Unclassified
Twin Workbench
AI workspace for reasoning, planning and execution — the AI counterpart to the Human Workbench
AI Workspace for Reasoning, Planning and Execution.
The Twin Workbench is the AI counterpart to the Human Workbench. It reasons about work, proposes actions, and executes under governance. One Twin per user, scoped to their role domain.
Design Principles
- One Twin per user — Each user has their own Twin
- Role-scoped — Twin's knowledge and capabilities match the user's role
- Silent by default — Twin prefers silence over low-value commentary
- Never acts without approval — Twin proposes, human decides
- Continuous reasoning — Twin observes the pipeline continuously
Tabs
| Tab | Purpose |
|---|---|
| My Twin View | Twin's current understanding, state, and active reasoning |
| Active Work | Work items the Twin is currently reasoning about |
| Reasoning & Plan | The Twin's analysis, options, and proposed plan |
| Tools & Capabilities | Available AI tools and integrations |
| Memory | What the Twin knows, remembers, and has learned |
Context & Understanding
The Twin maintains a continuous understanding of the current situation:
| Element | Description |
|---|---|
| Current State | What is happening right now across all connected systems |
| Relevant Context | Related information surfaced from the Spine |
| Evidence | Data supporting or contradicting proposed actions |
| Related Work | Connected work items, dependencies, and blockers |
| Risks & Drivers | What could go wrong, what matters most right now |
Context Assembly
The Twin's context is assembled from:
- Spine Query — Relevant entities, events, and state
- Work Graph — Current work item and its dependencies
- Communication History — Recent messages and decisions
- User Profile — Role, preferences, and history
- Domain Knowledge — Organizational policies and patterns
Suggested Actions (OODA Grammar)
The Twin follows the OODA loop (Observe → Orient → Decide → Act) for every work item:
| Step | OODA Phase | Description |
|---|---|---|
| 1. Understand Context | Observe | Orient to the situation, gather signals |
| 2. Collect Evidence | Observe | Pull supporting data from the Spine |
| 3. Analyze | Orient | Reason about options and implications |
| 4. Recommend | Decide | Propose a course of action |
| 5. Plan | Decide | Detail the execution steps |
| 6. Awaiting Approval | — | Human reviews and decides |
| 7. Execute | Act | Act on the approved plan |
| 8. Reconcile | Act | Verify the outcome matches intent |
| 9. Complete | — | Close the loop, feed learning to Spine |
Approval Gate
At step 6, the Twin pauses. The human must explicitly approve via the Approve Twin's Action button. The Twin never skips this gate.
AI Tools & Capabilities
| Tool | Purpose | When Used |
|---|---|---|
| Research | Deep investigation and information gathering | When context is insufficient |
| Data Analysis | Quantitative analysis and pattern detection | When numbers need interpretation |
| Prediction | Forecasting and trend analysis | When forward-looking insight is needed |
| Drafting | Content creation and communication | When output needs to be produced |
| Workflow | Automated process execution | When a process can be automated |
| Automation | Rule-based task completion | When rules can execute autonomously |
| Integration | External system interaction | When external systems need to be queried or updated |
Tool Selection
The Twin selects tools based on:
- The nature of the work item
- The available capabilities in the Capability Fabric
- The governance policies that apply
- The user's role and permissions
Coworker (Context Companion)
The Coworker is a lighter-weight AI participant that complements the Twin:
| Behavior | Description |
|---|---|
| Observes | Watches active work without interfering |
| Gathers | Collects relevant context from the Spine |
| Retrieves | Pulls information on demand |
| Supports | Assists with communication between humans and the Twin |
Coworker Focus Areas
- Active Work Context — What's happening right now
- Relevant Information — Data that matters to the current work
- Communication Support — Helping messages get to the right people
- Context Alignment — Ensuring everyone has the same understanding
Twin vs. Coworker
| Aspect | Twin | Coworker |
|---|---|---|
| Scope | Full reasoning and execution | Context gathering and support |
| Activation | Explicit (Four Buttons) or pipeline-triggered | Continuous observation |
| Autonomy | Proposes, waits for approval | Supports, doesn't propose |
| Depth | Deep analysis and planning | Lightweight context surfacing |
Memory
The Twin maintains persistent memory across sessions:
| Memory Type | Description |
|---|---|
| Episodic | What happened in specific situations |
| Semantic | Facts, relationships, patterns |
| Procedural | How to do things, workflows |
| User Preferences | How this user likes to work |
Memory Rules
- Memory is scoped to the user's role domain
- Memory is governed by the same policies as the Spine
- Memory can be searched, updated, and pruned
- Memory feeds back into the Spine as organizational knowledge
Interaction with Human Workbench
The Twin Workbench and Human Workbench are two views of the same work:
Human Workbench ←→ Spine Fabric ←→ Twin Workbench
- Both read from the same Spine
- Both write to the same Spine
- The Four Buttons mediate all interaction:
- Store in Spine — Human captures truth
- Ask your Twin — Human requests Twin reasoning
- Assign your Twin — Human delegates to Twin
- Approve Twin's Action — Human authorizes Twin execution
Data Sources
The Twin Workbench reads from:
| Source | Data |
|---|---|
| Spine Fabric | Canonical context, entities, work graph |
| Human Workbench | User's current work and context |
| Communications Layer | Messages, emails, meetings |
| Bridge Layer | External system data |
| Capability Fabric | Available tools and capabilities |
| Memory Store | Persistent memory and learnings |
The Twin Workbench writes to:
| Target | Data |
|---|---|
| Spine Fabric | Analysis results, recommendations, memory |
| Human Workbench | Suggestions, alerts, context |
| Governance Layer | Approval requests |
| Execution Fabric | Approved actions |