Pre-record your demos(steveharrison.dev) |
Pre-record your demos(steveharrison.dev) |
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.
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.
But, NEVER TELL ME IT'S A LIVE DEMO WHEN ITS PRE-RECORDED. You will lose my business forever if I catch it.
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.
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.
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?
I do live demos at work all the time and the best part is that when someone asks "can it frobulate with encapulation?" I can actually press the frobulate button with encapulation enabled and show them. Can't do that with a recording.
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.
Luckily it was a small group of devs and not a large venue. Still embarrassing...
> I prerecord all my live demos
> People are always impressed at how flawless my live demos are
> the i̶l̶l̶u̶s̶i̶o̶n̶ deception is perfect.
Seriously... just call it a demo or pre-recorded. As soon as you call it "live" you've crossed into the territory of fraud.
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
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?
I'm not sure what the problem is here. Customers don't want buggy products. Live demos are more informative to users who can't tell what the actual experience will be like vs what the envisioned experience is.
I'm sorry, but as a customer I don't care what's in a developer's dreams, I care what I can buy.
It's like expecting reality shows to be a 100% unfiltered feed of real life. Their purpose is to entertain, not to document.
> What's more, It's a Live is actually running the commands you're typing, so you have full interoperability with other programs.
So the actual 'demo' part is actually live. The only thing that's fake is the typing, which could be replaced honestly with copy & paste without changing anything except how exciting it is to see the speaker pretend to type at 100wpm.
With a teleprompter (or queue cards or notes) you are still speaking the words.
But with this system you are not. So it is more like a slideshow where your voice is pre-recorded per slide and you are pretending to speak.
I would still call a teleprompter a live setting. I wouldn't be surprised if Zuck had one. People can still stumble and maneuver and adapt in real time. Just like in a play. Rehearsal does not make the act not live. But if we want to get more fine grained there is a difference between a live demo and an on-the-spot demonstration. Think of it this way, people don't knock politicians for using teleprompters because they are lip syncing the words. They knock them for being overly rehearsed and not off the cusp. The complaint is actually beyond the teleprompter, the teleprompter is just a visual indication of this. But then imagine how people would respond if a politician went up on stage and pretend to actually speak but the words coming out were pre-recorded.
Or maybe we can meet more in the middle with something like autotune? I still don't think it is there because you'll still need to say the right words for autotune to work but even still people generally don't like that and feel like it is deceptive. So I might be being strict here, but we need to at least recognize that all these concerns arise from some form of deception. Where the expectation of what's happening doesn't meet reality.
No one will be upset if they never find out, but the real question is how they respond when they do. Does the person you're demoing to feel cheated, even a little, if you used a tool like this? I'd wager they would because they thought you were doing all the actions there in front of them.
Then that's documentation with extra steps.
> Your comment only makes sense if you think that the purpose of the demo
The purpose of a demo is to demonstrate things.
The purpose of a live demo is to demonstrate how things will work in the user's actual experience.
You can have pre-recorded demos and there's nothing wrong with that. The only issue is calling a pre-recorded demo "live". That's when you cross the line into deception.
Or as I said before
>>> Seriously... just call it a demo or pre-recorded. As soon as you call it "live" you've crossed into the territory of fraud.
> Their purpose is to entertain, not to document.
> "live" is showing it while both the demonstrator and the audience is in a room together.
> I'm not 100% sure
If it is live, it is being done in real time.
If it is pre-recorded it was done previously.
If you call a pre-recorded demo "live" then it is fraud.
What more is there to get?
This is very different from practicing a script and you actually doing all the actions in real time. You're allowed to rehearse for your live performance but unless you're doing it live then it's not live.
People also hate lip syncing because it's a deception of being live.
> Every time you press a key, It's a Live will write one character from the file into the terminal, making it look like you're typing every single command with the practiced ease of a consummate professional.
Good god, this isn't that hard man. You're fine to lip sync if you call it lip syncing. Just don't claim you're actually singing when you're not and you're all good. Why is this so difficult to grasp?