> For the complete documentation index, see [llms.txt](https://www.rolfsen.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://www.rolfsen.ai/index.md).

# Rolfsen.ai

**Engineered cognition, made explicit.**

Rolfsen.ai is a public body of work on the mechanisms beneath intelligent systems: memory, context, agency, tools, metacognition, and the architectures that make those mechanisms operational. The site is not organized as a pile of articles or a set of mutually exclusive categories. It is a graph of ideas: essays, concept hubs, reading paths, field studies, and working systems connected by the primitives they share.

***

## Core Questions

These are the questions the site is built to answer:

* [What is intelligence as an architectural property?](/canon/what-intelligence-actually-is.md)
* [What is an agent loop?](/canon/the-anatomy-of-an-agent-loop.md)
* [What is agent memory, and why is it not just storage?](/canon/memory-is-not-storage.md)
* [What is metacognition in AI agents?](/canon/metacognition-as-an-engineering-primitive.md)
* [What is context engineering?](/concept-hubs/context.md)
* [Why is documentation cognitive infrastructure?](/canon/documentation-is-infrastructure.md)
* [How should AI agent autonomy be designed?](/concept-hubs/autonomy.md)

***

## Start Here

### [The Thesis](/start-here/the-thesis.md)

Why intelligence should be treated as architecture, not product behavior or benchmark performance.

### [Reading Paths](/start-here/reading-paths.md)

Guided paths through the graph for agent builders, systems thinkers, tool users, and frontier readers.

### [Concept Map](/start-here/concept-map.md)

The core primitives and how they connect: intelligence, memory, context, agency, metacognition, documentation, and autonomy.

### [Tag Index](/start-here/tag-index.md)

The public vocabulary of the site. Tags are conceptual handles, not content buckets.

***

## The Canon

### [What Intelligence Actually Is](/canon/what-intelligence-actually-is.md)

Intelligence is not a performance metric. It is an architectural property, and the distinction reshapes how systems should be designed and evaluated.

### [The Anatomy of an Agent Loop](/canon/the-anatomy-of-an-agent-loop.md)

The recurring decision cycle that makes a system an agent, and why most implementations get the structure wrong.

### [Metacognition as an Engineering Primitive](/canon/metacognition-as-an-engineering-primitive.md)

The ability to think about thinking is not a philosophical luxury. It is a decomposable set of mechanisms that determines the ceiling of what agent systems can reliably do.

### [Memory Is Not Storage](/canon/memory-is-not-storage.md)

Memory is retained state that changes future behavior. Treating it as search over stored facts produces agents with good recall and poor continuity.

### [Documentation Is Infrastructure](/canon/documentation-is-infrastructure.md)

In AI-native systems, documentation is not a record of work already done. It is cognitive infrastructure for humans and agents.

***

## Core Concept Hubs

[Agent memory](/concept-hubs/memory.md), [context engineering](/concept-hubs/context.md), [agency in AI systems](/concept-hubs/agency.md), [metacognition in AI agents](/concept-hubs/metacognition.md), [documentation as cognitive infrastructure](/concept-hubs/documentation.md), [AI agent autonomy](/concept-hubs/autonomy.md), and [intelligence as architecture](/concept-hubs/intelligence.md) are recurring primitives. They appear across essays, frameworks, field studies, and libraries because the same mechanism looks different at each layer of the system.

***

## Proof of Construction

### [ThoughtFlow](/workshop/thoughtflow.md)

A pure-Python agent framework built around one contract: `memory = primitive(memory)`.

### [TaskAtlas](/workshop/taskatlas.md)

A lean Python library for representing goals, tasks, relationships, priority, evolution, and history.

***

## Current Inquiry

The primitive mechanisms of agent memory, and why most implementations confuse storage with cognition.