial evidence gathering by examining the AWS environment directly, AWS says. Teams can then add application context and business objectives, which the Agent uses to prioritize its findings. Reviewers therefore start with findings drawn from the environment and add the information needed to interpret them.

AWS says the findings cover individual resources, applications, and the overall architecture. A reviewer can therefore assess a proposed resource change in the context of the application using it and the wider system design. Recommendations also describe corrective actions and their possible effects on cost, helping teams understand both the engineering work and its financial consequences.

Once a team chooses an action, AWS says the implementation guidance can reflect the environment under review. It may provide steps for the AWS console, commands for the AWS command-line interface, or AWS CLI, and proposed changes to infrastructure as code, the configuration files used to define and manage infrastructure. Engineers thus get a more concrete starting point for implementation than the existing Tool’s improvement plan provides.

But those instructions still need to pass through the enterprise’s approval and change-management process. Calling the service agentic, meaning able to carry out tasks using AI, does not establish that it applies corrective changes autonomously. Engineers must decide whether and how to use the console instructions, commands, and configuration updates within their operating procedures.

Siddharth Mehta, research leader at consulting and services company Avasant, sees this approach as a way to speed up optimization for cloud architects, DevOps teams that combine software development and operations, and platform engineers. Avasant’s consulting and services business gives it a commercial interest in how enterprises adopt and manage cloud technology. Mehta believes direct analysis and detailed guidance can help teams move from occasional reviews and reactive corrections toward more continuous, AI-assisted optimization.

Refresh timing limits how current findings remain

More frequent optimization raises a practical question: how quickly do recommendations reflect changes in the environment? Nhat Bui, deputy director of engineering at Scopic Software, argues that agentic applications, in which AI agents carry out tasks, create both faster change and greater demand for review. His perspective comes from a software company with a commercial interest in how enterprises build and maintain applications.

“First, AI applications change faster and have less predictable cost profiles than traditional workloads, so a review done last quarter goes stale quickly. Second, more infrastructure is now being written by AI coding tools, and human reviewers can’t keep pace with that volume,” Bui said. He identifies two pressures on review teams: assessments lose relevance as systems change, while AI coding tools increase the amount of infrastructure people must examine.

The Agent’s refresh schedule determines how it responds to those pressures. “The Agent refreshes resource-and application-level recommendations weekly, while architecture-level recommendations are generated through on-demand reviews,” Bui said. Teams therefore get scheduled updates at the resource and application levels and must request reviews for architecture-level findings.

Between updates, infrastructure and consumption can change, particularly in agentic applications, Bui added. Before acting, reviewers need to consider what has changed since a recommendation was generated. Mehta’s expectation of more continuous optimization describes a shift in working practices, but Bui’s account of the refresh schedule sets a timing limit on the information supporting that work.

Business goals guide ranking while enterprise context guides approval

Even a current recommendation needs business context before a team can decide whether to act. Pareekh Jain, principal analyst at Pareekh Consulting, sees ranking against business objectives as a way to speed up cloud decisions. His consulting business has a commercial interest in enterprise technology decisions, and his assessment focuses on helping CIOs allocate investment and engineering effort.

“Instead of giving CIOs hundreds of technical recommendations, the agent can help identify which improvements matter most to the business. That means CIOs can decide where to prioritize cloud investments and engineering resources, and where to accept trade-offs,” Jain said. In his view, ranking eases the work of sorting recommendations while leaving investment priorities and acceptable compromises with the CIO.

For that ranking to help, the enterprise must supply objectives that reflect its needs. Mehta cautions that relevant information can also sit outside the AWS environment, including contractual service-level agreements, or SLAs, which set service-delivery commitments, and risk appetite, the amount of risk an organization will accept. These considerations can change whether a technically useful recommendation suits a workload.

“The Agent only understands the context available to it. It may not know about regulatory obligations, third-party dependencies, business-process dependencies, contractual SLAs, internal risk appetite, multi-cloud dependencies, and organizational policies not represented in AWS,” Mehta said. Reviewers must bring these obligations and relationships into the decision when the Agent cannot see them.

Those relationships can determine what happens when a proposed change is made. An outside provider or business process may depend on the resource being changed, while an application spanning multiple cloud providers may require a broader assessment. Regulatory duties and internal policies can also limit the choices, so enterprises with substantial requirements in these areas need to supplement the context held within AWS.

Mehta puts the limit directly: “Therefore, business-goal alignment improves prioritization but does not replace enterprise architecture or risk governance.” For architecture and platform leads, preparing context is therefore part of the review. Teams need a way to bring policies, commitments, and external dependencies into approval decisions, even when the Agent has already ranked a recommendation highly.

Access, coverage, and consumption shape the evaluation

Beyond preparing that context, an enterprise must establish whether it can access the preview. AWS makes the Agent available to customers with AWS Business+, Enterprise On-Ramp, Enterprise Support, or Unified Operations plans. The Well-Architected Tool requires no paid AWS Support plan, so the products have different access conditions.

Jain sees the higher-tier Support requirement as a potentially significant constraint on enterprise adoption. An organization with a qualifying plan already meets the condition, while one without a plan faces a separate Support purchase decision. Eligibility therefore determines which organizations can test whether direct analysis reduces their review workload.

For eligible organizations, the next question is whether the Agent covers the intended assessment. Mehta says it currently addresses cost, security, performance, and resilience, while the Tool covers all six areas, or pillars, of the Well-Architected Framework. Operational excellence and sustainability complete those six, so a full Framework review requires broader coverage than the Agent currently provides.

Coverage is separate from charges. AWS says neither the Agent nor the Tool has a specific additional fee, but enterprises can incur charges for the underlying AWS resources, services, and data they consume. That consumption can involve Trusted Advisor, Compute Optimizer, Security Hub CSPM (cloud security posture management), Resilience Hub, Cost Optimization Hub, and Cost Explorer.

An evaluation therefore needs to account for the services and data used during the review. A qualifying Support plan establishes access, but associated consumption can still carry charges. Teams should assess costs against the resources and services their evaluation actually uses, since the possible charges listed do not necessarily apply to every evaluation.

Key highlights

• Automate the review starting point: AWS’s Well-Architected Agent examines cloud environments directly and provides findings and implementation guidance. Architecture and platform teams retain responsibility for validating recommendations and routing changes through enterprise approval processes.
• Check recommendation freshness: Resource- and application-level recommendations refresh weekly, while architecture-level findings require on-demand reviews. Platform owners need to check intervening infrastructure and consumption changes before acting.
• Bring business obligations into approval: Business goals help the Agent rank findings, while regulatory duties, contractual commitments, and external dependencies may sit outside its available context. Architecture and risk owners need to assess high-priority recommendations against these obligations.
• Define the evaluation scope and cost: Access requires a qualifying AWS Support plan, and current coverage spans cost, security, performance, and resilience. Cloud platform and finance owners need to account for operational excellence and sustainability separately and assess underlying service charges, even though the Agent carries no additional fee.