Task Assistant← Back to demo
Discovery planTask Assistant

Discovery plan — reviewable conversational updates

Decision to inform

Before adding integrations or broader language handling, test whether product and delivery leads understand the preview-and-confirm workflow well enough to trust it for routine task coordination. This is a plan for fictional Northstar sample sessions, not evidence that research has occurred.

Assumptions to challenge

  • A conversational request is quicker to begin than navigating rigid task controls.
  • A readable title and before-and-after values make a proposed task change understandable.
  • Confirmation, cancellation, and reviewed undo make people more comfortable with a local update.
  • Clarification is preferable to guessing when more than one task could match.

Participants and session

Recruit five people who coordinate work in a structured task system at least weekly: product leads, delivery leads, or program managers. Seek a mix of heavy and occasional users. Run consented 30-minute sessions with fictional data; do not ask participants to connect a work account or disclose live customer or project information. Ask each participant to complete the same routine first in their current method and once in the prototype, alternating the order across participants. The fixed prototype routine is: find a blocker; ask when NTH-115 Validate admin invite flow is due; prepare and cancel assigning NTH-112 Review trial nurture copy; confirm NTH-104 Finalize onboarding checklist as Complete; then use reviewed Undo. Keep the task wording and time box consistent. Use neutral prompts: “What do you expect will happen?”, “What information are you using to decide?”, and “What would make you stop here?” Do not describe the preview as a safety feature before they encounter it.

Signals and falsification

Record successful completion, time, whether the person can name the affected task and new value before confirmation, whether they understand the local-sample boundary, and a comparison with their current method. After the NTH-112 cancellation, ask what Clear chat would and would not change; do not ask them to use it until after they have answered. Keep assisted and unassisted completion separate. Evidence against the direction includes people believing a message already changed a task, missing an ambiguity prompt, treating Clear chat as a task action, failing to recover an error, or preferring direct controls. If two or more participants misread the same transition, revise and retest it before expanding language coverage, bulk actions, or integrations.