The organization advances through a series of stages, each representing a more autonomous mode of operation:
| Stage | Description |
|---|---|
| Manual coder | The starting state. All output comes from direct coding effort. |
| Autocomplete | Basic tooling speeds up output. Tech debt compounds faster. |
| Edit bot | An automated refactoring agent runs in the background. |
| Single agent | A general-purpose agent interprets allocation policy directly. |
| Agent swarm | Multiple agents coordinate. Misinterpretation risk grows. |
| Data centre | Power becomes the primary constraint; fossil caps bite harder. |
| Orbital compute | Thermodynamic constraints enter play. |
| Matrioshka sphere | The late game. Civilization-scale consequences of early ethics choices arrive. |
Stage unlocks require meeting a threshold, then accepting the stage unlock through a review screen. The risks of each stage are shown before accepting. Deferring is always an option.
Autopilot modes become available after the edit-bot stage:
- Static autopilot enforces a simple rule (e.g., "maintain security above 20%") without exercising judgement.
- Adaptive autopilot makes small allocation adjustments toward a stated goal and surfaces suggestions for review. Ignoring a suggestion carries no penalty.
- When an autopilot action diverges significantly from stated intent, a misinterpretation warning appears.