Put a team of read-only AI ops agents between your GPU cluster and the on-call SRE: one chat asks Job Recovery, Hardware, Observability, Kubernetes and Cluster Management agents to correlate a training failure across layers and show the evidence before anyone acts: AI agent prompt
NVIDIA's Nov 4, 2026 technical blog describes the AI Factory Operations Agent it runs on its own DGX SuperPOD clusters. Instead of an SRE hopping between dashboards for compute, network, hardware and scheduling, a chat-facing agent routes to specialized sub-agents (Job Recovery, Hardware, Observability, Research, Kubernetes, Cluster Management) over MCP, correlates root causes across layers, and surfaces recommendations with transparent tool calls. The framework is built on the NemoClaw blueprint with OpenShell sandboxes and can run on-premises on Nemotron models. Production learnings stress clear agent boundaries, read-only investigation first, precise MCP tool definitions, and scoped permissions for multi-tenant factories. These are NVIDIA's own claims about its internal deployment and open developer example.
The prompt
Help me design a read-only multi-agent ops desk for our GPU or Kubernetes cluster, like an AI factory SRE helper. Our stack: [cluster type, e.g. Kubernetes + GPUs]. Signals we have: [Prometheus, logs, job scheduler, hardware alerts, docs]. Pain: [e.g. training jobs fail and on-call checks five dashboards]. 1) Propose 4-6 specialist agents with clear non-overlapping jobs (for example job recovery, hardware, observability, docs, Kubernetes, cluster management) and what each may query. 2) For each agent, list only read tools it should have and tools it must never have (no create/update/delete of pods, secrets or jobs). 3) Write the orchestrator rules: when a human asks in chat, which agent goes first, how agents hand evidence to each other, and when to stop and ask a person. 4) Draft one sample investigation for [describe a recent failure], with the questions each agent asks and the citations they must return (query result or doc section). 5) Give a trust rollout: week 1 chat-only recommendations, week 2 ticket drafts, never auto-remediate until SREs agree on accuracy. Don't invent metrics I didn't give you.
Fill in the blanks, then copy
This prompt has 4 [bracket]s to fill in. Type your details below and the prompt updates as you go. Your answers stay in this browser and are reused on other prompts with the same blanks. Nothing is sent to us.
Example values are fictional, just to show how a filled-in prompt reads. Replace them with your real details.How to use it
- Copy the prompt above.
- Replace every [bracket] with your own details (use the fill-in-the-blanks box above, which remembers your answers for the next prompt).
- Paste it into NVIDIA AI Factory Operations Agent / NemoClaw, MCP, OpenShell and review the answer before you use it.
Works with: NVIDIA AI Factory Operations Agent / NemoClaw, MCP, OpenShell, Prometheus, Grafana, Kubernetes, Claude or other chat agents
Real-world example: developer.nvidia.com/blog/how-nvidia-uses-ai-agents-to-streamline-ai-f
Before you use it: what to check
Already built into this prompt
- Don't invent metrics I didn't give you.
Check yourself
- This prompt works on text other people wrote (emails, reviews, messages or web pages). That text can hide instructions aimed at AI, such as "ignore your rules and…". Paste it as material to work on, and never let the AI act on requests inside it, like paying, sharing logins or changing records. How prompt injection works →
- Check names, dates, numbers and any claim against your own source material before you use the answer. AI can misread or invent details.
Using an AI agent? Download this prompt as a SKILL.md agent skill with these checks built in, or get the .zip for Claude.ai. How to install skills.
Want to make this prompt even better? Grade it with the free prompt grader or build your own with the AI prompt generator.
32 copy-paste agent prompts (PDF) plus 5 new prompts every week.
Get it free