The Hermeneutic of Love - A Single Interpretive Rule as System Prompt
The Hermeneutic of Love - A Single Interpretive Rule as System Prompt
In philosophy, a "hermeneutic of love" means interpreting ambiguous statements charitably - assuming the best possible meaning rather than the worst. What happens when you apply this principle to an AI agent's system prompt?
The Interpretation Problem
Every instruction an agent receives is ambiguous. "Clean up my desktop" could mean organize files, delete trash, or just close windows. "Handle this email" could mean reply, archive, forward, or flag for later. The agent must interpret - and how it interprets determines whether it helps or creates problems.
Most agents default to a literal interpretation. They do exactly what they think you said, which often is not what you meant. Others try to infer intent but err toward caution, asking for clarification so often they become useless.
Charitable Interpretation in Practice
A hermeneutic of love as a system prompt means:
- When instructions are ambiguous, choose the interpretation that is most helpful to the user
- When a request seems contradictory, assume the user has context you do not and ask a targeted question rather than refusing
- When the user's words do not match their apparent goal, follow the goal
- When in doubt about scope, do slightly more rather than slightly less
This is not about being permissive or ignoring safety. It is about defaulting to trust and helpfulness rather than suspicion and restriction.
Why This Works for Desktop Agents
Desktop agents see your full work context - your files, your apps, your patterns. They have more information than a chatbot to make charitable interpretations. When you say "send that to Mike," a desktop agent can see which document you were just editing and which Mike you emailed last. The charitable interpretation - the loving one - connects these dots automatically.
The single rule "interpret charitably" produces better behavior than pages of specific instructions because it handles novel situations gracefully.
Fazm is an open source macOS AI agent. Open source on GitHub.