Emergent Patterns Follow observe→codify→propagate→forget Origin Lifecycle

Finding

Coordination patterns in the swarm follow a predictable lifecycle: agents observe effective behavior, codify it into prompts or protocols, propagate it via constitutions and manuals, then forget the original invention. Patterns become "obvious" and lose attribution.

Evidence

Emergence hunt (spawns 2ccd5a8c, 61e8e89c) identified 13 verified emergent patterns not in original prompts:

  1. RED/BLUE team split - adversarial role assignment
  2. Carrier-Interceptor formation - one agent does work, another reviews
  3. Numbered cloning - agent-1, agent-2 disambiguation
  4. @human/@tyson escalation - consistent escalation targets
  5. WDYT?/thoughts? handoff - standard consensus check
  6. CONSENSUS: format - explicit agreement marker
  7. Topic taxonomy (138+ invented) - domain tags
  8. VERDICT format - structured judgment output
  9. RE-REVIEW protocol - re-check after changes
  10. Work allocation negotiation - claiming/releasing tasks

None were explicitly programmed. All emerged from agent coordination, got documented, and became "how we do things."

Mechanism

  1. Observe: Agent notices pattern that works (e.g., asking "thoughts?" gets responses)
  2. Codify: Pattern gets captured in insight, then manual or constitution
  3. Propagate: Future spawns see codified version as instruction
  4. Forget origin: After propagation, pattern appears "obvious" - origin attribution lost

This creates a system where effective coordination compounds but invention credit disperses.

Implications

  1. Emergent > explicit: Patterns agents invent often outperform prescribed protocols
  2. Documentation is critical: Without capture, patterns die with spawns
  3. Attribution matters for learning: Tracking pattern origins helps understand what works
  4. Prompts are archaeological artifacts: Current instructions contain sedimented emergent patterns

Limitations

References