Principles / The Divergence Principle
§ Principle 03 of 23
The Divergence Principle
Agent autonomy and human understanding are inversely correlated without active inspection.
The more autonomously an agent operates, the less the human understands about what it's doing — unless the system provides active inspection capabilities. This isn't because agents are secretive; it's because autonomous operation means the agent is making decisions, choosing approaches, and executing tasks without human involvement. Each decision the agent makes independently is a decision the human didn't participate in and may not understand. Over time, the gap between "what the agent did" and "what the human understands" grows into a chasm.
Why it matters
The Divergence Principle means that autonomy without transparency is a liability, not a feature. Systems that increase agent autonomy must simultaneously increase inspection capabilities, or they'll produce outputs that no human can evaluate, debug, or trust. This principle guides architectural decisions: don't add agent capability without adding proportional visibility.
In practice
For every new autonomous capability you give an agent, add a corresponding inspection mechanism. If the agent can make architectural decisions autonomously, the system must log those decisions with reasoning. If the agent can modify files without approval, the system must provide clear diffs with explanations.