AI Usage Guidelines for Network Engineers
Core Principles
Garbage In, Garbage Out (GIGO)
AI output quality depends entirely on the quality of the input.
| Do | Do Not |
|---|---|
| Provide clear, accurate, and complete prompts | Use vague or incomplete prompts |
| Include relevant technical context and constraints | Assume AI understands your environment automatically |
| Verify configuration examples before implementation | Paste sensitive customer or production data into public AI tools |
Writing Effective Prompts – SCOPE Framework
Use the SCOPE framework to improve AI response quality and consistency.
SCOPE
- S — Situation: Describe the environment, background, or technical context
- C — Character: Define the role the AI should act as
- O — Output: Specify the desired response format
- P — Purpose: Clearly explain the goal or task
- E — Examples: Provide examples of expected input/output where possible
Example Prompt
Situation: ISP hosting environment running Ubuntu 24.04 and NGINX
Character: Act as a senior Linux systems administrator
Output: Return a commented Bash script with a short explanation
Purpose: Create a disk monitoring script that alerts when usage exceeds 85%
Examples: Use syslog formatting similar to existing monitoring scripts
Prompt Best Practices
| Do | Do Not |
|---|---|
| Provide clear technical context | Use vague or incomplete prompts |
| Define the expected output format | Assume AI understands your infrastructure |
| Include examples where possible | Share passwords, API keys, or customer data |
| Specify versions, limitations, and constraints | Deploy AI-generated output without validation and testing |
Human in the Loop (MITL)
AI assists humans — it does not replace engineering judgement.
| Do | Do Not |
|---|---|
| Review all AI-generated configurations, scripts, and documentation | Allow AI-generated changes directly into production |
| Require human approval before production deployment | Treat AI responses as authoritative without verification |
| Validate outputs in staging or lab environments first | Use AI as the sole decision-maker for operational or security changes |
Approval & Change Control
All AI-assisted work must follow existing operational governance.
| Do | Do Not |
|---|---|
| Follow standard change management and CAB procedures | Bypass approval workflows because “AI suggested it” |
| Clearly identify when AI contributed to a deliverable | Deploy untested AI-generated automation |
| Maintain audibility for AI-assisted operational changes | Skip peer review for AI-created configurations or scripts |
Documentation Requirements
Prompt Documentation
Important prompts used for operational work should be documented.
Include:
- Purpose of the prompt
- AI tool used
- Date generated
- Reviewer/approver
- Final validated output
AI-Generated Documentation
Documents produced fully or partially by AI must be marked appropriately.
Security & Data Protection
Never Upload:
- Customer data
- Passwords, API keys, certificates, or secrets
- Internal IP schemes or sensitive network diagrams
- Confidential business information
Use only company-approved AI platforms for operational work.