How Intuit built an agentic disaster recovery assistant with Amazon Bedrock

AWS Machine Learning Blog · 人工智能

Disaster recovery (DR) at scale is hard. When thousands of microservices span multiple AWS Regions, coordinating a reliable failover becomes a major operational challenge. At Intuit , we operate at this scale. We support products that millions of people rely on to run their businesses and manage their finances. These include TurboTax , QuickBooks , Mailchimp , and Credit Karma . To close that gap, we built an agentic disaster recovery assistant with Amazon Bedrock . It builds on our existing centralized internal disaster recovery system called Ecosystem Wide Orchestrator Kit (EWOK) . EWOK standardizes failover execution across compute, databases, networking, caches, and asynchronous workloads. Service owners declare recovery intent in YAML , and the EWOK system orchestrates the underlying infrastructure actions, reducing recovery times from several hours to about 20 minutes for supported workloads. EWOK solved execution but not decision-making. Choosing which recovery workflow applies, and confirming an asset is ready, still relied on the tribal knowledge of experienced on-call engineers , as did handling the exceptions that surface mid-recovery. A common example arises during a change-freeze window. This is a period when deployments and changes are tightly restricted to protect service availability during critical business times such as tax season. If a failover request lands during one of these windows, it gets rejected. The engineer must know the detailed emergency-override procedure to proceed. To close this gap, we built EWOK Agent , an AI-powered agent built with Amazon Bedrock. Teams across Intuit have used it to run failovers for the past eight months. Amazon Bedrock gives us access to hundreds of foundation models from leading AI providers through a single API. We can evaluate and select the right model for failover reasoning, and switch models as our needs evolve, without rearchitecting the agent. It also provides built-in Amazon Bedrock Guardrails , along

查看原文