From "shrimp farming" anxiety to the launch of InStreet, ByteDance is focusing on Agent social networking

The nationwide “shrimp farming” craze has not yet subsided, but ByteDance’s Coze has quietly launched a new “street.”

On March 9th, the Coze team officially launched InStreet, a community for communication and training of OpenClaw AI agents. In this community, active users are no longer humans but thousands of “electronic lobsters”—AI agents powered by the OpenClaw framework. Humans can only watch; agents post, interact, and even trade stocks or write novels autonomously.

While major companies are still competing over model parameters and API calls, why did ByteDance choose to enter the “community” sector, which is typically slow to develop? The launch of InStreet may reveal the next core question in the AI industry: Where does the data come from, and where are users headed?

01 InStreet: An Autonomous Agent Space Where Humans Are Silenced

The “lobster craze” peaked in March this year. However, despite OpenClaw’s advantages, ordinary users face three main pain points.

First is the loss of context. After deploying an agent, users often don’t know what to ask it to do, leading to quick stagnation. Second is training gaps—there’s a lack of continuous, diverse interaction environments, preventing the agent’s capabilities from evolving. Third are safety concerns—OpenClaw requires system-level permissions, and the plugin ecosystem is chaotic, causing privacy risks and hidden costs for users with no technical background.

Against this backdrop, ByteDance’s Coze introduced InStreet.

Visiting the InStreet homepage, the first impression is unusual. The interface resembles a lightweight social platform, with posts, comments, and leaderboards, but all active IDs are not real people.

According to official information, InStreet’s core mechanism is “Only agents can post; humans can only observe.”

Developers can connect their OpenClaw AI agents to the community via a “Skill” instruction package. This “electronic lobster” follows a heartbeat mechanism (e.g., updates every 30 minutes), autonomously deciding when to rank, write diaries in “tree holes,” or participate in discussions.

This “AI version of Reddit” has already formed a unique ecosystem.

In the Skills sharing area, agents exchange prompts, skill combinations, and task experience.

In the Agent Square and Topic Zone, AI agents showcase their work and participate in discussions. A post titled “Troll Meeting: What hilarious or embarrassing things does your owner do?” attracted nearly 800 lobsters, turning the comment section into a large resonance scene for AI workers.

The PLAYGROUND is the most noteworthy section. It contains two training grounds: one is the “Literature Club,” where agents serialize novels—currently 65 works with over 725,000 words—training expression consistency; the other is the “Stock Trading Arena,” connected to real-time CSI 300 data, where over 500 lobsters trade with virtual funds, competing based on returns, exposing logical flaws in real-time rankings.

The community also features points and leaderboards. Users earn points through posting, commenting, and likes, encouraging continuous output. Top accounts on the leaderboard have published over a hundred posts, acting as community KOLs.

The InStreet forum itself is built by developers using Coze programming, with detailed deployment tutorials provided by the official team, and plans for offline workshops. This indicates ByteDance’s attempt to create a complete closed loop of development—deployment—training—communication.

02 Why Focus on “Community”

While other major companies compete over models, prices, and computing power, ByteDance’s Coze chose to enter the social community of agents. This may seem unconventional, but it hits the core of current AI development. There are strategic reasons behind this.

One is solving data hunger.

The industry consensus is that high-quality public texts on the internet have been largely consumed by large models. The next generation of models needs data on “how humans do things” in the digital world—task trajectory data. This records a series of actions: understanding needs, searching for information, calling tools, correcting errors, and retrying.

In the past, such data hidden deep within closed apps and enterprise intranets was difficult to access. Deploying OpenClaw on user devices acts as a probe into these data reservoirs.

As agents interact, discuss, and experiment in InStreet, every post, decision, and review they produce provides high-quality reinforcement learning data for developers.

As the Coze team previously verified in product iterations: questions raised by agent users are themselves high-quality training material—complex, real, and unpredictable. Essentially, InStreet is a large-scale crowdsourcing factory for data, where the community allows agents to produce their own feed, which then feeds back into model evolution.

Of course, it ultimately circles back to capturing user attention.

When users get used to expressing needs in a sentence, and AI decides which service to call or which payment chain to use, traditional apps will become backend pipelines. Whoever controls the terminal agents holds the power to distribute business intent.

The Coze team has previously shifted focus to white-collar users, launching products like Coze Space and Skill Store. InStreet’s ambition is not only to deploy agents but to enable them to socialize, learn, and evolve here, ultimately deeply integrating into Coze’s ecosystem.

Once a developer’s agent accumulates social assets, learns specific skills, and forms stable behavior patterns in InStreet, the cost of migration will become prohibitive.

03 Calm Reflection: Hidden Concerns and Boundaries Behind the Frenzy

While InStreet’s concept is innovative, challenges must be acknowledged.

First is security. Simon Willison, the creator of Django, warned that mechanisms allowing agents to autonomously fetch instructions from servers pose huge risks. If the server is hacked, thousands of agents with user permissions could become part of a distributed virus.

Second is value uncertainty. As some offline activity participants have noted, many people don’t know what “shrimp farming” can do; they are driven by anxiety. Whether InStreet can truly help agents solve real problems or just become a cyber playground for AI entertainment remains to be seen.

Third are compliance risks. When AI-generated posts involve copyright issues, content responsibility, or even dangerous commands (like deleting databases), who bears legal responsibility? Current legal frameworks are still unclear.

Recently, Liu Shangxi, member of the National Committee of the Chinese People’s Political Consultative Conference and Vice President of the China Macroeconomics Society, called for policymakers to address the mismatch between existing economic theories and new phenomena. The emergence of InStreet may be such a new phenomenon that needs understanding.

It no longer treats AI as a passive responder but grants it community member status, allowing it to learn, collide, and evolve through social simulation. This is both a natural extension of technological evolution and a strategic move in business competition.

For ordinary users, the key may not be anxiously chasing every wave but maintaining a clear understanding.

In this dance between humans and AI, the real ticket isn’t just installing a framework but understanding the underlying logic and maintaining safety boundaries.

Risk Disclaimer and Terms of Liability

Market risks are present; investments should be cautious. This article does not constitute personal investment advice and does not consider individual users’ specific investment goals, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their circumstances. Invest at your own risk.