How Intuit built an agentic disaster recovery assistant with Amazon Bedrock
Intuit engineered EWOK Agent, an agentic disaster‑recovery assistant that runs on Amazon Bedrock. The system lets on‑call engineers trigger production failovers with plain‑language commands, while automatically logging every action, enforcing policy constraints, and ensuring safety. EWOK Agent is built as a Bedrock agent that orchestrates underlying recovery scripts and services, providing a unified interface for complex, multi‑step recovery workflows. The design trades off some latency for comprehensive auditability and policy compliance, making it suitable for high‑stakes production environments.
⚡ Key Takeaways
- EWOK Agent is deployed on Amazon Bedrock and accepts natural‑language requests to initiate failover procedures.
- The architecture uses Bedrock’s agent framework to chain multiple recovery actions while enforcing policy rules and audit logging.
- The system prioritizes auditability and policy compliance over minimal latency, which can increase response time during a disaster scenario.
- Engineers integrate EWOK by invoking the Bedrock agent API endpoint and supplying a plain‑language prompt that describes the desired failover.
- The solution requires pre‑defined policies and audit hooks; without these, the agent cannot execute actions or guarantee compliance.
- WhyItMatters: By embedding policy enforcement and audit trails directly into the agent, production teams can safely automate complex disaster‑recovery tasks, reducing manual error and accelerating recovery times.
- TechnicalLevel: Intermediate
- TargetAudience: ML Engineers, DevOps Engineers
- PracticalSteps:
- Call the Bedrock agent API with a plain‑language prompt describing the desired failover (e.g., “Trigger a failover for service X to region Y”).
- Configure the agent’s policy set to define allowed actions and audit logging parameters.
- Verify that audit logs are captured in your monitoring system before executing a real failover.
- ToolsMentioned: Amazon Bedrock
- Tags: AGENTS, DEPLOYMENT, INFERENCE, AMAZON
🔧 Tools & Libraries
By embedding policy enforcement and audit trails directly into the agent, production teams can safely automate complex disaster‑recovery tasks, reducing manual error and accelerating recovery times.
✅ Practical Steps
- Call the Bedrock agent API with a plain‑language prompt describing the desired failover (e.g., “Trigger a failover for service X to region Y”).
- Configure the agent’s policy set to define allowed actions and audit logging parameters.
- Verify that audit logs are captured in your monitoring system before executing a real failover.
Want the full story? Read the original article.
Read on AWS ML Blog ↗