Product doctrine: the role-composed User Workbench
The IntegrateWise product is a role-composed workbench. Its job is to help a person complete current work with context already assembled and difficult preparation already done. It is not a thin shell around a generic assistant.
Product promise
The experience should feel like this:
My work already knows its context. AI prepared the hard parts. I decide.
The person remains the accountable actor. AI may retrieve, relate, explain, prepare, suggest, or propose; it does not silently create truth, apply policy, approve a decision, or execute a consequential action.
Composition and coverage
Every domain assembles its surface from the same inputs:
- role and department;
- active work focus;
- canonical work state and related evidence;
- the user’s authority and available capabilities; and
- a placement-aware ContinuityBundle.
Leads is one domain, not the model for the entire product. The same rule governs Accounts, Marketing, Product, Finance, Service, Education, Personal, and future categories. Each domain supplies semantic slots such as an overview, timeline, decision, meeting, preparation, or detail anchor. It must not create a separate “Twin product” for itself.
The Twin in the product
The Twin is a quiet, contextual cognitive participant. It can surface only grounded, placement-aware interventions:
NOTICE, EXPLAIN, RELATE, PREPARE, SUGGEST, and PROPOSE.
An intervention must identify the relevant evidence or freshness state, respect the user’s authority, and expose declarative next actions. Silence is correct when there is no useful, grounded intervention.
The Twin does not own:
- canonical work truth or memory;
- identity, policy, governance, or authorization;
- orchestration or execution; or
- a permanent rail, bubble, panel, or route that competes with the work.
The human control grammar
The Workbench makes the Twin useful through four stable, user-facing controls.
They are the human control grammar over connected work, not a set of generic AI
buttons:
| Work loop | User control | Meaning |
|---|---|---|
| Observe | Store in Spine | Preserve a fact, commitment, decision, evidence, or outcome through the governed truth pipeline. |
| Orient | Ask Your Twin | Understand the current work from its already-grounded Continuity Bundle. |
| Decide | Assign Your Twin | Delegate bounded investigation, analysis, drafting, and plan preparation. |
| Act | Approve Twin's Action | Authorize a reviewable, immutable ExecutionPlan across permitted connected capabilities. |
Connected-tool capabilities are a separate layer from these controls. The Twin
prepares and recommends; connected capabilities perform work only after the
required human approval and governance checks. The product must never describe
a prepared plan as completed work, or treat an AI message as an approval.
Interaction laws
- Start in the active canvas at the semantic place where the work happens.
- Use the least interruptive intervention that can help: notice before explanation, preparation before proposal, proposal before action.
- Do not route a person away from work merely to expose AI. A deeper Ask flow is allowed only when the person explicitly asks for it.
- A proposal is not an approval. Approval remains a visible, governed human decision.
- Do not fabricate an intervention from UI heuristics, stale local state, or model confidence alone. The platform must return a grounded bundle first.
Product acceptance checks
A new domain is product-aligned only if a user can stay in that domain’s work surface, see no generic assistant chrome, receive context-specific help when evidence warrants it, and retain clear decision rights. If it requires a permanent Twin destination to be useful, the domain has not yet been composed around work context.
The decisive product test is:
Can the user perform the approved capabilities of their connected work tools from this Workbench, without tool juggling or reconstructing context?
The required answer is yes. The Workbench composes what the user needs to
know, what the user is able to do, and the four controls by which the user
observes, understands, delegates, and authorizes consequential work.