Appendix
This appendix provides reusable reference materials for the AI Systems guide.
The main chapters explain the ideas.
This appendix turns those ideas into practical tools.
Use these templates to design AI workflows, review AI outputs, document AI use, and build reusable AI agents that remain human-led.
Appendix A: Human–AI–Human Workflow Checklist
Use this checklist before relying on an AI-supported output.
Human--AI--Human Workflow Checklist
1. Human framing
[ ] The problem or question is clearly stated.
[ ] The intended audience is known.
[ ] The purpose of the output is clear.
[ ] The risk level of the task is understood.
2. Context and grounding
[ ] Relevant background information has been provided.
[ ] Key data, documents, examples, or references have been included where needed.
[ ] Important constraints are stated.
[ ] Unavailable information is acknowledged.
3. AI assistance
[ ] The AI task is specific.
[ ] The expected output format is clear.
[ ] The AI is not being asked to make final human decisions.
[ ] Sensitive, private, or unsafe information is handled carefully.
4. Human evaluation
[ ] The output has been checked for correctness.
[ ] The output has been checked for completeness.
[ ] The output has been checked for bias, risk, or misleading claims.
[ ] Claims have been verified where needed.
5. Human decision and documentation
[ ] The final decision remains human-led.
[ ] AI use has been documented where appropriate.
[ ] Limitations are communicated honestly.
[ ] The output is saved, revised, or rejected based on review.Appendix B: AI Task Framing Template
Use this template before starting an AI-supported task.
AI Task Framing Template
Task title:
Purpose:
Who will use the output?
What decision, action, or learning goal should this support?
Background context:
Input materials provided:
Constraints:
Expected output format:
Quality criteria:
Known risks:
What the AI should not do:
Human review steps:
Final owner of the decision:A task that is poorly framed usually produces weak output.
A task that is clearly framed gives the AI system a better chance of producing something useful and reviewable.
Appendix C: Prompt Design Template
A good prompt is not only a question.
It is a small workflow specification.
Prompt Design Template
Role:
Act as ...
Goal:
Help me ...
Context:
Here is the situation ...
Inputs:
Use the following information ...
Constraints:
Do not ...
Prioritize ...
Assume ...
Process:
First ...
Then ...
Finally ...
Output format:
Return the answer as ...
Review requirement:
Before finalizing, check for ...This structure helps move prompting away from guessing and toward system design.
Appendix D: Context Package Template
Context packages are useful when AI needs to work with documents, data, examples, or project-specific information.
Context Package Template
Project name:
Task:
Relevant background:
Audience:
Source material included:
Important definitions:
Important examples:
Data notes:
Style or formatting requirements:
Things to avoid:
Uncertainties:
Expected output:A context package should help the AI stay grounded.
It should also help the human reviewer understand what the output was based on.
Appendix E: AI Output Evaluation Rubric
Use this rubric to review AI outputs before using them.
| Criterion | Low quality | Acceptable | Strong |
|---|---|---|---|
| Correctness | Contains clear errors or unsupported claims | Mostly correct but needs review | Accurate, checked, and well-supported |
| Relevance | Does not answer the task | Partly answers the task | Directly addresses the task |
| Completeness | Missing important pieces | Covers the basics | Covers key details without unnecessary excess |
| Clarity | Confusing or poorly structured | Understandable | Clear, organized, and easy to use |
| Evidence | No grounding or unclear sources | Some grounding | Claims are traceable to appropriate evidence |
| Risk awareness | Ignores possible harm or limitations | Mentions some limitations | Clearly identifies limits, uncertainty, and risk |
| Human usability | Hard to act on | Usable with edits | Ready for human review and adaptation |
The goal is not to make every AI output perfect.
The goal is to decide whether the output is safe, useful, and appropriate for the task.
Appendix F: AI Use Documentation Template
Use this when documenting AI support in learning, research, analysis, writing, or professional work.
AI Use Documentation Template
Project or task:
AI tool used:
Date of use:
Purpose of AI support:
Input materials provided to AI:
Type of output generated:
Human review performed:
Corrections or changes made by human reviewer:
Limitations identified:
Final decision owner:
Notes for reproducibility:Documentation does not have to be complicated.
It should make the workflow understandable and defensible.
Appendix G: Reusable AI Agent Specification Template
Use this template to define a reusable AI agent before building or using it.
Reusable AI Agent Specification
Agent name:
Agent purpose:
Primary users:
Supported tasks:
Tasks outside scope:
Required inputs:
Optional inputs:
Context sources:
Memory rules:
Workflow steps:
Decision logic:
Escalation conditions:
Output formats:
Review requirements:
Quality standards:
Safety boundaries:
Maintenance owner:
Version:
Last reviewed:An agent without scope can become unreliable.
An agent with clear scope can become a repeatable workflow component.
Appendix H: Agent Output Review Checklist
Before accepting an agent output, review it against this checklist.
Agent Output Review Checklist
[ ] The output matches the agent purpose.
[ ] The output stays within the agent scope.
[ ] The output uses the provided context appropriately.
[ ] The output does not invent unsupported information.
[ ] The output follows the required format.
[ ] The output is understandable to the intended audience.
[ ] The output identifies uncertainty where needed.
[ ] The output avoids unsafe or inappropriate recommendations.
[ ] The output is ready for human editing, approval, or rejection.
[ ] The final human decision is documented.Agent review is not a sign of failure.
It is part of responsible agent design.
Appendix I: AI Workflow Log Template
A workflow log helps teams and learners track how AI was used.
AI Workflow Log
Date:
Task:
Human owner:
AI role:
Inputs used:
Prompt or workflow summary:
Output received:
Review findings:
Changes made:
Final outcome:
Lessons learned:Over time, workflow logs help improve prompts, templates, review standards, and agent behavior.
Appendix J: Risk Level Guide
Not all AI tasks carry the same level of risk.
Use this simple guide to decide how much review is needed.
| Risk level | Examples | Review requirement |
|---|---|---|
| Low | Brainstorming titles, outlining notes, summarizing personal study ideas | Light human review |
| Medium | Drafting reports, interpreting non-sensitive data, preparing public communication | Careful review and editing |
| High | Medical, legal, financial, employment, academic grading, safety-critical, or sensitive personal decisions | Expert review, source verification, and strong human oversight |
High-risk tasks should never be treated as fully automated AI tasks.
They require human expertise, documentation, and careful judgment.
Appendix K: Example Human-Led AI Workflow
The following example shows how a responsible AI workflow can be documented.
Task:
Prepare a short explanation of a data analysis result for a non-technical audience.
Human framing:
The analyst identifies the result, audience, purpose, and limitations.
Context provided:
Clean summary table, chart, project goal, and notes about what the analysis can and cannot claim.
AI assistance:
The AI drafts a plain-language explanation.
Human review:
The analyst checks whether the explanation matches the data, removes overclaims, and adds limitations.
Final output:
A revised explanation that can be shared in a report.
Documentation:
The analyst records that AI was used for drafting language, not for generating the statistical result or final interpretation.This pattern can be adapted to research, learning, communication, coding, mentoring, and portfolio development.
Appendix L: CDI AI Systems Principle Summary
The AI Systems guide is built around a few core principles.
1. AI should assist human thinking, not replace human responsibility.
2. AI work should begin with a clear human-framed problem.
3. Context improves output quality.
4. Grounding improves reliability.
5. Evaluation is part of the system, not an afterthought.
6. High-risk work requires stronger review.
7. Reusable agents need clear purpose, scope, inputs, outputs, and maintenance.
8. AI-supported work should be documented when it influences decisions or public outputs.
9. Human judgment remains central.
10. Responsible AI practice is learned through repeatable workflows.These principles connect all four parts of the guide:
AI Foundations
↓
AI Systems
↓
Reusable AI Agent Template
↓
Applied AI AgentsClosing Note
Appendices are not just extra material.
In a systems guide, they are part of the operating infrastructure.
They help learners and teams move from ideas to repeatable practice.
Use these templates, adapt them, improve them, and keep them connected to human judgment.
The next section provides references and source material for further reading.