Chad Rolinski
Amazon (AWS)

Resilience Hub

Defined strategy and integrated AI seamlessly to make complex setup, analysis, and change easier for large companies in regulated industries.
AI-Powered Vision and Strategy Regulated Industries
RoleProduct Design Manager
Team2 Designers, 2 Principal PMs, 10 Engineers, 2 SDMs
Key outcomeReplaced three disjointed services with one connected path from understanding a system to testing it. Regulated industry customers adopted within the first month of launch.
AWS Resilience Hub, next generation. Proactively improve the resilience of your applications.

Context

Background

  • In regulated industries such as financial services, healthcare, and transportation, customers require resilient architecture to withstand major disruptions.
  • AWS helps these industries with three resilience services: Fault Injection Service to stress test the system, Resilience Hub to onboard and pinpoint resilience issues, and Application Recovery Controller to recover during an event. The services originally launched in 2021.
  • I took on design leadership for all three services in Q1 2025, with two designers joining my organization.

Customer problem

Three services for a customer journeyUnderstanding a system, testing it, and recovering it were split across three services with no clean path between them.
Architecture is complex and hard to understandCustomers pieced their architecture together from tables and resource screens, and nothing checked that picture against what was actually running.
Resilience testing is difficult to setupCurrent resilience testing was overwhelming and took too many steps to setup: create a template, create an experiment, add targets and metrics.

Business need

  • Resilient customers mean fewer outages and fewer headlines. Under the shared responsibility model, news of a disruption reflects on AWS whatever the cause.
  • The product also helps Amazon's own services, including AWS core infrastructure, keep high availability uptime.
  • Fewer outages mean more customer trust and more spend on AWS.

Resilient systems keep businesses running. In regulated industries such as financial services, healthcare, and transportation, customers must also prove to auditors that their architecture can withstand disruption. AWS serves them with three resilience services: Fault Injection Service to stress test a system, Application Recovery Controller to recover it, and Resilience Hub to tie the journey together.

The three launched in 2021 under two different leadership teams. I took on design leadership for all of them in 2025, with two designers joining my organization. Our research and journey mapping showed customers struggling to onboard, go deep in any one service, and connect the three to finish their work. With UX leading the design, we launched the next generation of AWS Resilience Hub in May 2026, followed by Recommended Resilience Tests in August 2026.

Launched May 2026With UX leading the design, we launched the next generation of AWS Resilience Hub in May 2026, followed by Recommended Resilience Tests in August 2026.
Getting startedCustomers choose to move to the next generation experience or continue with the original Resilience Hub, so the new unified service could roll out without disrupting anyone already relying on the old workflow.
AWS Resilience Hub next generation get started screen, with options to manage, assess, test, and report on resilience
Service topologyA full resource map, built automatically. Research showed customers struggled to understand their system from a series of disconnected tables. We redesigned that into a topology map customers can explore and check against their own mental model of their architecture.
AWS Resilience Hub service topology screen in dark mode, showing a resource map and dependencies for a sample banking application
Service functionsAI-generated groupings customers can refine. AI groups a customer's resources into service functions, the logical pieces of their application, like frontend delivery or payment processing. Customers can edit those groupings to make sure they reflect their service accurately.
AWS Resilience Hub service detail screen, showing AI-generated service functions for a sample banking application
Service Failure ModeAI-generated findings, ranked by severity. AI evaluates a customer's architecture against AWS Well-Architected best practices and surfaces specific failure modes. Customers review and act on each one with a clear view of where their system is exposed.
AWS Resilience Hub failure modes screen, listing AI-generated findings ranked by severity for a sample banking application

Challenge & Opportunity

Three services, one customer job

  • Fault Injection Service, Application Recovery Controller, and Resilience Hub shipped independently, though they served one job: understand a system, test it, and fix what breaks.
  • Every hand off cost context, a real risk for a customer preparing for an audit or running a system where downtime has consequences.

Large scale architecture was hard to see

  • Customers assembled the full picture of their system in their heads from tables and linked resource screens.
  • Nothing checked whether the declared picture still matched what was running, and many teams changed their own services independently.

Getting started with testing took too many steps

  • A customer built a template, deployed an experiment by hand, picked targets, then set up metrics separately to judge the result.
  • For anyone new to resilience testing, the path was long before it produced a first answer.

Three services, one customer job

Fault Injection Service, Application Recovery Controller, and Resilience Hub shipped independently, though they served one job: understand a system, test it, and fix what breaks. The Hub covered architecture, assessment, and policy, but had no clean bridge into testing, and recovery stood apart from both. Every hand off cost context, a real risk for a customer preparing for an audit or running a system where downtime has consequences.

Large scale architecture was hard to see

Customers worked through tables and linked resource screens and had to assemble the full picture of their system in their heads. The Hub could only reason about resources a customer explicitly imported, and nothing checked whether that declared picture still matched what was running. That gap is where resilience problems hide, and it widened constantly as many teams changed their own services independently.

Getting started with testing took too many steps

To run a test, a customer built a template, deployed an experiment by hand, picked targets from an application they had already defined in the Hub, then set up metrics separately to judge the result. For anyone new to resilience testing, the path was long before it produced a first answer.

Failure modesEvery finding across the entire system in one place. Findings roll up across all of a customer's systems and services into a single queue, so a team can triage by severity and category across every service at once.
AWS Resilience Hub failure modes list, showing findings across multiple systems and services with severity and status
Failure detailReasoning and recommendations for every finding. Each finding explains why it matters, in the context of a customer's own architecture, and comes with AI-generated recommendations for mitigating it, watching for it, and testing against it.
AWS Resilience Hub failure detail screen, showing a single point of failure finding with reasoning and AI-generated mitigation recommendations
Failure mode guidanceAssertions to steer AI Analysis. Customers can see and edit the assumptions driving each assessment, agent inferred or their own, so the AI grounds its findings in how the system actually behaves. They can also add service design files, giving the AI additional context to work from.
AWS Resilience Hub failure mode guidance screen, listing agent-inferred and user-added assertions that guide the assessment

Action

Set the vision for one service built around the customer's journey

  • I defined the strategy to rebuild Resilience Hub as one service that guides customers from start to finish, uniting work split across two leadership teams.
  • A System now reflects a customer's actual user journey, and each Service is a piece of that journey that could break on its own.

Pivoted onboarding from a linear chat to AI built into the experience

  • The first direction for onboarding was a linear chat experience. I pushed for a pivot to integrating AI seamlessly where it is more accurate, inside the workflow itself.

Made testing a one step start

  • Designed for prebuilt tests the product generates, targets, runs, and scores automatically, replacing the hand built template, experiment, and metrics workflow. We limited the set to four major tests to reduce decision paralysis.

Pushed for a visual topology of resources

  • I pushed for this, leveraging our success on another project where we reduced support tickets. The map provides a dynamic and auto generated view of a customer's resources to check against their own mental model in one view.

Integrated AI where it is most effective

  • A multi agent system analyzes a customer's topology against their policies, business context, and AWS best practices, then produces ranked findings and recommendations.
  • Customers can adjust AI findings by tailoring assumptions or adding system context via docs and files to improve analysis and findings.

Built in dependency detection

  • The product watches for changes in a customer's architecture and includes new dependencies in tests automatically.

Set the vision for one service built around the customer's journey

I defined the strategy to rebuild Resilience Hub as a single service that guides customers from start to finish, uniting work that had been split across two leadership teams. We redefined how a system is described: a System reflects a customer's actual user journey, and each Service is a piece of that journey that could break on its own. The product's structure follows the journey phases we found in research.

Pivoted onboarding from a linear chat to AI built into the experience

The first direction for onboarding was a linear chat experience. I pushed for a pivot to integrating AI seamlessly into the product at the points where it is more accurate, such as grouping resources into service functions and generating failure modes, so customers get its value inside the workflow they are already in.

Made testing a one step start

We replaced the hand built template, experiment, and metrics workflow with prebuilt tests the product generates, targets against a customer's discovered resources, runs, and scores automatically. We limited the set to four major tests to reduce decision paralysis.

Pushed for a visual topology of resources

I pushed for this. Research across several services in my org pointed to the same need: customers wanted to navigate their system visually. The design lead and I carried forward an earlier resource map we had shipped for the Load Balancing service, which had reduced support tickets for a top customer issue. The Hub's map is generated automatically from a customer's real resources, so they can see how services connect and check it against their own mental model in one view.

Integrated AI where it is most effective

We built a multi agent system that analyzes a customer's topology against their policies, business context files, and AWS best practices, then produces ranked findings and recommendations. Customers can adjust AI findings by tailoring assumptions or adding system context via docs and files to improve analysis and findings.

Built in dependency detection

The product watches for changes in a customer's architecture, surfaces them, and includes new dependencies in tests automatically. Customers mark each as a soft or hard dependency, which feeds the AI analysis.

Impact

Regulated industry customers adopted early

  • Key customers in financial services, transportation, government, and healthcare migrated to the new Resilience Hub during the first month of launch.

Found a risk two years of reviews had missed

  • An engineer from DoiT supporting an enterprise customer ran a failure mode assessment on a production system the team considered well architected. It surfaced an undocumented third party payment dependency that two years of reviews had missed. Read the full account.

Rearchitected Resilience Hub is now scalable

  • The rearchitected Next Generation Resilience Hub IA is designed to add mitigation in the future for full coverage of the customer journey in one service.

Regulated industry customers adopted early

Customers in financial services, transportation, government, and healthcare began adopting the Next Generation Resilience Hub within the first month of launch. The simplified path to testing is a key part of what is accelerating adoption.

Found a risk two years of reviews had missed

In one public example, a DoiT engineer supporting an enterprise customer ran a failure mode assessment against a production system the team already considered well architected. It surfaced an undocumented dependency on a third party payment provider that two years of manual reviews had missed, and the team fixed it before it caused a problem. Read the full account: The new AWS Resilience Hub found a failure mode missed for two years.

Rearchitected Resilience Hub is now scalable

The rearchitected Next Generation Resilience Hub IA is designed to add mitigation in the future for full coverage of the customer journey in one service.

Early adoption. Regulated industry customers in financial services, transportation, government, and healthcare adopted within the first month of launch.
One step testing. Customers pick a prebuilt test and run it against their own architecture.
Hidden risk discovery. A DoiT engineer supporting an enterprise customer found a third party dependency that two years of manual reviews had missed.
One connected journey. Understanding a system, assessing it, and testing it now live in a single service.

Let’s chat

Happy to connect, whether it’s a coffee chat or collaborating on an interesting design problem.

LinkedIn