WIP: natural language troubleshooting

original id: 36122830

Kubernetes is still too complex for many users. In AKS, even basic tasks like troubleshooting or onboarding often require switching between the Portal, CLI, and kubectl. This fragmentation slows teams down and increases the risk of errors. Per the Managed AKS AI Troubleshooting in CLI PRD, “The large surface area with Kubernetes - storage, add-ons, 3rd party tools, and Control plane exacerbates the problem with a potentially vast amount of telemetry chasing the “needle in haystack”. With a sprawl of tools, they are ultimately forced to create support cases to get to the cause or churn if they cannot resolve it in a timely manner.”

New users face steep learning curves, while experienced engineers still encounter friction when diagnosing issues across namespaces or interpreting logs. AI tools offer potential relief, but most are either too rigid, too generic, or disconnected from the actual cluster context.

Many AI integrations require users to bring their own models and configure endpoints manually. They often lack scoped awareness, leading to broad and inefficient analysis that doesn’t align with the user’s intent or environment.

What users need is a smarter, more approachable assistant that understands their cluster, operates within the right scope, and helps them take action confidently. The motivation behind this work is to reduce friction, build trust, and help users move from problem to resolution faster withing a single interface.