AWS Architecture Blog·June 22, 2026Architecting an AI-Powered Resilience Framework on AWS
This article outlines a five-layer AI-powered resilience framework on AWS designed to automatically discover system dependencies, generate targeted chaos experiments, and integrate with CI/CD pipelines. It addresses the challenges of traditional resilience testing, such as outdated documentation and the need for specialized expertise, by leveraging generative AI and AWS services to proactively identify and mitigate weaknesses before they impact customers. The framework emphasizes shifting left by embedding continuous resilience validation into the development workflow, significantly reducing Mean Time to Resolution (MTTR) and improving system reliability.
Read original on AWS Architecture BlogThe Challenge of Proactive Resilience Testing
Many production system failures stem from unproven resilience, where undocumented dependencies or overlooked configuration changes create vulnerabilities. Traditional resilience testing is often slow, requires specialized expertise, and struggles to keep pace with continuous deployments and evolving distributed system architectures. This leads to critical weaknesses being discovered in production, resulting in revenue loss and eroded customer trust. The core problem is the significant gap between designed system intent and its actual runtime behavior under stress.
AI-Powered Resilience Framework Architecture
The proposed framework utilizes a five-layer architecture to automate and continuously validate system resilience, integrating various AWS services with custom AI agents. The layers are: Discovery, Test Generation, Experimentation, Gap Analysis, and Continuous Validation. This comprehensive approach aligns with the AWS Well-Architected Reliability Pillar, specifically the "Test reliability" best practice area.
- Discovery Layer: Automatically identifies infrastructure components and their dependencies. This is powered by AWS Resilience Hub's native dependency discovery, augmented by custom agents on Amazon Bedrock AgentCore for code-level analysis (e.g., hard-coded dependencies, timeout configurations, retry logic) across AWS APIs, CloudFormation templates, and code repositories.
- Test Generation Layer: Creates targeted chaos experiments based on the discovered architecture. Generative AI analyzes failure modes and suggests relevant experiments, reducing the need for manual design and specialized chaos engineering expertise.
- Experimentation Layer: Safely executes generated tests using AWS Fault Injection Service (FIS). This layer ensures controlled blast radius and provides mechanisms for rapid rollback.
- Gap Analysis Layer: Identifies resilience weaknesses, potential single points of failure, and areas for improvement based on experiment results. It feeds insights back to the discovery and test generation layers, creating a feedback loop for architectural improvements.
- Continuous Validation Layer: Integrates resilience testing into CI/CD pipelines, performing drift detection and providing dashboards for ongoing monitoring. This layer ensures that resilience improvements persist as systems evolve, shifting left to catch regressions before they impact production.
Key Enablers
The framework significantly leverages AWS Resilience Hub (for native dependency discovery and generative AI failure mode analysis), AWS Fault Injection Service (FIS) (for controlled experimentation), Amazon Bedrock AgentCore (for custom AI agent hosting and secure, scalable execution of discovery and analysis tasks), and AWS Systems Manager (for managing recovery procedures and automation). AWS Config is also used for tracking resource changes, ensuring the architecture map stays current.