Ask W

W works best when your question is specific and grounded in the workspace you are already using. Treat W as a practical product judgement partner: ask it what to inspect, what to fix, and what is missing from the signals.

What W is for

Use W when you need to decide what to inspect or do next:

• "Who owns this blocked outcome right now?"
• "What is the highest-impact move this week?"
• "Which decision should I revisit first?"
• "What changed since the last source refresh?"

W answers using current signals, decisions, outcomes, and North Star context. If it is uncertain, it should surface what is missing instead of guessing.

W is strongest when the workspace already has traceable signals. If W cannot answer cleanly, use that result as a signal to clarify the underlying data first.

Product reference

Find it from any operating page by clicking Ask W in the lower-right corner. The same panel can answer with workspace context, and pages that support voice show the microphone control beside the message composer.

Where W shows up (decisions, connectors, support, onboarding)

W is contextual. Decisions, Connectors, Support, and Onboarding share one chat surface but load different tools.

In Decisions, W focuses on outcomes, signals, decisions, alerts, and next moves. In Connectors, it diagnoses setup and source health. In Support, it stays scoped to product help.

On docs pages, W stays scoped to documented product behavior. On onboarding screens, W can help capture goals and success criteria before source setup.

Question styles that work well

Use one of these shapes:

• Outcome + person: "What should Sam review for the activation outcome?"
• Outcome + timeframe: "What changed for enterprise onboarding this week?"
• Outcome + tradeoff: "Should we prioritize onboarding polish or source coverage?"
• Source + symptom: "GitHub is connected but no signals are appearing. What should I check?"
• Decision + confidence: "Which source supports the pricing decision?"
• Alert + action: "What should I do with this unresolved conflict?"

Short, specific questions work best.

What to expect from an answer

Useful W answers usually include:

• a direct recommendation or inspection target
• the signals it used
• what is missing or ambiguous
• the next screen or object to open
• one narrow action instead of a broad plan

If W points to a conflict, open the linked decision or outcome before changing direction. If W says signals are missing, add or refresh the source that should contain the answer.

Voice mode (start, mute, transcript, stop)

Open W and use the microphone button to start voice mode. While the session is active, you can mute or unmute the microphone, read the live transcript, and stop the conversation.

Voice mode uses the same workspace context as text chat. If the call ends before a useful result appears, restart voice mode and ask a narrower question.

Voice during onboarding

The onboarding voice flow is available at /onboarding/voice. W captures the goal and success criteria, then writes them into onboarding before source setup.

Use onboarding voice when you can explain the goal faster than you can type it. You can still edit the captured text before generating outcomes.

navigate_ui -- voice that drives the screen

When voice mentions a UI surface, such as "show me alerts", W can return a navigate_ui hint. The page scrolls to the relevant section and highlights it.

Use this when W references a signal or setting that is currently off-screen.

When not to use W

Do not use W to replace source review when the decision is high impact and the signal trail is thin. Add the missing source, review the linked signals, and then ask a narrower product judgement question.

Avoid questions like "what should we do?" without an outcome, timeframe, source, or owner. W will be more useful if you give it one concrete anchor.