On 9 December 1968, Douglas Engelbart sat on a stage at a computer conference in San Francisco and, for ninety minutes, in front of about a thousand people, demonstrated the mouse, hypertext links, windows, version control, video conferencing, and two people editing the same document at the same time.
The computer was thirty miles away in Menlo Park. They ran a leased line with custom modems for the data and two microwave links for the live video. In 1968.
It’s still called the Mother of All Demos, and it might be the best product demonstration anyone has ever given. The system he demonstrated, NLS, never became a product people bought. Within a few years the team had scattered, a lot of them to Xerox PARC.
A thousand people watched the future work, live, without a single bug, and it did not create demand. I think about that whenever a founder sends me a six minute screen recording and asks what I reckon.
A demo is proof, and proof is not desire
A demo answers how does it work, will it fit my stack, is this real. Those are excellent questions. They are also the questions somebody asks in the last ten minutes before they commit, and they are close to useless to a person who hasn’t decided they care yet.
On launch day, almost everybody watching hasn’t decided they care yet.
You can’t hear your own tapping
There’s a piece of research from 1989 by Camerer, Loewenstein and Weber with a perfect name: the curse of knowledge. Once you know something, you lose the ability to model not knowing it.
The version everybody remembers came out of Stanford the following year. Elizabeth Newton had people tap out the rhythm of well known songs on a table while a listener tried to name the tune, and asked the tappers to predict how often they’d be understood. They guessed about half. Listeners got a tiny fraction of that.
The tapper hears the whole song. Orchestration, lyrics, the chorus arriving. The listener hears knuckles on wood.
That is precisely what happens when the person who built the product records the demo.
Adding explanation subtracts understanding
Richard Mayer has spent decades running experiments on how people learn from a screen with a voice attached, and one of the findings is the one nobody wants: narrating text that is already on screen makes comprehension worse, not better. Two channels carrying the same payload compete for the same limited attention.
The instinct when a demo isn’t landing is to add. Add a callout. Add a subtitle. Add a line explaining the thing the viewer is already looking at. Every one of those makes it harder.
As a designer this is a familiar failure, because it’s the same one as a settings page that grew helper text under every field. The fix is never more explanation. It’s fewer things to explain at once.
The demo that failed live and sold anyway
On 21 November 2019 Tesla unveiled the Cybertruck. Franz von Holzhausen threw a metal ball at the armoured glass to prove it wouldn’t break. It broke. He threw another one at the second window. That broke too.
Five days later Musk said there were 250,000 preorders.
If demonstration created demand, that night should have killed the truck. It didn’t, because the demand wasn’t coming from the demonstration. It came from a shape people had already decided they wanted, from a founder they’d already decided to believe, in a category they already had opinions about. The demo was theatre on top of a decision that had already happened.
The most famous demo in tech was mostly theatre
The January 2007 iPhone keynote is the demo everyone wants to copy. Fred Vogelstein later reported what was actually going on up there. The phone couldn’t reliably survive the sequence Jobs performed, so the sequence was fixed: a specific path of actions in a specific order, rehearsed until it held. There were several prototypes on the podium and he moved between them. The signal indicator was rigged to show five bars no matter what. AT&T brought in a portable tower.
It’s remembered as a demo. It was a performance with a demo inside it, built by people who understood that proof of function was not what the room needed.
The room already hated its phone. That’s the part that made it work. The opening minutes were spent on the problem, on the state of phones as everyone in that auditorium already privately experienced them, so that by the time a list scrolled under a thumb the audience wasn’t evaluating a feature. They were watching a resolution.
So when does a demo sell?
When the viewer already has the problem and knows they have it.
Developer tools are the clean case. If somebody has just lost a week to a deployment pipeline, you don’t need to open with a story about the modern software landscape. Show the command. Show the output. They will finish the video in their head before you do. The desire was already installed and all you owe them is proof.
This is why the blanket advice that everything needs a story annoys me. If your audience is four hundred engineers who all lost the same Thursday to the same bug, a story is you spending the first thirty seconds of a conversation they were ready to have.
The good demos have a question in them
Blendtec started putting things in a blender on YouTube in October 2006. A man in a lab coat holds up an object, asks whether it will blend, and then finds out.
That is a demo, and it’s also a complete story, because there’s a question in it and the answer only arrives at the end. You can’t leave. That’s the whole mechanism.
Hitchcock explained the same idea to Truffaut with a bomb under a table. If it goes off with no warning you get a second of surprise. If the audience knows the bomb is there and the characters don’t, the identical scene is unbearable for ten minutes. Same event. The difference is whether the audience is holding a question.
Most software demos open by announcing the answer. “Here you can configure your workspace.” Now there’s nothing to find out, and the viewer is watching somebody else use a computer, which is one of the least interesting things a human being can watch.
What your six minute demo is actually judged on
Redelmeier and Kahneman published a study in 1996 on how patients remembered painful medical procedures. What predicted the remembered experience wasn’t total pain, and it wasn’t how long the procedure lasted. It was the worst moment and the final moment.
That has an unpleasant implication for demos. The flat four minute stretch in the middle where you tour the settings panel isn’t neutral filler people politely sit through. If it’s the worst part, it is disproportionately what they walk away remembering about your product. And whatever happens to be on screen when the video ends carries far more weight than its length deserves.
Which is a decent argument for ending on the result rather than on “and that’s the dashboard.”
Order, not replacement
None of this means fewer demos. It means the order is the entire decision.
Desire at the top, where the strangers are. Proof one click deeper, where the interested go looking. The long walkthrough after signup, where somebody is already trying to reach their first real result and genuinely wants to know which menu it’s under.
Often it’s the same footage. Different cut, different length, different job. The six minute walkthrough that fails as a launch asset is frequently excellent as the second thing somebody watches. It was never bad. It was early.
A demo answers a question the viewer has to already be asking. Everything before the demo exists to make them ask it.
Engelbart’s audience understood every single thing he showed them in 1968. They just weren’t people with that problem yet. It took the rest of the world about twenty years to catch up to that demo, and by then somebody else was selling it.