Unclassified

08 · Twin Execution

The Twin is the human-facing AI coworker, not a separate Workbench. Identity, relationship to user, personal context, AI continuity, business context, current work, capabilities, authority, governed execution, memory, handoff.

Twin Is the Human-Facing AI Coworker

Twin ≠ AI Workbench. The AI Workbench is the operating environment. Twin is the coworker who operates within it.

BUSINESS CONTEXT SPINE
        │
        ▼
  AI WORKBENCH (projection over BC)
        │
        ├── Twin  ← you are here
        │     identity: twin:{userId}
        │     bound to human, tenant, authority
        ├── Agent runtime
        ├── Model routing / ADK
        └── Capability access (governed)
        │
        ▼
  AI CONTINUITY MEMORY (serves BC)
        │
        ▼
   Shared Work / Evidence / Verify → Spine

Twin Identity & Relationship to User

Property	Value
Identity	`twin:{userId}` — one per human, tenant-scoped
Relationship	Knows personal context, current Work, and what the user is authorized to see
Projection	Hydrates the same Active Context Bundle as the Human Workbench for the same Work
Continuity	AI Continuity Memory — Twin's working continuity that serves Business Context
Authority	Same authorization surface as the user, but never owner of decisions or approvals

What Twin Does in a Work Context

Receives the Active Context Bundle: Business Context slice + Work + evidence + authority + AI Continuity relevant to this Work and user.


Assembles precise context, reasons over it, and prepares a plan or draft — scoped to the Work objective andBusiness Context.


Proposes a plan: capability, inputs, sequencing, and what will need approval. Surfaces business context, not raw tool payloads.


Submits the plan to governance. If authorized, hands subtasks to Spine Agents via Shared Work or executes via Hermes. Never bypasses approval.


Observes outcome, captures evidence, assists verification/reconciliation, and leaves continuity so the human or next agent continues without re-briefing.

Capabilities & Authority

Twin selects provider-neutral capabilities via the Capability Fabric, not hard-coded tool calls.
Twin never executes external mutations without a governance verdict.
Every Twin action is traced and evidence-linked to the Work and the Business Context it operated on.

Memory & Handoff

Twin's working notes live in AI Continuity Memory (serving Business Context, not replacing it).
On handoff, the next participant (human or agent) hydrates from Shared Work + Business Context, not from a pasted Twin transcript.

User thought ≠ AI inference ≠ Organizational truth

Promotion of a Twin inference to organizational truth requires the same path as any AI working item: Triage / Governance / governed write → verify → reconcile → Spine.

Where to Go Next

The operating environment Twin lives in.


When Twin hands a subtask to a governed organizational actor.


Three domains and G-10: AI Continuity scope vs persistent substrate.

</CardGroup>