Amazon EKS · 2025

AI troubleshooting

Enabling Amazon EKS users to easily troubleshoot in production environments

This project lets EKS users troubleshoot their production environments easily and on their own, without having to open a support ticket. Tickets can be time-consuming, and the quality of the answer isn’t guaranteed.

Timeline: 2024-2025
Role: Product design
Industry: Cloud computing
Disclaimer: Please note that because of an NDA, I can only share more details in a private setting.

The EKS console on a laptop: the Cluster insights tab, where an upgrade warning opens a popover with an Inspect with Amazon Q button

Context

The engineering team was working on an MCP server for the product, and it needed a way for our users to make use of it.

The problem

Our users could access the MCP server programmatically, but they didn’t have an intuitive way to use it.

What I did

User pain points

I started by synthesizing all the pain points from past user research, and quickly pinpointed the most crucial touch points where users could use the MCP server to troubleshoot.

Evolution of Define and ideate, a timeline: early exploratory designs for troubleshooting with GenAI in the EKS console in 2023; cross-functional workshops in August and October 2024, which identified 11 key customer problems; an in-person ideation session in November 2024; then a synthesis that revealed three themes focused on observability and recommendation

Design thinking activity artifacts, August to November 2024: a workshop board of how-might-we prompts and sticky notes across the cluster lifecycle, with the prioritized HMWs and the top 11 problems grouped by theme, beside a grid of hand-sketched concept sheets from the ideation session

The Network tab's service map: a pod's metrics popover has an Inspect with Amazon Q button

Two more ways in, circled: the Inspect with Amazon Q button in an upgrade warning’s popover, and the Inspect with Amazon Q column in the health issues table

Outcome

From the outset, user adoption has been a huge success, and the company is still seeing usage grow every month.

4,500

Clicks per week

+2%

Monthly growth in usage

Conclusion

What I’d do differently

In addition to letting users troubleshoot on their own, the system could access the MCP server automatically and troubleshoot for them. That would be a much more powerful feature, one that could drastically increase user adoption and retention for the product, as long as the results are accurate and helpful.

Let’s connect

For roles, collaborations, or just to say hi.

Email LinkedIn