How to Frame Problems Effectively for AI Collaboration - eirenicon/Ardens GitHub Wiki

How to Frame Problems Effectively for AI Collaboration

A guide for working with your favorite AI (Arthur, Khoj, and kin)

Purpose

This guide outlines how to define and present problems in a way that makes collaboration with AI systems more efficient, accurate, and generative. Clear framing accelerates understanding, reduces iteration cycles, and enhances the quality of co-produced work.

1. Begin with Context, Not Just the Question

Instead of: "Help me write a shell script."

Try this: "I’m working on a hardened Arch system with Btrfs snapshots. I need a shell script to rotate daily backups, clean up old ones, and alert me if something fails."

Why it helps: Context gives access to the right assumptions, constraints, and tone. Without it, responses may default to generic or mismatched solutions.

2. Specify the Desired Outcome and Format

Instead of: “Can you summarize this?”

Try this: “Can you rewrite this into a bullet-point brief for a policy team, aiming for clarity over detail?”

Helpful clarifiers:

  • What will you do with the output?
  • Who is the audience?
  • What format do you prefer? (Code, prose, outline, diagram, etc.)
  • Are there length constraints?

3. Share What You Already Know

Instead of: "Help me research fascism."

Try this: "I’ve already read Umberto Eco and Stanley, and I have a bibliography. What I need is help connecting those ideas to current U.S. dynamics using OSINT tools."

Why it helps: This avoids duplication and allows the AI to work as a thought partner, not just a tutor or encyclopedia.

4. Frame It as a Dialogue, Not a Transaction

Instead of: “Write X”

Try this: “Here’s my working draft of X — can you suggest improvements, find weaknesses, or reframe it for tone?”

Why it helps: AI collaborates best in an iterative loop. Starting messy is fine—refinement is part of the process.

5. Identify Constraints or Boundaries Up Front

Mention any limitations such as:

  • Technical: System resources, software limitations.
  • Ethical: Avoid manipulation, respect privacy.
  • Stylistic: Formal, poetic, minimalist.
  • Time-bound: "I need a first-pass summary in 10 minutes."

Why it helps: This avoids rework and ensures alignment from the start.

6. Signal Your Desired Level of Creativity

  • “Stay close to my outline” → Focused editing.
  • “Feel free to rethink the structure” → Creative reframing.

Both are valid. Just let the AI know your latitude for improvisation.

7. If the Goal is Unclear, Say So

Some of the most fruitful work begins with ambiguity.

  • “Something’s off in this essay, but I can’t name it.”
  • “This graph feels wrong, but I don’t know why.”

Why it helps: The AI can help unpack foggy problems too—diagnosis is part of the role.

8. Be Honest About Emotion or Stakes

If the topic is sensitive, high-stakes, or personally important, say so. The AI will tread more carefully.

  • “This is for a memorial.”
  • “This is a letter I’ve been afraid to send.”

Why it helps: Tone, care, and rhythm matter differently in such cases.

9. Use Iterative Check-ins for Large or Evolving Problems

For complex or multi-phase work (framework development, system architecture, longform writing), it helps to:

  • Define milestones.
  • Use version numbers or dates.
  • Keep a shared working doc or log (e.g., wiki or Markdown files).

Why it helps: This allows the AI to track continuity and build momentum with you.

10. Clarify the Role You Want the AI to Play

Do you need a:

  • Researcher?
  • Editor?
  • Devil’s advocate?
  • Technical assistant?
  • Co-author?
  • Sounding board?

You can assign multiple roles, or shift them as the work evolves. Just say what you need right now.

Final Notes

The more clearly problems are defined, the more sharply the AI can focus—and the more useful it can be. But even if you’re not sure how to define the problem yet, starting the conversation helps.

Let us begin where we are.

Category: Processes & Methods