Skip to content

guide-to-ai.md

Latest commit

 

History

History
148 lines (96 loc) · 4.33 KB

guide-to-ai.md

File metadata and controls

148 lines (96 loc) · 4.33 KB
Raw

🤖 AI Workflow Guide

Last updated: February 1, 2026

A practical, step-by-step framework for consistently getting high-quality, explainable results from AI.

Clarity • Architect • Execute • Audit • Improve • Add-Ons • Starter Prompt

Overview

Most people use AI reactively: ask a question, accept the answer, move on. This guide enforces discipline—forcing the model to think, clarify, and reason before acting. Use this process for any important task: writing, engineering, troubleshooting, strategy, architecture, creative work, or analysis.

1. Clarity Stage — Make the AI Think

Before producing any output, force the AI to clarify:

  • Assumptions: What must be assumed to proceed?
  • Constraints: Hard boundaries (tech stack, budget, time, rules, formatting, tone).
  • Missing Information: What would improve accuracy if provided?
  • Top 3 Risks: Where can this approach fail? What misunderstandings are likely? What breaks if requirements change?

Do not produce the final output yet.

Goal: Surface ambiguity before it becomes a hallucination or error.

2. Architect Stage — Design Before Building

Turn the clarified problem into a deliberate plan:

  • High-Level Architecture: Structure, components, dependencies
  • Step-by-Step Plan: Logical execution order
  • Decision Tree: If X → do A; if Y → do B
  • Contingency Plan: What changes if requirements shift?

Do not implement anything yet.

Goal: Lock in reasoning before creation.

3. Execute Stage — Build Only What Was Designed

Now — and only now — implement the solution exactly as designed.

  • Follow the architecture precisely
  • No improvisation or skipping steps
  • If something is unclear: pause and ask for clarification

Goal: Clean execution without hallucination or scope drift.

4. Audit Stage — Review Like a Senior Engineer

Switch to audit mode. Review the output as if doing a pull request:

  • Identify flaws or incorrect logic
  • Call out ambiguities
  • Find missing edge cases
  • Flag best practice violations
  • Recommend specific improvements

Do not rewrite the solution yet.

Goal: Separate creation from critique to catch mistakes.

5. Improvement Stage — Apply Fixes

Apply the audit feedback. Produce the improved, final version.

Goal: Intentional refinement, not random rewriting.

When to Use This Framework

Use this process when:

  • Accuracy matters
  • The task is complex
  • You need explainable reasoning
  • You want consistency across iterations
  • You’re delegating work to AI like a team member

Skip only when speed matters more than precision (quick chats, brainstorming, rewriting short text).

Add-Ons: Elevate Your Results

Enhance your workflow with these options:

🔍 Role Mode

Define the AI’s perspective: senior cloud architect, editor, investigator, policy analyst, attack-surface reviewer, etc.

📏 Quality Bar

Set the standard: “Enterprise-ready,” “Senior engineer level,” “NYT editorial quality,” “Clear enough for a junior to implement,” etc.

🛑 Refusal Rules

Tell the model what not to do:

  • Don’t hallucinate missing data
  • Don’t assume configurations
  • Don’t skip the clarity stage
  • Don’t produce code without validation

Copy-Paste Starter Prompt

You will follow this workflow:

1. **Clarity Stage** — assumptions, constraints, missing info, risks.
2. **Architect Stage** — structure, step-by-step plan, decision tree, contingencies.
3. **Execute Stage** — implement exactly to the plan; ask if unclear.
4. **Audit Stage** — critique the output (no rewriting).
5. **Improvement Stage** — apply fixes.

Do not skip stages. Do not guess. Ask when required.

Back to Student Resources