Can we entrust the AI agent with the K8s operation?
The agent that was helping me write the code is now running kubectl directly and even started the primary response to the failure alarm. So, is it okay to leave the K8s operations to him? Claude Code·Gemini CLI·Codex CLI's 9 model combinations were put into the same 10 K8s failures. All three tools were usable in practice, but the most expensive model always did better in operation...