ADR-021: AI Agent-Orchestrated Case Management Framework

Status: Accepted Version: 1.0 Date: 2025-10-01 Supersedes: N/A Related ADRs: N/A Related PRDs: PRD-001

Context

See: PRD-001 (Automatic Application Synthesis)

Decision

We formally adopt AI Agent-Orchestrated Case Management, modeled after the CMMN (Case Management Model and Notation) standard, as the canonical reference domain for SEA™ Forge. This system treats projects as “Cases” chartered by a Desired Outcome, moving away from linear waterfall pipelines in favor of event-driven, discretionary work patterns.

The system architecture MUST include the following components:

1. The Hierarchical Artifact Pipeline

The system will manage the lifecycle of meaning through four distinct stages of reification:

1. Cognitive Artifacts: Ephemeral aids (lists, sketches, notes) intended to reduce mental load.
2. Intellectual Artifacts: Synthesized constructs (SOPs, domain concepts, personas) derived from cognitive artifacts.
3. Information Products: Refined intellectual artifacts mapped to market/domain value (software systems, campaigns).
4. Intellectual Capital: Standardized information products transformed into enterprise assets via an Exchange Rate Policy.

2. Hybrid Project Membership

• Human Members: Strategic intent, ethical oversight, sovereign judgment.
• AI Members (G3): Specialized agents created on-the-fly using the project Knowledge Base.
• Project Manager Agent (PM-Agent): A built-in conductor managing AI members ensuring cognitive ergonomics.

3. CMMN Execution Model

• Case Plan Items: Tasks/stages forming the unit of work.
• Discretionary Tasks: Activated by PM-Agent or human choice.
• Sentries: Entry/exit criteria gating activation.

Rationale

CMMN provides a proven, standards-based model for knowledge-intensive, event-driven work patterns. By combining this with AI agent orchestration and a hierarchical artifact pipeline, we enable:

• Non-linear, discretionary work patterns
• Clear progression from cognitive aids to enterprise assets
• Human-AI hybrid collaboration with clear role boundaries
• Traceable, governed state transitions

Constraints

• MUST use CMMN-based execution model for case management
• MUST track artifact lifecycle through four reification stages
• MUST support hybrid human and AI project membership
• MUST generate PlanItemTransitionRecord for every state transition
• MUST require TransitionToken evidence for ProjectionToken creation
• MUST assign typed Identity Tokens to all artifacts and log in IFL

Quality Attributes

• Round-trip integrity for all state transitions
• Provenance gating preventing “teleportation”
• Immutable lineage for all artifacts
• Economic isomorphism enabling value flow tokenization
• Event-driven, discretionary work patterns

Bounded Contexts Impacted

• Case Management
• Artifact Pipeline
• AI Agent Runtime
• Knowledge Layer
• Internal Federated Ledger

Consequences

Positive

1. Round-Trip Integrity: Every state transition generates a PlanItemTransitionRecord.
2. Provenance Gating: ProjectionToken creation requires TransitionToken evidence (no teleportation).
3. Immutable Lineage: All artifacts assigned typed Identity Tokens logged in IFL.
4. Economic Isomorphism: Alignment of artifact pipeline to Intellectual Capital allows tokenization of value flows.

Negative

• Requires implementation of CMMN execution engine
• Additional complexity in artifact lifecycle management
• Overhead of transition record generation

Additional Notes

References

• SDS-050: Semantic Identity Provenance