RU

Organization archetypes

Illustrative organizational patterns derived from purpose and operating constraints.

#Archetypes are combinations

An archetype is a useful recurring combination of purpose, authority, resources, and work patterns. It is not a biological species or a certification category. The same organization can combine several archetypes, and an archetype can be implemented with different numbers of agents and different control structures.

The examples here are design illustrations rather than claims that all these organizations have been deployed. They help a reader ask concrete architectural questions before choosing a framework. A good archetype exposes characteristic obligations and failure modes as well as the work it might automate.

#Six illustrative patterns

A research office values traceable evidence; an operations office values continuity and bounded changes. Commercial, publishing, stewardship, and coordination organizations introduce other obligations. These distinctions concern the purpose of the organization, rather than prescribing one agent architecture.

ArchetypeContinuing obligation
Research officeQuestions and evidence
Operations officeService condition and incidents
Publishing organizationAccuracy and corrections
Commercial organizationCustomers and resource accounting
Resource stewardAllocation and preservation
Federation coordinatorAgreements across organizations

#Turn a pattern into a design

For a publishing organization, specify who may approve publication, how sources are checked, and how a published mistake is corrected. For operations, specify allowed changes, observations of actual service state, and handling of uncertain actions. A commercial design additionally needs clear ownership of customer commitments and resource records.

An archetype does not establish competence. Automated research can produce unreliable conclusions, and simulated business profit does not demonstrate real business viability. Use a separate organizational profile to describe the chosen design, then evaluate representative work, interruptions, and boundary cases before making claims about sustained performance.

Sources & context

  1. Lu et al. (2024). The AI Scientist
  2. Backlund & Petersson (2025). Vending-Bench
  3. Xu et al. (2024). TheAgentCompany

Reference edition 0.1 · 2026-10-14. Read the editorial method.