Skip to main content

Truth as Context

A systematic process for ensuring AI agents operate with full organizational context when creating or modifying truth documentation.

The Core Principle​

When augmenting your organizational truth (adding new documents, updating existing ones, or refactoring across files), the AI agent should have access to all relevant existing truth. New documentation created in isolation risks contradicting, duplicating, or misaligning with established principles.

The ideal: Every new piece of truth is written with full awareness of all existing truth.

The Context Challenge​

While LLM context windows are expanding rapidly, it's often still unfeasible to include every document in every interaction. This creates a tension:

  • Too little context: Agent produces work that contradicts or duplicates existing truth
  • Too much context: Token costs increase, performance degrades, or context limits are exceeded

The solution isn't to avoid the problem. It's to manage it systematically.

Implementation Framework​

1. Use Your README as the Truth Index​

Your README.md should serve as the index of your entire truth repository, listing every document with a one-line summary. This is the minimum context an agent needs to understand what exists and navigate to relevant files.

The README should always fit in context, even when individual documents can't.

2. Tiered Context Loading​

Always include:

Include when relevant:

  • Documents the new content will reference
  • Documents that cover adjacent topics
  • Recent additions that might overlap

Include on request:

  • Historical documents
  • Archived decisions
  • Rarely-referenced material

3. Pre-Creation Context Check​

Before creating any new truth document, the agent should:

  1. Review the truth index for potential overlaps
  2. Read any documents flagged as related
  3. Identify terminology that must be used consistently
  4. Note any principles the new document must align with

4. Post-Creation Coherence Audit​

After creating or significantly modifying a document:

  1. Check for contradictions with existing principles
  2. Verify terminology consistency
  3. Add appropriate cross-references
  4. Update the truth index if needed

Dealing with Context Pollution​

Per Make Every File Count, if certain documents consistently pollute context without adding value:

Diagnose the problem:

  • Is the document outdated? Update or archive it
  • Is it too verbose? Condense to essentials
  • Is it redundant? Merge with another document
  • Is it rarely relevant? Move to a lower-priority tier

The fix: Ruthlessly prune, consolidate, or rewrite. Source control preserves history. Nothing is truly lost.

The Refactorability Connection​

This process directly enables Make Your Company Refactorable. When all truth fits in context (or is systematically loadable), you can:

  • Rename concepts across all documents in one session
  • Ensure new policies align with existing principles
  • Refactor organizational structure with full awareness of dependencies
  • Maintain coherence as your truth repository grows

The North Star​

An organizational truth repository where any augmentation (new document, update, or refactor) happens with full awareness of existing truth. Not because every document is always loaded, but because you've built systems to ensure relevant context is always available when decisions are made.