Unclassified

07 · AI Workbench Execution

What happens when AI receives work — Business Context, AI Continuity, Active Context, context assembly, reasoning, model selection, planning, capability, governance, execution, observation, continuation.

What Happens When AI Receives Work

The AI Workbench is a first-class working environment over business context. It is not a chat panel, an API payload, or the Twin itself.

BUSINESS CONTEXT SPINE
        │
        ▼
  Projection (governed, participant-scoped)
        │
        ▼
  ACTIVE CONTEXT BUNDLE
  Business Context slice · Work · Evidence · Authority · AI Continuity
        │
        ▼
  ┌─────────────────────────────────────────┐
  │           AI WORKBENCH                  │
  │  ┌─────────────────────────────────┐   │
  │  │  Context assembly               │   │
  │  │  Reasoning & orchestration      │   │
  │  │  Model routing (AI Gateway)     │   │
  │  │  Effort / adaptation            │   │
  │  │  Capability selection (Fabric)  │   │
  │  │  Planning                       │   │
  │  │  Governance check (may I?)      │   │
  │  │  Hermes execution (if authorized)│   │
  │  │  Observation & continuation     │   │
  │  └─────────────────────────────────┘   │
  │              │                        │
  │              ▼                        │
  │      AI CONTINUITY MEMORY              │
  │      (serves Business Context)         │
  └─────────────────────────────────────────┘
        │
        ▼
  Shared Work · Evidence · Verify → Spine → Rehydrate

Twin lives inside the AI Workbench. The Twin is the human-facing AI coworker. The AI Workbench is the operating environment.

Internal Flow (derived OODA)

The Workbench internal planning loop is OODA-derived — observe, orient, plan — but it is not exposed as buttons. It is the internal machinery.

AI Workbench hydrates Active Context Bundle: Business Context projection + Work + AI Continuity relevant to this Work. No unrestricted Spine access.


Twin/ADK assembles the precise context needed — entity, history, related Work, decisions, and what the participant is authorized to see.


Reasoning is scoped to the Work's objective and Business Context. Model selection routes through AI Gateway; effort adapts to Work complexity.


A plan is formed: which capabilities, in what order, with what inputs. Capabilities are resolved via [Capability Fabric](/platform/capability-fabric-view) — provider-neutral, not hard-coded tool calls.


Before any external mutation, governance answers: *may it be done?* Identity → authority → policy → approval → authorization.


Hermes executes the authorized plan. Outcome + evidence return. Verification and reconciliation determine what updates the Spine. Both workbenches rehydrate.

Key Separations

Cognition vs execution. Model routing, agent runtime, capability resolution, authority, and execution remain separated. No monolithic AI Orchestrator.
Twin vs Workbench. Twin is the coworker. The Workbench is the environment the Twin (and agents, and ADK) operate in.
AI Continuity vs Business Context. AI Continuity maintains Twin/agent working continuity while remaining grounded in the authoritative Business Context Spine.

Where to Go Next

The Twin's identity, relationship to user, authority, and memory.


Governed organizational actors with identity, role, authority, and evidence.


Where this Workbench sits in the end-to-end Work pipeline.

</CardGroup>