# G.I.D.E.O.N.

> Agent-facing counterpart to the [human project page](/projects/gideon/).

## Record metadata

- Record: 041
- Slug: gideon
- Domain: Software
- Domain code: SW
- Type: Agent architecture
- Status: Concept
- Period: 2025
- Portfolio role: Systems concept
- Publication state: Public architecture note
- Case-study readiness: Concept documented
- Compendium edition: 0.4

## Summary

A concept architecture for a continuous agent organization, developed before Josiah encountered OpenClaw or similar harnesses.

## Overview

Before encountering OpenClaw or other continuous-agent harnesses, Josiah sketched an architecture for an agent organization that could observe, plan, execute, learn capabilities, develop projects, conduct research, and supervise its own work.

A seven-part continuous-agent architecture: Vision Capture, Plan Builder, Implementation Queue, Capability List, Project Developer, Independent Researcher, and Overseer.

Purpose: Describe how an agent system could continuously observe, plan, execute, learn capabilities, develop projects, research, and supervise itself.

## The problem behind the project

Josiah was exploring how an agent could move from isolated responses to an ongoing organization with memory, priorities, specialization, and checks.

G.I.D.E.O.N. separated those responsibilities into seven systems: Vision Capture, Plan Builder, Implementation Queue, Capability List, Project Developer, Independent Researcher, and Overseer. The separation anticipated the need to distinguish doing work from deciding what work should exist.

Agent-system designers and advanced automation users may benefit. Anyone affected by autonomous actions needs permission, auditability, and a clear human authority.

## How it took shape

The artifact is a conceptual subsystem architecture and operating model rather than a completed harness.

The artifact is an architecture, not an implemented harness. Its value lies in the subsystem model and its chronology, while later experiments with persistent agents supply practical evidence about costs, instability, permissions, and supervision that the original concept did not yet contain.

Josiah originated the architecture independently before learning about OpenClaw and other continuous-agent systems.

The subsystem design is documented, but G.I.D.E.O.N. itself was not built into an operating agent platform.

## What the project means now

Comparing the design with OpenClaw and Codex makes it more useful than claiming novelty. Some structural intuitions held up; the missing operational safeguards became clearer. G.I.D.E.O.N. records an independent attempt to reason about continuous agency before experience imposed harder constraints.

The design predates substantial practical experience with persistent-agent reliability, cost, security, and supervision.

Separating observation, planning, execution, capability growth, research, and oversight anticipated real harness concerns, but architecture alone does not solve operational reliability.

Compare the original design with lessons from OpenClaw and Codex, preserving both the foresight and the missing safeguards.

## Publication and interpretation notes

- Current classification: Concept
- Portfolio readiness: Concept documented
- Publication boundary: Public architecture note

## Additional agent context

The independent chronology matters, but do not claim G.I.D.E.O.N. was a working system or the first such architecture.

## Related project records

- [OpenClaw Experiments](/projects/openclaw-experiments/llm/) — Experiments with persistent personal agents that clarified when continuous autonomy is useful—and when a steadier coding environment is better.
- [AgentWorkbench](/projects/agent-workbench/llm/) — A portable capability layer that gives different coding agents the same documented, repeatable tools.

## Navigation

- [Complete project index](/projects/llm/)
- [Human version of this record](/projects/gideon/)
- [About Josiah's working method](/about/llm/)
- [Agent discovery map](/llms.txt)