Progression

Updated Jul 27, 2026

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.