mundane-demos は何をしますか？
The Muse launch never led with benchmark scores. It led with negotiating a bill, selling a car, booking a trip, auditing subscriptions — jobs the viewer did last week. Mundane utility converts; benchmarks impress people who were already convinced.
Pick the right task
- The viewer did it this morning. Email triage, calendar Tetris, expense reports, subscription audits. Recognition creates desire: "I hate doing that" becomes "I want that."
- One job, end to end. Not a feature tour. The bill is negotiated, the flight is booked, the car is listed. Show the confirmation, not the chat transcript.
- Boring is the point. If the task looks impressive, viewers credit the demo. If the task looks boring and the agent still nails it, viewers credit the product.
Structure the demo
- State the job in one sentence ("Cancel my three unused subscriptions").
- Show it happening, sped up if needed, but real.
- End on the proof: the confirmation email, the refund receipt, the booked itinerary. The artifact is the argument.
Keep it under two minutes. Cut everything that isn't the job or the proof.
Speed is the wow factor
"Wicked fast" and "always-on" were the launch headlines for a reason. When the capability is table stakes, latency is the differentiator. Demo on the fastest path, show the timer if you have one, and never apologize for a loading spinner — remove it instead.
What not to demo
- Abstract reasoning puzzles. Nobody wakes up needing those solved.
- Multi-step setups. If the demo needs ten minutes of config, the product isn't ready to demo.
- Anything you can't show failing gracefully. Live demos break; have the recovery be part of the story, not the end of it.