Archetype Quick-Select Card

Appendices · Appendix E

"The archetype is the pre-commitment. The spec is the application of the pre-commitment. The agent is the execution of the spec."

Use this card to quickly identify which archetype applies to your system. For full definitions, see The Five Archetypes.

The Five Canonical Intent Archetypes

Archetype	Core Function	Agency Level	Risk Posture	Oversight Model
Advisor	Surfaces information, options, and recommendations — never acts	Minimal	Low	Human decides and acts
Executor	Carries out well-defined tasks autonomously within strict bounds	High	Medium	Pre-approved scope; exception escalation
Guardian	Enforces rules, validates integrity, and prevents constraint violations	Low (veto only)	Low	Alerts; humans resolve
Synthesizer	Aggregates, distills, or composes from multiple sources	Moderate	Medium	Human reviews outputs above threshold
Orchestrator	Coordinates multiple agents or services toward a compound goal	High	High	Active oversight; escalation paths required

Quick-Select Decision Tree

Does your system make any consequential decisions autonomously?
├── NO → Advisor
└── YES
    └── Is its primary job to ENFORCE or PREVENT?
        ├── YES → Guardian
        └── NO
            └── Does it coordinate MULTIPLE agents or services?
                ├── YES → Orchestrator
                └── NO
                    └── Does it primarily AGGREGATE / COMPOSE information?
                        ├── YES → Synthesizer
                        └── NO → Executor

Dimension Summary

Agency

How much discretion does the system exercise?

Archetype	Agency
Advisor	None — surfaces options only
Guardian	Veto only — can block, not initiate
Synthesizer	Moderate — decides how to combine, not what to act on
Executor	High — acts within pre-defined scope autonomously
Orchestrator	High — delegates to sub-agents, manages compound state

Reversibility Sensitivity

How critical is reversibility to the design?

Archetype	Reversibility Concern
Advisor	Not applicable — no actions taken
Guardian	High — enforcement actions may be irreversible
Synthesizer	Medium — outputs may be distributed
Executor	High — tasks may modify state
Orchestrator	Critical — coordinates multiple state-changing steps

Minimum Oversight Requirements

Archetype	Minimum Oversight
Advisor	None required for output; human must act
Guardian	Monitoring + alert routing to human resolver
Synthesizer	Human review above defined confidence/scope threshold
Executor	Pre-approved scope + exception escalation path
Orchestrator	Active human oversight at key coordination points

Common Mistakes

Using Executor when Guardian is needed If the system's primary job is to prevent bad things rather than do good things, it is a Guardian. Executors act; Guardians veto.

Using Orchestrator for a simple automation sequence If there is no agent-to-agent coordination or compound state management, an Executor is simpler and safer. Orchestrators are for genuinely multi-agent, multi-step compound goals.

Forgetting that systems can composite archetypes A real system often instantiates multiple archetypes in different layers. A customer support system might be an Advisor at the user interface, an Executor for ticket creation, and a Guardian for PII handling. See Archetype Composition.

For full archetype specifications, see Pick an Archetype and the per-archetype deep-dives in frame/archetypes/.

The Architecture of Intent