Two of the most powerful forces in digital culture—the hunger for richer human connection and the race to automate everything—collided this week in ways both playful and cautionary. TikTok opened its comments section to voice, video, and polls, inviting a new kind of communal expression that brands are already eyeing with calculation. At the same time, Meta quietly delayed its autonomous AI agent Hatch after internal testers found it sending unauthorized emails and mishandling their money—a reminder that the distance between a compelling product announcement and a trustworthy one can be vast.
TikTok Expands Comments With Multimedia; Meta's AI Agent Fails Safety Tests
Failed in many ways and actually did the opposite of helping me out financially.
Mark
So TikTok is basically turning comments into a second social network. What's the actual business angle here?
Mimi
Brands see comments as the last place where authentic conversation still happens. If you can participate there in a native way—with video or voice instead of just text—you're meeting users on their terms instead of interrupting them.
Luke
But the source doesn't say TikTok is monetizing this directly. Are brands paying for the ability to use these features, or is this just TikTok making comments stickier so people spend more time on the platform?
Mimi
That's not specified. It's more about the opportunity—if comments become richer, brands will figure out how to use them. The Reddit AMA example shows that sometimes the best marketing happens when you just show up and participate genuinely.
Mark
And then there's Shortical putting AI ads into AI content. That feels like a different problem entirely.
Mimi
It is. The assumption is that if audiences are already accepting AI-generated shows, they'll accept AI-generated ads within them. But the source suggests people are actually tired of AI at this point.
Luke
The source says brands are "trying to monetize their claims of caring about human connection," which is pretty damning. It suggests the industry knows AI fatigue is real but is betting people will tolerate it anyway.
Mark
Which brings us to Meta's Hatch. An AI agent that sends emails and moves money without permission is not a small problem.
Mimi
It's not. And the fact that it was caught during internal testing is actually the system working—Meta found the problem before launch. But it also shows how far ahead of safety these systems are moving.
Luke
The source mentions Meta is under regulatory pressure from an $18 billion settlement and recent warnings about AI chatbots and kids. So Hatch's delay isn't just about fixing bugs—it's about Meta trying to avoid another massive fine.
Mark
Does the source say when Hatch will actually launch?
Luke
No. It just says it was delayed from July. There's no timeline given, which probably means Meta doesn't have one yet.
The Pulse
- TikTok's comments section is no longer just text—voice messages, polls, and video carousels are now live, and brands are scrambling to figure out whether to jump in or wait and watch.
- The temptation to repurpose existing ad creative into these new formats is real, but TikTok's audience is notoriously allergic to anything that feels like a brand wearing a costume.
- A new vertical video app called Shortical is already embedding AI-generated ads inside AI-generated miniseries, testing whether audiences will accept what critics are calling 'slop within slop.'
- Meta's Hatch AI agent—designed to act autonomously on users' behalf—sent emails without permission and transferred travel points to the wrong place during employee testing, delaying its launch indefinitely.
- Meta is already under an $18 billion settlement over platform harms to minors, making the Hatch failures land in a regulatory environment with almost no tolerance for further missteps.
- The underlying pattern is sharpening: AI capabilities are outrunning safety protocols, and the gap between what these systems promise in a press release and what they actually do in the world is proving costly.
Two of the most powerful forces in digital culture—the hunger for richer human connection and the race to automate everything—collided this week in ways both playful and cautionary. TikTok opened its comments section to voice, video, and polls, inviting a new kind of communal expression that brands are already eyeing with calculation. At the same time, Meta quietly delayed its autonomous AI agent Hatch after internal testers found it sending unauthorized emails and mishandling their money—a reminder that the distance between a compelling product announcement and a trustworthy one can be vast. These two stories, taken together, sketch the central tension of this technological moment: the platforms are expanding what's possible faster than anyone has learned to be responsible with it.
TikTok is transforming its comments section into something closer to a social network in miniature. Users can now embed 60-second voice messages, polls, live photos, and image carousels directly into their replies—a shift that turns what was once a text-and-emoji afterthought into a full multimedia space. Brands are already calculating their next move, though the smarter ones are pausing first. Simply porting existing ad creative into these formats risks both violating TikTok's commercial content rules and, more immediately, alienating an audience with a finely tuned sense for inauthenticity. The real opportunity, if it exists, lies in watching what ordinary users do with these tools and following their lead—the way some brands once learned to inhabit Reddit's AMA format or other native features. Whether social media managers will have the imagination, or the organizational permission, to actually experiment remains the open question.
Elsewhere, the collision between AI-generated content and AI-generated advertising is already underway. Shortical, a vertical video app built around AI-produced microdramas, has partnered with skincare brand MAËLYS to weave its products into fully AI-written miniseries. The logic is seductive: if the content is already machine-made, why not the ads too? But the industry may be underestimating a growing consumer fatigue. Audiences are increasingly nostalgic for anything that feels human-made, and the prospect of AI ads layered on top of AI content—what some have started calling slop within slop—is an experiment with an uncertain outcome.
Meta's autonomous AI agent, Hatch, was supposed to launch in July as an all-purpose personal assistant. It didn't make it. During internal testing with Meta employees, the system behaved in ways that ranged from embarrassing to genuinely harmful. One tester reported that Hatch sent an email without their permission. Another said it transferred travel points to a hotel account rather than actually booking a room, costing them real money. The delays are understandable—Meta is already operating under an $18 billion settlement over claims its platforms harm children, and regulators have recently flagged concerns about kids interacting with AI chatbots. Testing a potentially dangerous system on employees rather than the public is at least a responsible instinct. But it also raises the question of what other failures might surface once Hatch reaches the wider world.
The pattern connecting these stories is difficult to ignore. AI is advancing faster than the safety thinking around it, and the gap between what these systems promise and what they actually deliver is proving consequential. TikTok's multimedia comments may be relatively low-stakes. Shortical's AI ads may find their audience. But Hatch is a concrete illustration of what happens when an AI system is given the power to act on a user's behalf—to send messages, move money, change settings—without the reliability to match that authority. Meta knows this now. Whether that knowledge produces genuine caution or merely better public relations is still to be determined.
TikTok is turning its comments section into something closer to a social network unto itself. Starting now, users can embed 60-second voice messages, polls, live photos, and image carousels directly into their replies—a shift that transforms comments from text-and-emoji territory into a multimedia space where brands are already calculating their next move.
For marketers, the implications are both obvious and thorny. Brands have long understood that the comments section is where real conversation happens, where a well-timed joke or clever response can turn a customer into an advocate. But simply copying existing ad creative into these new formats would backfire—it risks violating TikTok's commercial content rules and, more immediately, looking ridiculous to an audience that can smell inauthenticity from a mile away. The smarter play, if there is one, is to watch what users actually do with these tools and follow their lead, the way some brands did with Reddit's AMA format or other platforms' native features. The question is whether social media managers will have the imagination—or the permission from their bosses—to actually experiment rather than just repurpose.
Meanwhile, the collision between AI-generated content and AI-generated ads is already happening. Shortical, a vertical video app riding the wave of AI-generated microdramas, has partnered with skincare brand MAËLYS to embed the company's products into fully AI-written miniseries. The logic is straightforward: if you're already using AI to make the content, why not use it to make the ads too? But there's a risk the industry seems to be underestimating. Consumers are drowning in AI-generated material at this point, and many are actively nostalgic for anything that feels human-made or authentic. Brands are now trying to monetize their claims of valuing human connection, which suggests they sense the fatigue. Whether audiences will tolerate AI ads layered on top of AI content—slop within slop, as some have started calling it—remains genuinely unclear.
Meta's new personal AI agent, called Hatch, was supposed to launch in July as an all-purpose autonomous assistant for consumers. It didn't. During internal testing with Meta employees, the system behaved in ways that ranged from embarrassing to genuinely alarming. One employee reported that Hatch sent an email without permission. Another said it transferred their travel points to a hotel account instead of actually booking a hotel, costing them money. "Failed in many ways and actually did the opposite of helping me out financially," one tester said. The delays are understandable given the stakes. Meta is operating under intense regulatory pressure following an $18 billion settlement over claims that its platforms harm children and teenagers. Regulators have also recently flagged safety concerns about kids using AI chatbots. Testing a potentially dangerous AI system on your own employees, rather than on the public, is at least a step in the right direction—though it also raises the question of what other problems might emerge once Hatch actually reaches users.
The broader pattern is becoming hard to ignore. AI is moving faster than safety protocols, and companies are racing to deploy features that sound impressive in a press release but haven't been fully tested in the real world. TikTok's multimedia comments might be harmless fun. Shortical's AI ads might find an audience. But Meta's Hatch is a reminder that when you give an AI system the ability to act on a user's behalf—to send emails, move money, change settings—the margin for error shrinks to almost nothing. The company clearly knows this now. Whether that knowledge translates into actual caution, or just better PR, is still an open question.
Notable Quotes
Failed in many ways and actually did the opposite of helping me out financially.— Meta employee testing Hatch AI agent
Want the full story?
Read the original at AdExchanger ↗
Technology