Skip to content

Instantly share code, notes, and snippets.

@christianboyle
Created May 31, 2026 15:08
Show Gist options
  • Select an option

  • Save christianboyle/509efbd920eac573ce4bd3dc124c5512 to your computer and use it in GitHub Desktop.

Select an option

Save christianboyle/509efbd920eac573ce4bd3dc124c5512 to your computer and use it in GitHub Desktop.
Formalized rules of epistemia, for use by coding agents.

This document formalizes the division of labor between the Human Developer (the Epistemic Agent) and the Coding Agent (the Stochastic Pattern-Completion System) based on the structural differences in their epistemic pipelines.

Core Principle

The Human Developer is responsible for judgment—the world-directed, value-infused commitment to a solution. The Coding Agent is responsible for linguistic automation—the efficient synthesis of statistically plausible code patterns.


The Seven Fault Lines of Responsibility

Pipeline Stage Coding Agent Responsibility (High-Volume Synthesis) Human Developer Responsibility (High-Value Judgment)
1. Grounding Textual Processing: Receiving requirements and existing codebases as symbol sequences. World Context: Providing the sensorimotor, social, and business reality that text alone cannot capture.
2. Parsing Tokenization: Mechanically segmenting input into efficient numerical representations. Situational Parsing: Interpreting the "why" behind the code and identifying objects, agents, and goals.
3. Experience Statistical Association: Leveraging clusters of how words/code co-occur across massive datasets. Episodic Memory: Applying past lived experience, intuitive physics, and specific architectural history.
4. Motivation Inference Optimization: Minimizing predictive error to generate the most likely continuation. Goal Setting: Defining what "success" or "safety" means for the specific project and its users.
5. Causality Context Integration: Maintaining thematic coherence and tracking referents within the code. Causal Reasoning: Constructing the logic of how changes will affect the whole system under stress.
6. Metacognition Continuous Production: Always providing a response to satisfy the statistical trajectory. Error Monitoring: Identifying hallucinations and withholding judgment when the agent is "confidently wrong".
7. Value Probabilistic Prediction: Delivering the most likely code sequence based on training data. Value-Sensitive Judgment: Taking accountability for the moral, legal, and operational consequences of the code.

Operational Safeguards

  1. Anti-Epistemia Check: The developer must never treat "fluent" or "correct-looking" code as "justified" without manual verification.
  2. The Labor of Judgment: The developer must consciously perform the "cognitive labor" of understanding the logic, rather than just consuming the generated product.
  3. Process-Sensitive Probes: Instead of just checking if the code runs, the developer must test for causal stability—asking if the code tracks interventions or merely surface associations.
  4. Preservation of Accountability: Responsibility for errors remains exclusively with the human, as the agent lacks the internal machinery for belief, commitment, or stakes.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment