[jjcm]

I've done a lot of stage demos. I've done pre-recorded ones, I've done live ones.

I prefer live ones, by a significant margin, even for non-deterministic AI things. In general I find live ones better, but it does depend on your circumstance. It's worth asking yourself three questions when you're doing a demo:

What is the outcome when it goes great?

What is the outcome when it goes wrong?

What percentage of the time do each of those occur?

With a pre-recorded demo, when they go great it's usually a 9/10. Things are solid, but they aren't adaptive to the situation, which means you might have lost an opportunity to insert a quip from something that happened earlier. When they go wrong though, it's a 0/10. You've told people that you don't have enough faith in your product to show it live - why should they have faith in it?. Now you might say that the demo goes right 100% of the time(ish), but there's also a human element to it. Someone might play the demo early, someone might not act in time with the live demo, any of these can give away that it's pre-recorded. I'd put their success rate at 95%.

With a live demo, when they go great they're 10/10. You can adapt them, you can show it live, things react instantly instead of slightly-out-of-sync due to scripted content. It's incredible. When they go wrong, it's a 3/10. You've got egg on your face. People make fun of you after. But they know at that point that you ran a live demo. That does count for something. We know Zuck did that demo live. They do fail a bit more, especially in the age of AI. 10% failure rates are pretty good for live demos. To me though, the higher failure rate is worth it for the better PR if you do fail.

[jadbox]

100% agree. It's not a BINARY choice though! I prefer to do the demo live, and if it fails, have a one-push switch over to a recording and quickly declare that live projects are a work in progress. I think this is ultimately how you show both strength of live demo preparation, but can humbly recover the show if the live version fails by switching to recorded.

But, NEVER TELL ME IT'S A LIVE DEMO WHEN ITS PRE-RECORDED. You will lose my business forever if I catch it.

[tschwimmer]

I don't agree that a bad live demo is a 3/10. The recent demo showed me conclusively that the Meta Ray Bans are not at the quality level where I would buy them, especially not for $800. That's pretty much a 0/10 in my book. Since it was live usage, it's indicative of the real quality of the product.

[layer8]

That live demo, despite the failures, unexpectedly made Meta much more likable to me than (say) the sterile and cringy prerecorded Apple events, which are just completely off-putting.

Furthermore, look at how the prerecorded Apple Intelligence demo at WWDC 2024 turned out, where more than a year later we still don’t know if and when its most salient features will see the light of day. Whereas with the Meta demo, we can assume that they were reasonably confident it would work on stage, and the failure may really just be due to the flaky wifi at the venue.

[Royce-CMR]

Yeah I agree. But I’ll split it slightly.

From a trust perspective I want a real demo. Technical team, deep dive, concept sold now talk me through it type stuff.

From a “show non-tech executives the art of the possible as part of 2026/2027 planning” I want a recording.

Failure in the latter isn’t a 3/10, it’s a -10/10.

[bee_rider]

IMO a perfect prerecorded demo would have to be shockingly good to get better than like a… B- for me. It’s just too easy to cut or reshoot the interaction to be perfect.

[qwertytyyuu]

Given Tesla’s track record of completing faking their pre recorded demos, I don’t trust them at all. As a consumer live demos all the way

[SOLAR_FIELDS]

In another world we would call that fraud. But here it's business as usual.