# A Field Guide to Fable: Finding Your Unknowns > [!question] Lingering Questions > - The tweet gives the principle (map vs. territory) with no concrete workflow example — what does "checking the territory" actually look like as a repeatable step in a Fable session, beyond "run the tests"? > - Does this generalize past coding agents to any AI tool operating on a domain the prompt-writer only partially models (e.g. AI-assisted pastoral or governance work)? > [!tip] Article Ideas > - "The map is not the territory, for AI agents" — a short piece translating Korzybski's line into a concrete AI-assisted-development checklist, using WordPress plugin/theme examples (test-first, diff review, log inspection). > [!link] Cross Reference > - [[3. Resources/MOCs/AI Agent Safety MOC]] > - Same-batch companion piece "I Finally Get Fable 5" makes the same point operationally: verifying an agent's work against the real repo cost more tokens than the work itself — the territory pushed back. > [!quote] Vocabulary > - **Map vs. territory** — Korzybski's distinction, applied here to AI agents: prompts/skills/context are the map; the real codebase and its constraints are the territory. The map can guide but never fully substitutes for the territory. ## Metadata - Source: [A Field Guide to Fable Finding Your Unknowns [01kx13e2dfjeh2vjxhzr7e19g2]](3.%20Resources/Readwise/Tweets/A%20Field%20Guide%20to%20Fable%20Finding%20Your%20Unknowns%20%5B01kx13e2dfjeh2vjxhzr7e19g2%5D.md) - Author: trq212 via x.com - URL: https://x.com/trq212/status/2073100352921215386/?rw_tt_thread=True ## Author's thesis Working with Claude Fable 5 keeps resurfacing a classic epistemic lesson: prompts, skills, and context ("the map") are only representations of the work, never the work itself ("the territory") — the codebase, environment, constraints, and edge cases. The author wants readers to design AI-agent workflows that loop continuously through real execution and observation rather than trusting the prompt on faith. ## Key arguments (≤3) 1. **The map is not the territory, applied to AI**: what you give Claude (prompts, skills, context, schemas) is a representation, always narrower and cleaner than the real system. 2. **The gap is a feature to exploit, not a bug to hide from**: Fable continuously exposes where your mental model of a task diverges from how the system actually behaves once code runs. 3. **Productive use requires a loop**, not a one-shot prompt: propose via Fable → run in the real system → observe the mismatch → refine the map → repeat. ## Claim atoms ### AI prompts only approximate real systems > [!quote] ID:rw_fable-fieldguide-h1 > The map, a representation of the work to be done, is my prompts and skills and context, it's what I give Claude. Where/why: Opening framing of the thread — establishes "the map" as everything the user hands the model, distinct from the real environment. ### Real work happens in the territory > [!quote] ID:rw_fable-fieldguide-h2 > The territory is where the work needs to happen, the codebase, the real world, its actual constraints. Where/why: Companion half of the map/territory pair — grounds the abstraction in concrete terms (codebase, constraints) rather than leaving it purely metaphorical. ## Permanent candidates - **AI prompts are maps, not the territory** (promoted — see linked Permanent note): the single unifying claim behind both quoted atoms above — one idea, not two, since H1 and H2 are the same claim stated from each side. ## Processing checklist - [ ] Find a concrete Fable-session example (from this or another source) where map/territory divergence was caught and corrected, to anchor the abstract claim - [x] Connect into [[3. Resources/MOCs/AI Agent Safety MOC]] ## Links - [[3. Resources/MOCs/AI Agent Safety MOC]] — because applies-in: treating agent output as provisional until checked against the real system is a core safety discipline, not just a productivity tip. - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — because applies-in: the map/territory framing generalizes to how any knowledge worker should treat AI-generated plans.