§2. System Overview
Three Pillars
Mind Extended is organized around three pillars:
| Pillar | Function | Question it answers |
|---|---|---|
| Memory (foundation) | How information is stored, updated, and discarded | "Where does this live?" |
| Access (interface) | How agents interact with the system | "How do I reach this?" |
| Action (output) | How ideas become reality through execution | "How does this get done?" |
These pillars are not arbitrary organizational categories. Each derives from specific theoretical insights about how external structures participate in cognition. Memory draws from Luhmann's insight that a well-designed system is generative, not merely archival — it returns what you did not know you knew — and from Hutchins' concept of sedimented reasoning, where past cognitive work is embedded in artifacts for future use. Access draws from Engelbart's observation that augmentation requires not just better artifacts but also shared language, methodology, and training — without these, the tools remain inert. Action draws from Vygotsky's scaffolding: the system provides structured support through the Zone of Proximal Development, enabling forms of thinking the operator could not achieve alone, and from Hutchins' core claim that the unit of analysis is the system acting, not the individual.
Feedback Loop
The three pillars form a continuous cycle:
Action (Execution) → Registry → Memory → Planning → Action (Execution)
Work produces records. Records become memory. Memory informs planning. Planning drives new work. The system improves through this cycle — each iteration refines memory, sharpens access patterns, and improves action efficiency. This is Engelbart's co-evolution in practice: the four means of augmentation (artifacts, language, methodology, training) change the operator, who in turn finds ways to change these elements. The system improves recursively.
Agents
An agent is any entity — human, artificial, or hybrid — that perceives, remembers, and acts within the extended mind system. Mind Extended recognizes three types:
| Type | Code | Description | Examples |
|---|---|---|---|
| Human | H | A biological cognitive agent. Initiates goals, makes value judgments, holds embodied experience. | The system's operator, a collaborator, a client |
| Non-Human | NH | An artificial cognitive agent. Executes tasks, processes information, follows protocols. | A Claude Code session, a CI/CD pipeline, an automated skill |
| Human-Non-Human | HNH | A hybrid cognitive unit where human and artificial agents collaborate as a single cognitive process. Neither is merely a "tool" of the other — both contribute irreplaceably. | An operator working with Claude Code to write a spec, a designer iterating with an AI on a visual concept |
The HNH agent is the most characteristic unit of Mind Extended. The extended mind thesis predicts exactly this: when the collaboration is tight enough — when the conditions for genuine cognitive extension are met (availability, trust, accessibility, and what Clark 2025 calls maintainability for human-AI systems) — the boundary between "human thought" and "AI assistance" dissolves into a single cognitive process.