Research note
Agentic commercial intelligence systems need more than prompts.
The Intellectus operating model combines domain memory, workflow state, and human review so AI agents can support decisions instead of producing disconnected drafts.
The pattern
Useful commercial agents are not generic chat interfaces. They need a domain model, a source of truth, and a way to route work back into the systems people already use. Duke uses procurement data, buyer history, and tender context. Clara uses brand memory, campaign state, assets, and approval feedback.
1. Model the domainRepresent tenders, buyers, brands, campaigns, assets, and approvals as first-class objects.
2. Keep statePersist the context that changes decisions, not just the last prompt.
3. Route judgmentUse people where consequences are strategic, legal, financial, or reputational.
Where the model is visible
Duke shows the procurement version of this model: source coverage, normalized data, and procurement-specific workflow. Clara shows the marketing version: brand memory, campaign planning, and always-on execution support.