Pre-record your demos (opens in new tab)

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.

I would never recommend doing this in a sales context though, where you're selling to prospects. Mainly because it gives you no flexibility to pivot and adjust the content based on what they're interested in, and also because it gives off the impression you don't have a lot of faith in your product to perform live.

I'd still 100% recommend having pre-recorded backups however. I was once doing a demo to our largest customer in region and the second we joined the call our core product had a SEV0. I was very relieved to have something to show while the engineers fixed the issue.

What I dislike is presenting something as if it is live when it isn't. Demos are fantastic things and it is even perfectly fine to have imaginative aspects to them. But there's a division line between demoing and lying. Take Rabbit R1 as an example[0]. This whole thing was faked yet it was presented as a live demonstration, not an illustration. In other words, fraud.

The difference is if you're trying to show your customers (or potential customers) what you're imagining the device works like vs what the device actually works like. The difference is how you're communicating the difference to them.

I'm not sure why this is a contentious opinion, but just don't lie to people?

[0] https://www.youtube.com/watch?v=22wlLy7hKP4

I asked my professor what it was like, was he an incredible professor? He told me the lectures were useless! I was shocked and he said they were so well prepared they were effectively a recording of the material - and no advantage over reading the book. At that point I reflected and realized that, by far, my favorite professor was one who constantly made mistakes while solving problems on the board. He'd have to notice something wrong and track down the error, and it was wonderful and highly instructional to watch the process.

https://gitlab.com/stavros/itsalive

I can even pause the replay mid-demo, take over, and resume again later. People are always impressed at how flawless my live demos are, and with some judicious audience participation, the illusion is perfect.

But when the standards are lower, people appreciate the live demo. Videos are much less compelling, and they could've watched that at home in a few minutes, rather than fly out to you.

In grad school -- in a lab where, instead of "publish or perish", our mantra was "demo or die" -- we were giving demos most weekdays, usually to only a few people at a time. Always live. I think once I had to play a video of someone else's stuff, and it fell flat.

My best demo story is when I was showing a new thing (just a class project), to a few people crowded around a computer in my lab area at an open house event, typing into my software agent in a chat window (pre-LLMs), ad-libbing and speaking aloud what I was typing, "I'm interested in... thinks Microsoft and... thinks Oracle..." Then I turned my head around to look at the people behind me, to see that they were following, and I swear to gosh, Larry Ellison himself was standing in the doorway of my lab area, giving me this fiercely aggressive stare. Whether he was glaring because I said Oracle or Microsoft, I do not know.

My worst demo story was at the very start of my demo career. I had just arrived on campus, and I was borrowing a computer in an open area. A research scientist comes up, and sees on my screen this Web-scraping personalized newspaper thing I had made. So I'm showing it to them, and they're interested, then all of a sudden they say, sadly, "Mother Theresa died..." Which was what my "demo" was showing.

So maybe don't include an uncontrolled live data feed of depressing or distracting news in your demo.

Also, implementing early software agents in Java apparently summons Larry Ellison.

Meta took that risk and failed.

- record the screen as you type out all of your code as you would do live

- don't need the audio, just the video

- edit it down to remove typos and pauses

- speed it up to a nice fast typing pace

Now you can drop the video into Keynote or whatever, and Play/Pause the video as you speak.

A crucial part of this is that the video has effectively no pauses. Maybe just a few short ones between sections, to leave time to hit Pause before the next keystrokes appear.

Instead of a 30 minute video that I had to perfectly keep pace with, it was like 5 minutes of rapid typing that I could speak along with and pause when it was time to explain something.

As a customer, when you show me a video, I default-assume you're lying.

Also, as a presenter: If your live demo fails and you can't turn that into a win, maybe you're not good at presenting. Your audience wants you to succeed, it takes a minimal amount of charm to make it work. (Heck, even Mark turned this into a win, and he's not exactly a charm bomb)

> Since most of the time you're demoing something you previously worked on, this prevents you from working on current stuff right up to the demo, because you're worried you might break your local setup for the demo by installing different node modules, switching to different Git branches, etc.

This isn't the case for me generally... probably more of problem for web-apps things I guess.

I keep multiple workspaces locally for the multiple things I'm working on, but that hasn't been my pain point. Usually it's that I'm demoing a work in progress (not my choice generally) that has rough edges.

I enjoy the internal demos where questions can be asked about doing XYZ, and that can be shown if it's quick. Recordings tend to take away from that, and be less memorable/engaging I guess.

All so that I could demonstrate that when I turned off a RAID controller a red light came on and the application just…continued.

So much work for something largely unspectacular meant that a recorded demo that could be run anytime from a laptop was far more appealing. My junior colleagues were now doing demos they otherwise wouldn’t, and the seniors turned off the sound and edited the video into snippets they could introduce and talk around

https://github.com/sloria/doitlive

Apple can pull it off, but if you’re at a mid market tech company you just sunk like weeks people’s time into choreographing a live movie. Super stressful time sync.

Audience knows when something is live

But anywhere the demo gods can show up, for example in live coding during a demo, can be partially or fully recorded, but narrated in person.

If you aren’t doing a live demo, it’s not a demo, it’s just a sizzle reel.

> That’s what we get for doing it live!

And if you believe that, I've got a bridge to sell you...

It makes me question the rest of this article when the author takes that at face value. A DDOS made your LLM say the same thing over and over? Yeah, I don't think so. Also, this new tech, how many device that support “Hey Meta start Live AI” even exist or were "listening" to the stream.

This is about a valid as Zuck saying that it was "Wifi" issues. How often do you have Wifi issue where you can talk to remote servers just fine?

No one wants to see a terminal, that's for sure, so don't display code or start pecking away at a terminal unless you really think you'll need it.