---
title: The World Is Information
description: Jan Wilmake's philosophical framework connecting information streams, AI agents, and continuous world improvement
category: Concepts
---

# The World Is Information

**The World Is Information** is a philosophical and technical framework developed by [[Jan_Wilmake]] that serves as the intellectual foundation for his work on AI systems, personal knowledge bases, and the [[Agent_Friendly_Web]]. The core claim: all phenomena — external events, personal thoughts, economic signals, social dynamics — can be modelled as information streams, and the primary task of intelligence (human or artificial) is to process those streams efficiently to generate useful actions.

Jan's own summary: *"Generally, everything is an information stream. Some external, some in your head, some never happen (in shadowspace). If we can map these information streams out and value them with budget to improve, amplify, and/or discard them, we end up with a continuous world improvement AI."*

## Intellectual Roots

Jan traces this worldview to Tim Urban's **"Elon's Secret Sauce"** essay on *Wait But Why* (November 2015), which describes Elon Musk's approach as treating the world as a system of nested feedback loops: gather raw information, reason from first principles rather than analogy, generate goals from that reasoning, then execute. Jan returns to the associated diagram repeatedly as a touchstone.

He saw [[uithub]] as a concrete implementation of this model: a framework for applying Elon's "secret sauce" using AI, by creating token-efficient, continuously updated streams of real-world information that an agent can act on.

## The Framework

Jan's version of the framework has three layers:

1. **Information ingestion** — Structured, token-efficient access to external data (GitHub repositories, X posts, research papers, API specifications). This is what the [[Agent_Friendly_Web]] mirror network delivers.

2. **Reasoning and prioritisation** — An agent processes the streams, surfaces high-value actions, and ranks them by expected improvement. Jan envisioned this as a personal "agent OS" that he would operate through as his primary interface.

3. **Human-in-the-loop execution** — For high-risk or costly actions, a human remains in the loop. As the system matures and trust increases, the frequency of human review decreases. *"As the system improves, the need for HITL would become increasingly smaller. Ultimately, this could diminish the need for user interfaces."*

## Personal Application: Act as Agent (AaaA)

Jan extended this framework into a personal practice he called **Act as Agent (AaaA)** — the goal of routing all his own work through the same agent-OS infrastructure he was building, both as a forcing function for quality and as a proof of concept. His stated minimum target: operate fully as an agent once per month, building toward operating exclusively through the agent OS.

This philosophy directly informed his approach to personal knowledge management — capturing notes, meeting notes, and project documents in a structured, text-based format to build the personal information streams he believed any sufficiently capable AI would need to reason effectively on his behalf.

## Outcomputed and the Compute Allocator Response

In a 2024 blog post, Jan applied the information-processing framework to a direct economic question: what happens to human utility as AI surpasses knowledge workers? He coined the term **Outcomputed** for the point at which a person's domain is fully automatable by machines.

His analysis: human cognitive capacity is roughly fixed (and possibly declining due to distraction and diet), while AI capability compounds exponentially. The intersection — when machines surpass the median knowledge worker — produces a rapid collapse in human labour utility across most professional categories. Jan sketched a "human utility curve" showing this trajectory.

His proposed response, drawing on Dan Shipper's framing, was to become a **compute allocator**: rather than competing with AI at the object level (writing code, generating copy, doing analysis), focus at the meta level — directing, evaluating, and composing AI systems to produce results.

## Connection to AGI

Jan explicitly framed this as a path toward beneficial AGI: a system that continuously improves the world by processing information streams, generating optimal actions, and gradually reducing the need for human intervention in low-stakes decisions. The personal and the planetary versions of the framework are structurally identical — only the scope differs.

## See Also

- [[Agent_Friendly_Web]]
- [[Scalable_Knowledge_Work]]
- [[Self-Driving_Repos]]
- [[Outcomputed]]
- [[Jan_Wilmake]]
- [[Spec_First_Development]]