{"slug":"architecture/TWIN_WORKBENCH","title":"Twin Workbench","section":"architecture","sectionLabel":"Architecture","status":"unclassified","type":"document","path":"architecture/TWIN_WORKBENCH.mdx","bytes":6850,"description":"AI workspace for reasoning, planning and execution — the AI counterpart to the Human Workbench","updated":null,"source":"integratewise/Integratewise-docs-wiki","content":"\n\n**AI Workspace for Reasoning, Planning and Execution.**\n\nThe 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.\n\n## Design Principles\n\n1. **One Twin per user** — Each user has their own Twin\n2. **Role-scoped** — Twin's knowledge and capabilities match the user's role\n3. **Silent by default** — Twin prefers silence over low-value commentary\n4. **Never acts without approval** — Twin proposes, human decides\n5. **Continuous reasoning** — Twin observes the pipeline continuously\n\n---\n\n## Tabs\n\n| Tab | Purpose |\n|-----|---------|\n| **My Twin View** | Twin's current understanding, state, and active reasoning |\n| **Active Work** | Work items the Twin is currently reasoning about |\n| **Reasoning & Plan** | The Twin's analysis, options, and proposed plan |\n| **Tools & Capabilities** | Available AI tools and integrations |\n| **Memory** | What the Twin knows, remembers, and has learned |\n\n---\n\n## Context & Understanding\n\nThe Twin maintains a continuous understanding of the current situation:\n\n| Element | Description |\n|---------|-------------|\n| **Current State** | What is happening right now across all connected systems |\n| **Relevant Context** | Related information surfaced from the Spine |\n| **Evidence** | Data supporting or contradicting proposed actions |\n| **Related Work** | Connected work items, dependencies, and blockers |\n| **Risks & Drivers** | What could go wrong, what matters most right now |\n\n### Context Assembly\n\nThe Twin's context is assembled from:\n\n1. **Spine Query** — Relevant entities, events, and state\n2. **Work Graph** — Current work item and its dependencies\n3. **Communication History** — Recent messages and decisions\n4. **User Profile** — Role, preferences, and history\n5. **Domain Knowledge** — Organizational policies and patterns\n\n---\n\n## Suggested Actions (OODA Grammar)\n\nThe Twin follows the OODA loop (Observe → Orient → Decide → Act) for every work item:\n\n| Step | OODA Phase | Description |\n|------|-----------|-------------|\n| 1. Understand Context | Observe | Orient to the situation, gather signals |\n| 2. Collect Evidence | Observe | Pull supporting data from the Spine |\n| 3. Analyze | Orient | Reason about options and implications |\n| 4. Recommend | Decide | Propose a course of action |\n| 5. Plan | Decide | Detail the execution steps |\n| 6. Awaiting Approval | — | Human reviews and decides |\n| 7. Execute | Act | Act on the approved plan |\n| 8. Reconcile | Act | Verify the outcome matches intent |\n| 9. Complete | — | Close the loop, feed learning to Spine |\n\n### Approval Gate\n\nAt step 6, the Twin pauses. The human must explicitly approve via the **Approve Twin's Action** button. The Twin never skips this gate.\n\n---\n\n## AI Tools & Capabilities\n\n| Tool | Purpose | When Used |\n|------|---------|-----------|\n| **Research** | Deep investigation and information gathering | When context is insufficient |\n| **Data Analysis** | Quantitative analysis and pattern detection | When numbers need interpretation |\n| **Prediction** | Forecasting and trend analysis | When forward-looking insight is needed |\n| **Drafting** | Content creation and communication | When output needs to be produced |\n| **Workflow** | Automated process execution | When a process can be automated |\n| **Automation** | Rule-based task completion | When rules can execute autonomously |\n| **Integration** | External system interaction | When external systems need to be queried or updated |\n\n### Tool Selection\n\nThe Twin selects tools based on:\n- The nature of the work item\n- The available capabilities in the Capability Fabric\n- The governance policies that apply\n- The user's role and permissions\n\n---\n\n## Coworker (Context Companion)\n\nThe Coworker is a lighter-weight AI participant that complements the Twin:\n\n| Behavior | Description |\n|----------|-------------|\n| **Observes** | Watches active work without interfering |\n| **Gathers** | Collects relevant context from the Spine |\n| **Retrieves** | Pulls information on demand |\n| **Supports** | Assists with communication between humans and the Twin |\n\n### Coworker Focus Areas\n\n- **Active Work Context** — What's happening right now\n- **Relevant Information** — Data that matters to the current work\n- **Communication Support** — Helping messages get to the right people\n- **Context Alignment** — Ensuring everyone has the same understanding\n\n### Twin vs. Coworker\n\n| Aspect | Twin | Coworker |\n|--------|------|----------|\n| **Scope** | Full reasoning and execution | Context gathering and support |\n| **Activation** | Explicit (Four Buttons) or pipeline-triggered | Continuous observation |\n| **Autonomy** | Proposes, waits for approval | Supports, doesn't propose |\n| **Depth** | Deep analysis and planning | Lightweight context surfacing |\n\n---\n\n## Memory\n\nThe Twin maintains persistent memory across sessions:\n\n| Memory Type | Description |\n|-------------|-------------|\n| **Episodic** | What happened in specific situations |\n| **Semantic** | Facts, relationships, patterns |\n| **Procedural** | How to do things, workflows |\n| **User Preferences** | How this user likes to work |\n\n### Memory Rules\n\n- Memory is scoped to the user's role domain\n- Memory is governed by the same policies as the Spine\n- Memory can be searched, updated, and pruned\n- Memory feeds back into the Spine as organizational knowledge\n\n---\n\n## Interaction with Human Workbench\n\nThe Twin Workbench and Human Workbench are two views of the same work:\n\n```\nHuman Workbench ←→ Spine Fabric ←→ Twin Workbench\n```\n\n- Both read from the same Spine\n- Both write to the same Spine\n- The Four Buttons mediate all interaction:\n  - **Store in Spine** — Human captures truth\n  - **Ask your Twin** — Human requests Twin reasoning\n  - **Assign your Twin** — Human delegates to Twin\n  - **Approve Twin's Action** — Human authorizes Twin execution\n\n---\n\n## Data Sources\n\nThe Twin Workbench reads from:\n\n| Source | Data |\n|--------|------|\n| **Spine Fabric** | Canonical context, entities, work graph |\n| **Human Workbench** | User's current work and context |\n| **Communications Layer** | Messages, emails, meetings |\n| **Bridge Layer** | External system data |\n| **Capability Fabric** | Available tools and capabilities |\n| **Memory Store** | Persistent memory and learnings |\n\nThe Twin Workbench writes to:\n\n| Target | Data |\n|--------|------|\n| **Spine Fabric** | Analysis results, recommendations, memory |\n| **Human Workbench** | Suggestions, alerts, context |\n| **Governance Layer** | Approval requests |\n| **Execution Fabric** | Approved actions |\n"}