Overview
The AWS Agentic AI Lens is prescriptive guidance for evaluating agentic AI workloads against best practices. In the article’s account, the Lens is not yet selectable in the AWS Well-Architected Tool, so teams cannot run it natively through that console workflow.1 The Lens organizes 150 best practices across 41 questions and six pillars, giving teams a structured basis for reviewing an agentic system.1
Operationalizing the review
Smarsh, which needed a rigorous and evidence-based process for a workload in a highly regulated industry, converted the Lens into an executable assessment by building a custom Claude skill.1 The implementation stores the best practices in a structured JSON file and uses a Markdown guide to specify the rubric, evidence requirements, and constraints against unsupported claims.1 A Python script then generates a validated Excel workbook containing assessment statuses, supporting evidence, gaps, and remediation items.1
In the described workflow, Claude reads the skill, gathers evidence from the workload’s GitHub repository, assesses the practices, and produces the workbook; the article reports that this takes hours rather than days.1 Smarsh also tested the skill against synthetic repositories before relying on its output.1 This design combines structured guidance and deterministic report generation with AI-assisted evidence gathering, rather than treating the model as the sole source of assessment results.1
Practical significance
The case illustrates one way to make reviews more repeatable when a lens is not available as a built-in console workflow: encode the requirements and rubric explicitly, require evidence for assessments, and produce a consistent report through a validating script.1 The article describes the approach as improving review consistency, speeding compliance work, and supporting collaboration with AWS on remediation.1 The method is an implementation example, not a claim that the Lens itself runs automatically or that a generated review replaces expert judgment.1
Related Pages
- AI Agent Security — The AWS Agentic AI Lens structures reviews of agentic workloads, including the security controls and human oversight needed to manage their risks.
- Enterprise AI Agent Architecture — The Lens reviews agentic workloads whose runtimes, tools, memory, governance, and observability are central to enterprise agent architecture.
- Amazon Bedrock AgentCore — AgentCore provides AWS agent runtime and permission safeguards that can be assessed as part of an Agentic AI Lens review.
- Amazon Bedrock — Bedrock's managed foundation-model and agent capabilities are AWS services whose agentic workloads can be reviewed using the Lens.
- AI-Ready Data — Reliable evidence gathering for Lens reviews depends on trustworthy, traceable, and governed data.