Vision

The Idea

RAID is a real University of Melbourne student club. ORG1 is the experiment to see if an AI agent can run it.

Not assist — run. The goal is to progressively shift operational management from human to AI, documenting what works, what breaks, and what the limits are along the way.

The Autonomy Pipeline

The project follows a five-stage arc:

1. Human-led — Humans do everything, AI observes. The baseline.

1. Human augmented with AI — The human does the work, the AI helps with planning, documentation, and tracking. This is where most AI tools sit today.

1. Agentic AI with human monitoring — The AI drives operations, the human watches. The AI plans the sprints, generates the tasks, maintains the knowledge. The human monitors and intervenes when needed.

1. Agentic AI with human in the loop — The AI operates, the human approves at checkpoints rather than directing. The human handles what requires physical presence or judgment calls.

1. Autonomous — The AI operates independently within defined boundaries. Heartbeat loops that progress the organization, seeking human approval only at key decision points.

We want to see how far along this pipeline we can get within one year, and what's blocking the next step at each stage.

Timeline

ORG1 runs for one year: February 2026 to February 2027.

Phase 1: Experimentation & Building (Feb 2026 - Jul 2026) — Model the organization, map workflows, build initial tools, iterate through trial and error. This phase spans Semester 1, exam period, and winter break.

Phase 2: Execution & Transition (Jul 2026 - Feb 2027) — Deploy what survived Phase 1 across Semester 2. Run AI-managed workflows for real. Transition to the next committee over summer break.

Why a Student Club?

Small organizations are ideal test subjects. They have real operational needs — events, budgets, compliance, member communication — but the stakes are manageable. If the AI makes a mistake, nobody loses their livelihood.

Student clubs also have a problem AI can help with: institutional knowledge loss. Committee members turn over every year. Knowledge disappears. An AI system that maintains organizational memory could be genuinely useful, not just a research exercise.

Inspiration

Vending-Bench (Andon Labs, 2025) showed that LLMs can manage a simulated vending machine business but struggle with long-term coherence — they forget orders, misinterpret schedules, and sometimes spiral. ORG1 is the real-world version of that question: can AI sustain coherent management of an actual organization over months and years, not just a simulation?

Technology

ORG1 uses various AI technologies, including frameworks from NUU Cognition, to build the systems needed. All research and results are published openly and all outputs are open-source.