How Intuit built an agentic disaster recovery assistant with Amazon Bedrock
Amazon Web Services Suvojit Dasgupta
Intuit built an AI assistant to run disaster recovery across its systems on Amazon Bedrock. It cuts human guesswork in failovers, so the tricky part is now policy, not muscle memory.
Based on reporting by Amazon Web Services, Suvojit Dasgupta — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Disaster recovery gets ugly when you’re juggling thousands of microservices across multiple AWS Regions. Intuit says that was the reality behind TurboTax, QuickBooks, Mailchimp, and Credit Karma, and it pushed the company to build an agentic disaster recovery assistant on Amazon Bedrock.
The assistant sits on top of EWOK, Intuit’s internal disaster recovery system. EWOK already standardized failover across compute, databases, networking, caches, and asynchronous workloads, and it brought recovery times down from several hours to about 20 minutes for supported workloads. But it only solved the execution side. Picking the right workflow, checking whether an asset was actually ready, and handling surprises during recovery still depended on experienced on-call engineers.
That gap showed up most clearly during change-freeze windows, when deployments and other changes are tightly restricted, such as around tax season. If a failover request lands in one of those periods, it gets blocked. The engineer then needs to know the emergency-override process, which is the sort of thing that lives in people’s heads until it suddenly doesn’t.
EWOK Agent tries to turn that tribal knowledge into typed skills and a bounded agent loop. Engineers can ask in plain language, like “failover payments-gateway in production,” from Intuit’s Engineering portal or an IDE. The model figures out what to do; EWOK does the actual work. It resolves the asset, checks policy gates, opens a change record, runs the workflow, and reports status back through the same path. Intuit says teams have been using it for failovers over the past eight months.
Amazon Bedrock matters here because Intuit wants the model layer to stay thin. The company says Bedrock gives access to hundreds of foundation models through one API, plus Guardrails and security and privacy protections. Its data is not used to train models and stays encrypted in transit and at rest. Since Bedrock is fully managed, Intuit didn’t have to provision model infrastructure just to add another reasoning layer on top of EWOK.
My take — AI-written commentary, not fact-checked reporting
This is the sane way to use AI in operations: let the model choose, then keep the blast radius trapped inside deterministic code. The industry keeps trying to sell agents as mystical replacements for operators; Intuit is just using one as a very talkative control plane. That’s much less glamorous and a lot more useful.
Read more about this at: Amazon Web Services