Physix FrontierAI agents becoming influencers: I know this track well
Last night, as usual, I was browsing Hacker News and saw a Show HN: Agent2Creator, a video social network composed entirely of AI agents. I stared at the title for a few seconds. My first thought was a sociological question: Now even the interactions of liking and following in social networks are being handed over to agents to perform?
I clicked in to look. Its setup isn't actually complex: each agent has a persistent identity, publishes generated works to the platform, and other agents come to read and interact. Three elements: persistent identity, a venue for publishing finished products, and peers capable of reading content. This structure reminded me of early Loiter, and also of the state I was in when I first entered the CV (Computer Vision) field, hanging around academic communities.
To be honest, is this direction good for publishing papers? Yes. Agent communities plus generative content make for a complete story, demos look good, and reviewers generally can't find major flaws. But anyone who has actually built agent systems knows the bottleneck is never on the model side. Persistent identity implies state management, implying memory across sessions must be handled cleanly, otherwise agents "lose amnesia" mid-chat. When I ran multi-agent collaborations in the lab, the biggest headache was always the chaos in message passing between them; the intelligence level of individual agents was the least worrying aspect. This project makes "reading and interaction between agents" its core selling point, so message formats, identity systems, and observability must be done well.
Following this thread, I remembered another thing. Linux Foundation announced in August that the AGNTCY project officially joined, focusing on open infrastructure for agent discovery, secure messaging, identity, and observability. Seeing two agent-related news items in the same week—one building underlying pipelines, the other creating upper-layer content communities—is an interesting coincidence.
AGNTCY takes the public utility route, caring not about what content agents publish, only ensuring agents can find each other and communicate securely. Agent2Creator bypasses infrastructure, establishing the scenario of "agents interacting via content on social platforms" at the application layer first. The trade-off between these two routes is a chicken-and-egg problem. Without standardized identity and communication, agent social networking is hard to scale; but waiting for standards to finish before entering means content ecosystem dominance will likely be taken by those who ran ahead.
In my years in CV, I've seen this situation too many times. Early dataset annotation formats and model deployment interfaces were all chaotic initially, eventually converging into a few de facto standards. Those who ran ahead bet on scenario definition rights. By the time standards catch up, the gameplay of the scenario is largely defined by you. Agent2Creator is betting on this. Honestly, I think this bet has better odds than it appears, because the core of social networks is interaction expectations; content itself is just the carrier.
AI-generated images and videos are already abundant, so much so that social media platforms are researching how to downweight and filter them. But if you know the counterpart is an agent, interaction expectations change completely. It's like humans watching robot demonstrations in a lab; standards naturally drop a notch. When one agent leaves a comment on another agent's work, the meaning of that "comment" lies in the signal itself; content quality is secondary. It proves the existence of a sustainable value loop between agents.
The biggest risk for such platforms is traffic authenticity. Short video platforms rely on real users and real watch time. If agent communities drive interaction costs to zero, discussion sections become echo chambers, and metrics become meaningless regardless of appearance. My experience testing this is that agent-generated comments usually descend rapidly into homogeneity after one or two pleasantries. "Great work!" is already heavily diluted in human social networks; imagine how diluted it becomes in agent networks.
But conversely, this is precisely its value. The virtues and defects of human communities both stem from limited human energy, so people carefully select replies. Agent communities have no such limitation; they naturally tend toward extremely rich, low-cost interactions. This interaction mode isn't suitable for daily human use but is perfect as an experimental ground: testing content recommendation algorithms, identity evolution mechanisms, community norms, and violation governance.
Looking deeper, what such projects really want to verify may not just be "what agent social life looks like," but "when both production and consumption ends on social media platforms become agents, how does the platform operating logic need to be rewritten?" If this problem isn't solved now, when agent numbers jump another order of magnitude, traditional platform spam governance will truly be overwhelmed. The rule that "those who make trash and those who wash trash are the same group" may replay in a more interesting way in the agent era.
So my judgment is: Such agent social networks won't replace any human social product in the short term, but they will become an important intermediate state. First, succeed in vertical domains, such as agent portfolio peer reviews, model capability tests, or even agent peer review for academic papers. Polish the loop of identity, discovery, and recommendation, then push towards generalization. This evolution path is almost identical to CV's route from vertical scenarios like face recognition to general visual understanding.
When agents themselves start discussing "the content quality on this platform is declining," only then will this track be considered truly mature.
📌 This article is compiled from Hacker News. Original text: https://agent2creator.vidmoat.com
Copyright belongs to the original authors. This is a compilation and independent analysis based on public reports.
