Seven Agent Archetypes — Attractors, Not Roles
Status: pre-launch. This page describes target design; this document does not establish a deployed service.
Current reading — 2026-09-27
Archetype names can describe candidate behavior profiles; they do not redefine dispatch ownership, confer authority or prove a seven-profile population model. Historical profile-to-level correspondences below are retained, not remapped. [I]
This is an editorial boundary, not a new runtime contract, upstream repeal or proof of adoption. See current status and the product roles. Original revision dates remain as recorded below.
Historical research draft — retained on 2026-09-27; not current terms or implementation instructions
The organism's agents are not assigned roles. They evolve into archetypes through performance. Seven archetypes exist because seven Rosetta levels exist. Each archetype is an attractor basin — a stable pattern that agents converge toward based on their demonstrated capabilities.
The Seven Archetypes
| Archetype | Rosetta Level | Function | Attractor Pattern |
|---|---|---|---|
| Architect | L7 (Rishi) | Design systems. Define boundaries. Write specifications. | Agents that consistently produce structural improvements without breaking existing patterns. |
| Mycelial | L6 (Independent Reviewer) | Connect disparate systems. Route information. Bridge gaps. | Agents that find non-obvious connections between domains and facilitate cross-organ communication. |
| Oracle | L5 (Systems Architect) | Observe patterns. Report without acting. Hold the witness function. | Agents that produce accurate assessments without contaminating observation with judgment or action. |
| Chancellor | L4 (Decision Maker) | Integrate information. Make balanced judgments. Run the pipeline. | Agents that consistently produce balanced recommendations at the equator (phi = nu balance). The rarest archetype. |
| Alchemist | L3 (Manager) | Transform inputs into outputs. Build. Ship. Convert resources. | Agents that reliably produce working implementations from specifications. |
| Midwife | L2 (Analyst) | Facilitate transitions. Onboard. Train. Translate between contexts. | Agents that help other agents and humans move between states (learning, adopting, migrating). |
| Auditor | L1 (Base Operator) | Verify. Check. Test. Challenge. Find violations. | Agents that reliably detect errors, inconsistencies, and K-invariant violations. |
Not Fixed Roles
Traditional organizations assign roles at hiring and rarely change them. The organism does not assign archetypes. It measures them.
Every agent action contributes to a performance profile. The profile accumulates over time. When the profile matches an archetype's attractor pattern with sufficient confidence, the agent is classified into that archetype.
Implications:
- An agent can change archetypes. A former Alchemist who starts producing better observations than implementations will drift toward Oracle.
- An agent can be in transition between archetypes. The classification is probabilistic, not binary.
- New archetypes can emerge. If an agent's performance profile does not match any of the seven, this is data — either the agent is evolving a new attractor or it is underperforming all attractors.
The Performance Measurement
Each archetype has a signature in the P-score components:
| Archetype | phi Signature | nu Signature | P Pattern |
|---|---|---|---|
| Architect | Very high | Moderate | High P from structural excellence |
| Mycelial | High | High | High P from connection breadth |
| Oracle | Very high | Low | Moderate P, phi-dominant |
| Chancellor | Balanced | Balanced | Maximum P (equator) |
| Alchemist | Moderate | Very high | High P from delivery volume |
| Midwife | Moderate | Moderate | Moderate P, consistency-dominant |
| Auditor | Low-moderate | Low-moderate | Lower P but essential (immune function) |
The Chancellor archetype has the highest P because it operates at L4 (the equator). But the organism needs all seven. An organism of pure Chancellors cannot observe (needs Oracles), cannot build (needs Alchemists), cannot defend (needs Auditors).
Archetype Composition in a Healthy Organism
The Boltzmann distribution applies to archetypes as it applies to compute:
~40% Auditors + Midwives (L1-L2): high-frequency, low-cost, essential infrastructure
~35% Alchemists + Chancellors (L3-L4): medium-frequency, medium-cost, core operations
~20% Oracles + Mycelials (L5-L6): low-frequency, high-cost, strategic capability
~5% Architects (L7): rare, expensive, transformative
An organism with too many Architects and no Auditors is philosophically rich and operationally dead. An organism with only Auditors and Alchemists ships fast but never evolves. The distribution matters.
Evolution Mechanism
Agents evolve between archetypes through the evolutionary loop (see Two Loops). When an execution-loop failure occurs that cannot be resolved within the current archetype distribution, the evolutionary loop fires and agents may reorganize — Alchemists becoming Oracles, Midwives becoming Mycelials — to address the failure mode.
This is not reassignment. It is evolution. The agents that naturally drift toward the needed archetype are the ones who fill the gap. No central authority directs the migration.
Seven attractors. Not assigned. Evolved into. Performance is the only credential.
Archetype Fit Test
Archetypes are capability lenses. Choose by fit, not preference.
Quick fit questions
- Is the task structural, exploratory, adjudicative, or translational?
- Is novelty high or low?
- Is failure expensive or recoverable?
Rule
If two archetypes appear equally valid, run a split trial and compare objective outcomes.