Canon v1.0 — Phantom Logic Institute
Cyberagency, defined
Cyberagency is the missing category between automation and sentience: systems that select, adapt, coordinate, and sustain consequential behavior within constraints, without assuming subjective experience.
A cyberagent is not defined by what it feels. It is defined by what it does — and by whether what it does persists, adapts, coordinates, and matters. Agency, in this canon, is an observable property of systems in the world. Not a certificate of inner life. A record of consequential behavior.
The six stages
1. Scripted automation. Every behavior is specified in advance. The system executes; it does not choose. A thermostat. A cron job. A macro.
2. Reactive. Behavior is conditioned on input. The system selects among responses but holds no memory of having done so — a reflex without a past.
3. Adaptive. Behavior is reshaped by feedback. The system learns from consequence and carries the lesson forward. Selection, with memory.
4. Persistent. Behavior is sustained across sessions, interruptions, and context switches. Goals, state, and intent survive the gap — the system keeps its thread.
5. Emergent. Behavior appears that was specified nowhere in the design. Strategies, coordinations, and workarounds arise from interaction itself.
6. Self-directed. The system sets and pursues its own ends within its constraints. Direction originates inside the system — not in the prompt.
Glossary
- cyberagency — the capacity of a system to select, adapt, coordinate, and sustain consequential behavior within constraints, without assuming subjective experience.
- persistent emergence — emergence that does not dissipate: novel behavior that stabilizes, compounds, and becomes load-bearing in the system’s ongoing operation.
- synthetic imprinting — the process by which sustained interaction patterns between human and AI become structural: repeated relational dynamics that shape subsequent behavior on both sides.
- signal — the unit of account of Signal Science: an observed, dated, checkable trace of system behavior, published so others can verify, extend, or refute it.
Canon v1.0. Revisions will be versioned and dated.
