I’ve been exploring how far we can push fully autonomous, multi-agent investigations in real SRE environments — not as a theoretical exercise, but using actual Kubernetes clusters and real tooling. Each agent in this experiment operated inside a sandboxed environment with access to Kubernetes MCP for live cluster inspection and GitHub MCP to analyze code changes and even create remediation pull requests.