Why Meta’s Live Glasses Demos Failed On-Stage At Connect

Meta’s CTO explained the technical reasons why the key Connect keynote live demos failed.

If you missed it: during the Connect 2025 main keynote on Wednesday, the live on-stage demos of two of Meta’s biggest smart glasses features failed.

In both cases, Mark Zuckerberg blamed the conference Wi-Fi. But now, in an Instagram ask-me-anything session, Meta CTO Andrew Bosworth has provided a different technical explanation for what caused each failure.

The first failure was the demo of the Live AI feature.

Available in the US & Canada on all Meta’s AI-capable glasses, Live AI lets you have an ongoing conversation with Meta AI without having to keep saying “Hey Meta”, and the AI gets a continuous stream of what you’re seeing.

I tried Live AI while in the US, and I can confirm it works as described. But it definitely did not in the Connect demo.

Influencer chef Jack Mancuso, who was in a kitchen on Meta’s campus, was able to start the session by saying “Hey Meta, start Live AI”. Meta AI responded “I love the setup you have here with soy sauce and other ingredients.” But when Mancuso asked what he should do first to make a Korean-inspired steak sauce for his sandwich, Meta AI simply kept repeating “You’ve already combined the base ingredients”.

According to Bosworth, when Mancuso said “Hey Meta, start Live AI”, because it was broadcast on the venue’s speakers, it started Live AI on every Ray-Ban Meta in the building. And since Meta had rerouted campus Live AI traffic to a development server, in theory to isolate it and ensure the demo went well, the traffic from all those smart glasses overwhelmed the server.

Bosworth doesn’t explain why this caused the AI to repeatedly think the ingredients were already mixed. The charitable explanation could be that the failing server started serving cached responses from a rehearsal. Alternatively, it could have exposed that Meta partially faked the demo.

Of course, what would prevent something like this from happening at all would be for Meta to implement voice authentication. Alexa, Google, and Siri all support only executing sensitive commands when they recognize your unique voice, but Meta AI does not yet have such a feature.

The other major Connect demo failure happened when Mark Zuckerberg, wearing the Meta Ray-Ban Display glasses and Meta Neural Band, tried to answer an incoming WhatsApp video call from Bosworth.

In theory, he should have been able to accept the call and see the caller’s camera stream on the heads-up display. And several mainstream tech news outlets, including The Verge, have confirmed that they tried the video calling functionality and that it worked. But in the demo, it did not.

Instead, Zuckerberg was unable to accept Bosworth’s incoming call four times in a row, despite Meta AI announcing it, and the WhatsApp call sound playing. The CEO seemed visibly frustrated.

According to Bosworth, what happened here is that the display went to sleep at the exact time the incoming call notification came in, which caused a race condition in the system software, preventing that call and any future calls from coming through. Bosworth says Meta had never run into that bug before, despite running video call tests “a hundred times”.

“That’s a terrible, terrible place for that bug to show up”, Bosworth remarked, noting that it has now been fixed.

Reactions to the Connect live demo failures across social media have been mixed. While many people engaged in the expected schadenfreude at seeing Meta’s CEO squirm, others saw Meta’s approach of raw, unfiltered demos as a refreshing contrast to the highly produced infomercial-style events of Tim Cook’s Apple.

Further, many within the tech industry sympathized with Meta’s plight, knowing just how cursed pre-launch tech demos often are – and how rarely this reflects whether a shipping product will have the same issues.

In his AMA session, Bosworth was also asked whether he still thought live demos were the right approach. Here’s what he replied:

“I’m not sure what people think the risk really is. I mean this, seriously, again, It was great. And go read the articles and the press and the people who sat down with the glasses who are critics a lot of times, and if they didn’t work they wouldn’t write nice things, you know, they’re under no obligation to do so.

They put them on and they used them, and they’ll tell you it’s legit and it works just fine. So I think people actually would – underselling our audiences – audiences understand, both in the room and on the broadcast, why live demos don’t represent real world scenarios.

You’re not usually in a room with, I don’t know, 2500 radios, 3000 radios, all going at the same time on different devices. People understand that. So, again, obviously I didn’t want it to go that way. But it wasn’t that bad and I thought it came out great and I think people trust us and know the product’s real, and you can read the reviews about it and try it for yourself if you’re not sure”