- What: Intuit developed EWOK Agent, an AI-powered assistant for automated disaster recovery.
- Technology: Built using Amazon Bedrock foundation models, the Converse API, and Bedrock Guardrails.
- Performance: Reduces service recovery times from several hours to approximately 20 minutes.
- Deployment: The tool has been active across Intuit’s engineering teams for the past eight months.
- Impact: Automates complex failover decisions for critical services including TurboTax and QuickBooks.
Intuit has successfully integrated generative AI into its core infrastructure operations by building an "agentic" disaster recovery assistant using Amazon Bedrock. The tool, known as EWOK Agent, was designed to manage large-scale service failovers across thousands of microservices that support millions of customers using products such as TurboTax, QuickBooks, Mailchimp, and Credit Karma. By shifting from manual, engineer-led decision-making to an AI-driven model, Intuit has reportedly reduced recovery times for supported workloads from several hours to roughly 20 minutes.
Bridging the Gap Between Execution and Reasoning
The development of the EWOK Agent marks an evolution of Intuit’s existing disaster recovery infrastructure. For several years, the company relied on its "Ecosystem Wide Orchestrator Kit" (EWOK), a centralized system that standardizes failover execution across databases, networking, compute, and caches. While the original EWOK system successfully handled the technical execution of infrastructure changes via YAML-based intent declarations, it still required human intervention for the critical "reasoning" phase of a disaster.
According to a case study published by Amazon, Intuit’s disaster recovery process previously relied on the "tribal knowledge" of experienced on-call engineers. These engineers were responsible for selecting the correct recovery workflows and navigating complex operational exceptions, such as "change-freeze" windows during peak business periods like tax season. During these windows, standard deployments are restricted, and engineers must know specific emergency-override procedures to proceed with a failover. The EWOK Agent was built to codify this expertise and automate these high-pressure decisions.
Technical Architecture and Amazon Bedrock Integration
To build the EWOK Agent, Intuit utilized Amazon Bedrock, a fully managed service that provides access to a variety of foundation models (FMs) through a single API. This architecture allows Intuit to evaluate and switch between different models as their needs evolve without requiring a total rearchitecture of the agent.
The agent operates through a "thin layer" that connects foundation models to specific "skills"—encoded representations of failover knowledge. According to the technical documentation provided by Amazon, the system leverages the Amazon Bedrock Converse API and its tool-use (function calling) capabilities. This setup allows the AI to translate a plain-language request from an engineer—such as "failover payments-gateway in production"—into a series of validated, policy-aligned recovery steps.
A core design principle of the system is the "crisp boundary" between the AI's reasoning and the system's execution. As stated in the AWS blog post, "The model decides what to do, and the EWOK Agent deterministically executes how." This ensures that while the AI handles the complex logic of which assets to move and when, the actual infrastructure changes are performed by a predictable, auditable execution engine.
Security and Operational Guardrails
Security was a primary concern for Intuit, given that the agent operates on production financial systems. The company implemented Amazon Bedrock Guardrails to ensure safety and policy compliance. Furthermore, the integration utilizes standard AWS security protocols, ensuring that Intuit’s data is not used to train the underlying foundation models and remains encrypted both in transit and at rest.
The EWOK Agent is delivered to Intuit’s staff as a plugin, allowing engineers to trigger failovers directly from the company’s internal engineering portal or their preferred integrated development environment (IDE). This accessibility, combined with the AI’s ability to handle exceptions autonomously, reduces the cognitive load on engineers during critical outages.
Industry Impact
The deployment of the EWOK Agent demonstrates a significant shift in how large-scale tech enterprises manage "high-stakes" infrastructure. By moving disaster recovery from a manual, "tribal knowledge" model to an agentic AI model, Intuit has addressed a common bottleneck in cloud operations: the speed of human decision-making during a crisis.
For the broader industry, this implementation serves as a blueprint for how generative AI can be applied to "deterministic" systems. Rather than using AI to write code or generate text, Intuit is using it as a reasoning engine to orchestrate existing, reliable automation. This approach mitigates the risks associated with AI hallucinations by strictly limiting the agent's actions to a predefined set of "skills" and authenticated APIs.
What’s Next
Intuit’s engineering teams have already been using the EWOK Agent for eight months to manage failovers. While the current implementation is specific to Intuit's internal EWOK system, the design pattern—using a bounded agentic loop over a deterministic executor—is being promoted by Amazon as a reusable architecture for other organizations using Amazon Bedrock. No specific timeline for new features or public release of the underlying EWOK code was provided.