The Directive Gap
🌳 Evergreen ·
The directive gap is the distance between the human’s goal (with all their ambient knowledge) and the context actually available to the agent during execution. The human who said “implement feature X” had knowledge about the codebase, conventions, and likely obstacles that was never made local. This gap is the root cause of most suggestible-actor failures. When it is wide, the agent confabulates.
Every prescription for the suggestible actor (actionable error messages, hard boundaries with signposts, documentation as local context, CI/CD gates that report remediation) is a strategy for closing this gap. The core principle is to convert ambient knowledge into local context: make explicit what the human developer would have carried implicitly.
Connections
- 🌳Confabulation Is Plausible
- 🌳Convert Ambient Knowledge into Local Context
- 🌿Directive gap closure is indistinguishable from aligned confabulation
- 🌿Reverse-Pass Observer Role
- 🌿Spec and Code as Co-Artifacts Stable Relative to Model
- 🌳Susceptibility Peaks at Failure
- 🌳The Suggestible Actor: Four Properties
- 🌳Two-Layer Governance Model
- 🪶The Architecture Orphaning Problem with AI Agents
- 🪶The Hidden Directive Gap
- 🪶The Suggestible Actor: A New Model for AI-Assisted Software Development