
ai-problem-framing
Turn vague intent into solvable structures with explicit objectives, constraints, success criteria,
Overview
AI Problem Framing is Layer 2 of AI fluency—the ability to transform fuzzy intent into structured problems that AI can solve well. This is the highest-leverage skill because AI amplifies your framing, not your intent.
Core Principle: AI doesn't solve problems—it amplifies the framing you give it.
Fluency Signal: Get high-quality output on the first iteration.
When to Use This Skill
- Before any significant AI delegation
- When AI output keeps missing the mark
- When iteration feels random rather than directional
- When defining AI-assisted workflows
- When training others on effective AI use
The Framing Framework
Every AI task requires three elements:
1. Explicit Objectives
What you actually want, not what you're asking for.
| Weak (Vague Intent) | Strong (Explicit Objective) |
|---|---|
| "Analyze this data" | "Identify the 3 metrics most correlated with customer churn" |
| "Write about X" | "Draft a 500-word explanation of X for non-technical readers" |
| "Help me think about Y" | "Generate 5 distinct strategic options for Y with trade-offs" |
| "Improve this code" | "Reduce the time complexity of this function from O(n²) to O(n log n)" |
Test: Would two different AIs produce similar outputs from this objective?
2. Constraints
Boundaries that shape the solution space.
Types of constraints:
- Format: Length, structure, output type
- Scope: What's in/out of bounds
- Style: Tone, voice, technical level
- Resources: What information to use (or not use)
- Quality: Standards that must be met
Example:
Objective: Write a product description
Constraints:
- Maximum 150 words
- Include price and 3 key features
- Match brand voice (casual, confident)
- Do not mention competitors
- Must include call-to-action
3. Success Criteria
How you'll know the output is good.
Success criteria should be:
- Specific: Measurable or evaluable
- Complete: Cover all important dimensions
- Prioritized: Which matter most if trade-offs needed
Example:
Success criteria (in priority order):
1. Factually accurate (all claims verifiable)
2. Addresses all 3 user questions
3. Under 500 words
4. Professional tone
5. Includes recommended next steps
Problem Decomposition
Breaking Work into AI-Appropriate Chunks
Large problems require decomposition:
- Identify sub-tasks - What distinct pieces of work exist?
- Sequence them - What depends on what?
- Assign ownership - Human, AI, or hybrid?
- Define interfaces - What moves between steps?
Decomposition Pattern
[Original Problem]
↓
[Sub-task 1: Research] → AI (good at synthesis)
↓
[Sub-task 2: Analysis] → Human (requires judgment)
↓
[Sub-task 3: Drafting] → AI (good at generation)
↓
[Sub-task 4: Review] → Human (requires accountability)
↓
[Sub-task 5: Refinement] → AI + Human (iteration)
Artifact Mapping
For each sub-task, define:
- Input artifact: What goes in
- Output artifact: What comes out
- Verification method: How to check quality
Human vs AI Responsibility
What Must Stay Human
- Final decisions: AI advises, humans decide
- Accountability: You own the output
- Value judgments: Ethics, priorities, trade-offs
- Verification: Checking against reality
- Context integration: Understanding full situation
What AI Does Well
- Pattern synthesis: Combining information
- Variation generation: Multiple options
- Format transformation: Restructuring content
- Search augmentation: Finding relevant information
- Draft creation: First-pass content
The Boundary Decision
For each task element, ask:
- Can AI do this reliably? (Capability)
- Can I verify the output? (Verifiability)
- Do I understand it enough to catch errors? (Expertise)
- What's the cost of AI error? (Risk)
If any answer is unfavorable, keep it human.
Practices
Problem Statement Template
Before delegating, complete:
## Problem Statement
**Objective:** [What I actually want to achieve]
**Context:** [Background AI needs to understand]
**Constraints:**
- [Constraint 1]
- [Constraint 2]
- [Constraint 3]
**Success Criteria:**
1. [Most important criterion]
2. [Second most important]
3. [Third most important]
**Out of Scope:** [What I don't want]
**Artifacts:**
- Input: [What I'm providing]
- Output: [What I expect back]
**Verification:** [How I'll check quality]
Task → Artifact Mapping
For complex work:
| Task | Input | Output | Owner | Verification |
|---|---|---|---|---|
| Research competitors | Industry list | Competitor profiles | AI | Spot-check 2-3 |
| Identify gaps | Profiles + our features | Gap analysis | Human | N/A (judgment) |
| Draft positioning | Gap analysis | Positioning options | AI | Review all options |
| Select positioning | Options | Decision + rationale | Human | N/A (decision) |
Constraint-First Framing
Instead of starting with what you want, start with what you don't want:
- List everything that would make output unacceptable
- Convert to positive constraints
- Prioritize constraints
- Then add objectives
This prevents scope creep and ensures critical requirements aren't missed.
Assessment Criteria
Layer 2 Complete When:
- Can write task specs that multiple AIs would handle consistently
- Distinguishes objectives from activities
- Specifies measurable success criteria before delegation
- Correctly assigns tasks to human vs AI ownership
- Has reduced iteration cycles through better framing
Common Framing Failures
Failure 1: Activity vs Outcome
Wrong: "Analyze customer feedback" Right: "Identify the top 3 complaints and their frequency in customer feedback"
Failure 2: Implicit Constraints
Wrong: "Write a blog post about AI" Right: "Write a 800-word blog post about AI for marketing professionals, avoiding jargon, in our brand voice (examples attached)"
Failure 3: Missing Success Criteria
Wrong: "Make this better" Right: "Improve clarity for non-technical readers; success = a 10th grader can understand the main point"
Failure 4: Unbounded Scope
Wrong: "Help me plan my project" Right: "Create a 10-item task list for the research phase of the project, with dependencies and estimated durations"
Related Skills
- ai-instruction-design — Translating framing into prompts
- ai-cognitive-readiness — Knowing when to frame vs not delegate
- ai-fluency-antipatterns — Vague Intent Delegation anti-pattern