Blog
LLM agents, AI SRE, and Kubernetes operations.
Technical writing from the team behind KubeIntellect — grounded in a peer-reviewed framework, not hype.
A Kubernetes AI agent should never say 'healthy' about pods it couldn't see
A one-line bug could make a truncated cluster listing look complete: on a 401-pod cluster, one CrashLoopBackOff sorted past an 8,000-character cap and the agent reported '85 pods, no issues.' The cause, why green tests missed it, and the fix.
Read →KubeIntellect is peer-reviewed and open access — read the paper
The KubeIntellect framework is published in the Journal of Grid Computing (Springer, 2026) and is open access. 93% query resolution on a live cluster, +25 points over a raw GPT-4o baseline. Links to the DOI, ACM, and arXiv.
Read →How to choose an AI tool for Kubernetes: the capabilities that matter
A capability-first guide to evaluating any AI tool for Kubernetes — the difference between look, explain, reason, and act — and where to draw the safety line.
Read →When an AI agent hits a wall, it should write its own tool
How the KubeIntellect agent synthesizes, validates, and registers new Kubernetes tools at runtime — and the guardrails that make it safe. 81.8% success on novel requests, from the published evaluation.
Read →Why more AI agents is not more capability
An honest engineering lesson from building an LLM operator for Kubernetes: going from the 13 specialized agents of the published v1 design to a lean coordinator made it better and cheaper. More agents is not more capability.
Read →