A field guide to Claude Fable 5: Finding your unknowns

• Introduces "map vs territory" mental model: your prompts are the map, the codebase is the territory, and unknowns are the gap between them
• Categorizes unknowns into 4 types: known knowns (in your prompt), known unknowns (aware you don't know), unknown knowns (too obvious to write down), unknown unknowns (haven't considered)
• Provides specific techniques for each phase: blind spot passes and brainstorms (pre-implementation), implementation notes (during), quizzes and explainers (post)
• The balance problem: too specific and Claude can't pivot when needed; too vague and Claude makes wrong assumptions based on "best practices"
• Real example: author used these techniques to edit Fable's launch video end-to-end despite being a video editing novice

The core insight is that Fable represents a fundamental shift: model quality is no longer the bottleneck—your ability to clarify unknowns is. The author introduces a "map vs territory" framework where your prompts and context are the map, the actual codebase and constraints are the territory, and unknowns are the dangerous gap between them. The more ambitious the work, the more unknowns Claude will encounter, and each unknown forces Claude to guess what you want.

The author breaks unknowns into four categories: known knowns (what's in your prompt), known unknowns (what you know you haven't figured out), unknown knowns (things so obvious you'd never write them down but would recognize when you see them), and unknown unknowns (what you haven't considered at all). The best agentic coders have few unknowns because they're deeply in-sync with both the codebase and model behavior—but this is a learnable skill.

The article provides concrete techniques for each phase of work. Pre-implementation: use "blind spot passes" to surface unknown unknowns, brainstorm multiple approaches to identify unknown knowns (especially for visual work), have Claude interview you about ambiguities, provide source code as references (richer than screenshots), and create implementation plans that surface likely decision points. During implementation: maintain an implementation-notes.md file where Claude logs deviations from the plan. Post-implementation: create pitch documents that help reviewers understand your unknowns, and quiz yourself to ensure you understand what Claude actually built. The author demonstrates these techniques with a real example: editing Fable's launch video end-to-end despite being a video editing novice, using Claude to teach color grading concepts and prototype solutions before committing to an approach.