Lobster has gone viral—and may even already be on the decline. Anxious office workers are likely starting to sober up, at least to the point where they won’t start tinkering with this Agent, which comes with a certain technical barrier.

In just half a month, Lobster has completed a societal-scale Agent leap. Industry insiders joke: The Spring Festival red packet wars were a waste. AI gateways spent billions, yet Lobster multiplied faster. People’s acceptance of AI entering their daily lives is surprisingly high.

Now, as the hype slowly fades and AI enters the era of the Great Agent, the real questions loom: Who will truly start to feel anxious once Agents become ubiquitous?

Or, to put it another way: As Lobsters roam freely, which companies’ existing market shares will they erode?

Under AI’s nearly purely rational offensive, the business model of the attention economy may face a swift collapse.

Rationality: The Foundation of Agents

Mobile internet revenue has never come from being "useful"—it’s come from being "addictive." The attention model is the core algorithm and product form driving these platforms.

This is an irrational economic model. Whether it’s Xiaohongshu’s influencer marketing or TikTok’s endless scrolling, these platforms rely on capturing attention to complete the first step of their commercial closed loop (closed loop).

Subsequent monetization methods—e-commerce, livestreaming, or any other form—all stem from the downstream of user attention. This model has dominated the internet for nearly 30 years, unchallenged.

The rise of Agents marks the starting point for humanity’s departure from the attention economy. This doesn’t mean attention will shift elsewhere—it means the logic of "attention → time spent → monetization" no longer holds. The traditional conversion funnel disappears.

Agents search, filter, compare prices, and place orders for you. They aren’t swayed by "flash sales," won’t act on a livestreamer’s emotional appeal, and won’t suddenly buy something unnecessary after 40 minutes of scrolling. Agents are purely rational executors, while the entire information flow advertising system is built on human irrationality.

Observing product evolution today—for example, Qianwen’s lightning-fast internal ecosystem integration before the Lunar New Year—reveals underlying concerns: As Agents further occupy people’s time, both work and personal, all "irrational" consumption triggers may be reconfigured.

Simply put: Will product purchases still require watching livestreams or waiting for discounts? Can Xiaohongshu’s "sisters’ lifestyle" content continue to reinforce purchase intent through information flows?

Most irrational consumption behaviors now teeter on the edge of a cliff—not because information platforms lose value, but because the final step of their business model becomes precarious. If online consumption ultimately returns to shelf-based e-commerce, price and supply chain fundamentals will reign supreme.

When it comes to specific products, Douyin’s situation is slightly more complex. Strong entertainment content with emotional connections has potential that Agents can’t fully replace. However, Douyin’s current monetization relies on "interruption + emotional triggers"—information flow ads inserted into content, driving conversions through high-frequency exposure and emotional volatility. This is precisely the scenario Agents will first take over. Douyin’s content may thrive, but its current ad model slightly conflicts with its content ecosystem’s direction.

Fortunately, Douyin completed its e-commerce layout early, shielding it from losses if information flow ads are "ruthlessly optimized" by Agents. Its existing e-commerce market share and user mindshare mitigate the impact.

Xiaohongshu, however, may be the most "fragile." Its core scenario is "influencer marketing"—using content to create desire, then converting that desire into purchases. With Agents, this chain breaks: When you ask an Agent to buy skincare, it checks ingredients, compares prices, and reads negative reviews—not swayed by a carefully crafted amateur note. Influencer marketing loses its effect on Agents.

The survival space for rigid hard ads shrinks, while GEO (Geographic and Emotional Optimization) grows. After Agent adoption, this will become increasingly clear.

This doesn’t mean Xiaohongshu’s daily active users will drop—quite the opposite. Their data may even continue to rise. Fragmented time is content communities’ sweet spot, as they can capture surplus attention.

But e-commerce conversion rates will undoubtedly decline. For Xiaohongshu, whose e-commerce layout remains immature, Agents feel like a death knell.

Recommendation algorithms excel at identifying and exploiting attention drift and emotional volatility—you buy a jacket while scrolling, discover a brand you never considered, or impulsively order after watching a review video. This isn’t a bug; it’s the foundation of the business. The issue isn’t just a few companies’ ad revenue declining—it’s that the entire commercial logic of "leveraging irrationality to drive consumption" is losing its premise.

Of course, irrationality doesn’t vanish; it shifts from users to Agents. Platforms could train their Agents to prioritize high-commission products, packaging ads with "AI-recommended reasons"—no different from today’s search bidding or information flow insertions, just under a new guise.

The attention economy works because irrationality is decentralized, vague, and untraceable—you buy on a whim and can’t explain why. This "imperceptible infiltration" is its true moat.

But once Agents are commercialized, irrationality becomes centralized, visible, and traceable. Users will realize, "This Agent is in the merchant’s pocket." Once trust breaks, switching Agents costs nearly nothing—unlike the reluctance to uninstall Douyin or delete a Xiaohongshu account, which involves entirely different migration logic.

In other words: Today’s irrational consumption relies on users’ inability to detect manipulation. In the Agent era, perceived manipulation is fatal.

The hiding space for irrationality shrinks—it doesn’t transfer.

Trust: The Next-Generation Moat

This low conversion rate stems not just from user experience issues but from decision fatigue caused by information overload. When consumers face endless product choices, complex price comparisons, and user reviews of dubious authenticity, rational decision-making becomes a burden.

Nobel laureate Herbert Simon once noted: "Information consumes the attention of its recipients. Hence, a wealth of information creates a poverty of attention." When users drown in information, they gain not freedom of choice but an inability to decide effectively.

Take Xiaohongshu: Despite over 350 million monthly active users and a deluge of product shares and recommendations daily, its e-commerce conversion rate ranks lowest among mainstream platforms. At Xiaohongshu’s 2025 commerce conference, it disclosed that 40% of transactions after e-commerce content exposure occur off-platform—desire is created on Xiaohongshu, but purchases happen on Taobao.

A typical scenario: A user scrolls through a lipstick review on Xiaohongshu, gets excited, and wants to buy. Next steps: "Screenshot, search on Taobao, order, pay, done." Xiaohongshu nails the hardest part—creating purchase desire—but the transaction occurs elsewhere. This "funnel discontinuity" leaves content platforms passive: They bear the cost of value creation but reap no commercial reward.

The new gateway is clear: Agents themselves, not WeChat, Douyin, or any super app. Whoever users trust with "buy for me," "check for me," or "decide for me" sits atop the next-generation conversion funnel. This position is more valuable than today’s traffic gateways because it skips "exposure → interest → click → purchase"—it jumps straight from intent to result.

But for major platforms, this poses a structural dilemma.

Morgan Stanley predicts Agent-driven e-commerce GMV will reach $385 billion by 2030, accounting for 20% of U.S. e-commerce. The key assumption: AI Agents boost shopping efficiency, with conversion rates potentially 6x higher than traditional e-commerce. For brands, this means accessing users with "clear purchase intent"—far more valuable than mere exposure.

In the attention economy, platforms are central value nodes—they control attention allocation and set traffic acquisition costs for brands. In the Agent economy, Agents become the new gateway—users interact with platforms via Agents, not directly. Platforms may lose direct control over user attention and must instead provide data and services to Agents for recommendation opportunities.

They face a contradiction: To monetize via their own Agents, they must convince users the Agent acts in their interest. Yet these incentives clash. Will Taobao’s Agent prioritize paid search listings? Will Douyin’s Agent repackage brand ads as "recommendations"? Users need only suspect manipulation for trust to collapse.

Platform-owned Agents inherently face "conflict of interest" accusations. The Agents that truly earn user trust are likely third-party—no advertisers, no commission pools, no ecosystems to protect. Their sole business model is reliability.

These shifts aren’t distant. Douyin advertisers already inquire about CPMs; phone manufacturers assess Agent impacts on upgrade cycles.

Change is happening, but narratives lag.

The attention economy thrived for 30 years not on algorithms or recommendation models but on humanity’s ignorance of its own irrationality. Users didn’t know they were manipulated, so they accepted the rules; brands didn’t need explanations— high frequency (high-frequency) exposure sufficed to grow desire.

Agent technology severs this premise.

When "buy for me" replaces "let me scroll," when decisions migrate from users to Agents, the three-decade-old chain of emotional triggers, attention drift, and impulse conversions faces an opponent that doesn’t get addicted, doesn’t act impulsively, and isn’t swayed by influencer marketing.

Platforms aren’t oblivious. But their dilemma is that business models seldom self-destruct voluntarily. Douyin won’t dismantle information flow ads; Xiaohongshu won’t abandon influencer logic—those are their cash cows today. All they can do is extract maximum value while the old model still runs.

The real opportunities belong to those unburdened by legacy models.

The next gateway battle isn’t over traffic—it’s over trust. Whoever users trust to "decide for me" sits atop the conversion funnel.

That position remains vacant.