The supervisor + sub-agent pattern.

JustineAI is a workflow, not a chatbot. The supervisor agent holds the PI workflow shape and decomposes each matter into discrete sub-agent tasks — intake, case-theory construction, medical-records analysis, ICD-10/CPT coding, settlement-range computation, demand-package construction. Each sub-agent has a scoped responsibility and a specific Eve-Genesis (Law Edition) training emphasis. The conversational surface is a query layer over the supervisor; it is not the substance of the product.

What the platform actually does.

1. 01

   Supervisor agent owns the matter-level workflow state and the per-status sub-agent dispatch.

2. 02

   Sub-agents are scoped specialists with their own prompts, allowed tool surfaces, and refusal postures.

3. 03

   Sub-agent outputs feed back to the supervisor, which advances the matter’s workflow state.

4. 04

   The conversational layer queries the supervisor; it does not bypass it.

5. 05

   Audit trail records the supervisor sequencing decisions and the sub-agent actions as distinct events.