The selected option specifically A states "Kubernetes provides autoscaling, auto-restarts, dynamic task scheduling, error isolation with containers, and integrated monitoring.", which matches the operational requirement rather than a superficial wording match. Kubernetes is better for long-running AI services because it supplies restart, scheduling, monitoring, and autoscaling primitives. Slurm remains strong for batch
/HPC jobs. Option A wins because it optimizes the system boundary around the risky component rather than hoping the base model behaves consistently. The NVIDIA implementation angle is not cosmetic here: NIM microservices and the NIM Operator fit Kubernetes production operations; Triton provides serving primitives and Prometheus-exportable inference metrics for GPUs and models. The durable control mechanism is independent scaling of agent components so embeddings, reranking, reasoning, and guardrails do not share one rigid capacity pool. That is why the other options are traps: CPU-only or memory-only scaling signals rarely capture the saturation profile of GPU-backed LLM inference. For certification purposes, read the question as asking for controlled autonomy, not raw LLM creativity.