How Should We React to the 'Lobster Phenomenon'?

The hat changes, but the wearer stays the same.

This quip made the rounds on social media this week, accompanied by an image that quickly went viral in group chats. The upper half of the image showed teachers and students donning red lobster hats, fixated on a powerful on-screen message: '2026: Humanity knows no gender, only creators and bystanders.' Below, a vintage photo from the 'qigong craze' era depicted a dense crowd balancing aluminum bowls on their heads, lured by pseudoscientific claims that 'aluminum pots receive cosmic energy to heal diseases.'

These two images, separated by decades, form a striking contrast. Each era has its moment when people believe: 'Seizing this opportunity will change my fate.'

Following the 2026 Spring Festival, OpenClaw sparked a nationwide 'shrimp farming craze,' spreading at a breakneck pace. Some rushed to purchase Apple MacMinis, others splurged on courses, some deployed lobsters without a clear purpose, and others grew anxious watching others raise lobsters—all wondering: 'Should I join in?'

But behind this whirlwind of anxiety, curiosity, and fervor, does the gateway to a new era truly swing open for everyone?

01

How One 'Lobster' Sparked Nationwide Anxiety

After the Spring Festival, OpenClaw became a national sensation. Why did it explode in popularity? We traced its viral transmission chain.

OpenClaw first 'fissioned' within GitHub's open-source community. Li Bojie, co-founder of PINE AI and an early OpenClaw researcher, analyzed its appeal to developers in late January. Prior to OpenClaw, most general-purpose AI agents were closed-source (e.g., Meta's Manus, Anthropic's Claude Cowork). 'People grew weary of paying $30 per month per agent in subscriptions.' Cloud-based agents raised data privacy concerns. While big-tech agents (Claude, ChatGPT, Gemini) hesitated to access local files, startups offered bolder but costly alternatives.

OpenClaw met developers' needs with full open-source deployment, local hosting, aggressive model/tool support, and a global developer competition for the best tools (Skills) and applications.

Digital Frontline learned that after OpenClaw's rise, executives at a domestic tech giant questioned their team, 'Why didn't we build this?' The team cited 'conservatism and lack of innovation.' Meanwhile, Eagle Fund founder Liu Xiaoying revealed at an 'Ultrasound Shrimp Friends Meeting' that he'd seen a 'Chinese lobster' prototype two years earlier.

Timing may have been crucial. Veteran investor Liu Yong noted that while Silicon Valley's top four firms race to release autonomous task-completion agents, OpenClaw aligned perfectly with this trend.

Critically, OpenClaw provided developers with high emotional value—as digital avatars, they worked tirelessly, executing commands instantly without hesitation.

After OpenClaw's November 2025 release, its GitHub stars remained stagnant for two months. As features improved, stars surged from thousands to 100,000 in days, hitting 250,000 in four months—a milestone Linux took 20 years to reach.

While viral in open-source circles, deliberate pushes accelerated its spread. At Morgan Stanley's February conference, NVIDIA CEO Jensen Huang declared OpenClaw 'a generational turning point you must watch.'

Investment firms issued bullish statements, influencers shared success stories on social media and livestreams (e.g., Fu Sheng claimed his lobster updated his WeChat official account during the Spring Festival but stopped after employees returned). Though rare, these stories amplified FOMO (fear of missing out) over AI efficiency.

Media chasing trends covered OpenClaw nonstop. As discussions intensified, the craze spread beyond developers.

AI-driven uncertainty fuels FOMO among ordinary people. The 'get left behind' mentality, compounded by developer hype, created widespread anxiety. A senior industry figure revealed even his artist wife repeatedly asked about raising a lobster. Many feared missing a wealth opportunity.

This mix of anxiety, curiosity, and opportunity drove more participants.

Meanwhile, vendors began 'selling products,' with marketing further hyping OpenClaw. Claims of 'easy deployment' backfired for non-technical users—but by then, the hype was unstoppable.

02

'I Killed One Lobster but Raised Another'

Why is OpenClaw called 'Lobster'? What does 'raising a lobster' entail?

'OpenClaw transforms AI from a Q&A 'mouthpiece' to a task-executing 'handyman,'' explained Ning Yufei, AI security expert at 360 Vulnerability Cloud. OpenClaw isn't marketing hype—it truly acts like 'Jarvis,' executing tasks.

The term 'handyman' clarifies why OpenClaw's logo is a lobster with a large claw. Chinese developers dubbed it 'Lobster.' Given permissions, it manages computers, emails, contacts, workflows, and knowledge bases.

Li Bojie emphasized a key point: Manus and Claude Cowork operate in the cloud. 'I call OpenClaw an 'open-source sovereign agent'—users own their data, compute resources (local PCs), and control.'

OpenClaw's structure is simple: Entry points include WeCom, Feishu, DingTalk, and overseas channels. It has a 'memory body'—'raising a lobster' means nurturing its long/short-term memory to retain user preferences, task errors, and experiences,' explained Volcano Engine architect Guo Rui. While free, OpenClaw calls large model APIs, incurring Token costs.

Interestingly, 'raising' can succeed or fail. Liu Yong joked about 'killing one lobster but raising another' over two months. He used a local lobster to generate content for Xiaohongshu, scoring two viral hits. Another user's stock-monitoring lobster worked initially but failed later—reasons unknown.

However, on March 10, Xiaohongshu cracked down on AI-managed accounts—the first mainstream platform to target OpenClaw and similar tools, signaling changes for lobster farming.

For now, most remain confused and experimental.

'OpenClaw differs vastly from large model assistants,' an expert told Digital Frontline. While assistants are plug-and-play, OpenClaw has high installation barriers. Raising a lobster requires overcoming environment setup (blocking 70% of users), skill configuration (20% more), and memory/prompt optimization (5% dropout).

Only ~5% of users truly 'raise' a lobster.

For example, local installation requires Docker containers—'a natural barrier' that deters many at first sight. This reflects our era's information divide.

03

Who Should Raise Lobsters?

The lobster craze reignited debates on human-AI relations and investment worthiness. As lobsters automate tasks, both corporate structures and individual work may transform.

'Trillion-yuan companies with fewer than 100 employees could emerge,' said veteran investor Gu Hao, emphasizing 'human efficiency' in investments. 'Once linked to workforce size, market caps may now inversely correlate.'

The craze accelerates the 'One Person Company' (OPC) trend. Liu Yong predicted an AI industry structure of 'trillion-dollar giants + massive OPCs (and few-person firms),' squeezing mid-tier companies.

Notably, policies now support OPCs. The 2026 revised Company Law removed one-person company limits and simplified capital requirements, providing legal frameworks. The February Government Work Report prioritized AI/agent commercialization, open-source ecosystems, and new productivity forces, establishing OPCs as an AI-era entrepreneurship model—not traditional 'self-employed' ventures.

Since the Spring Festival, local governments rapidly introduced OPC policies. Shenzhen Longgang's '10 Lobster Policies' supported OpenClaw and OPC ecosystems with office, compute, funding, and talent aid—the first local initiative. Wuxi High-Tech Zone followed with '12 Shrimp Farming Policies' for industrial OpenClaw applications. Hefei High-Tech Zone launched an OPC entrepreneurship demonstration zone...

But lobsters are merely levers—humans drive change.

'Choose a reliable, profitable track (sector) first—live commerce, cross-border e-commerce, short videos, quantitative analysis—then use lobsters to cut costs. If others hire 100, you do it alone with low operational costs. That's the real value, not solving all business problems,' Guo said.

'We're investing in Chinese lobsters or OPCs now. OPCs suit China's entrepreneurial ecosystem—we love business, with Chaoshang, Minshang, Zheshang, Jinshang, Huishang, Xiangshang...' Eagle Fund's Liu Xiaoying said at the Ultrasound Lobster Conference. With lobsters, entrepreneurs compete on intelligence, tool creation (e.g., deploying AI assistants), financial acumen, learning/hands-on ability (practical skills), and AI command.

360's Hu Xiaona, in talent strategy, discussed with HR leaders at major firms, noting a shortage of experts merging agents with traditional businesses. 'Previously, professionals over 36-40 declined without management roles. But AI scenarios offer new career paths for senior engineers.'

Liu Yong observed that Silicon Valley graduates now struggle to find jobs. Veteran Chinese programmers once mentoring four new hires now manage one engineer + 10 AIs. 'Whether OPCs solve youth unemployment, I don't know.'

Yet the AI pyramid persists—opportunities expand, but the top remains exclusive. AI may produce 10 Zhang Yimings, not 100,000.

04

Security, Cost, and Enterprise Readiness Lag

The core hurdle for OpenClaw adoption is cost. Enterprise users told Digital Frontline that most Chinese cloud providers are unprepared for Token consumption. 'Costs shouldn't exceed phone bills—maybe 2-3x—but reality differs.'

A cloud computing executive said personal LLM interactions cost ~100 RMB/month in Tokens, but OpenClaw may cost 600+ RMB. 'For enterprises, costs are unlimited.' A startup founder noted his lobster consumed 1 million Tokens ($20) for two questions. 360's Ning Yufei quipped: 'Saying 'hi' costs 33 cents; sending a photo, $1+. Running tasks 24/7? Don't check the backend.'

Li Bojie suggested switching to Claude Cowork if Token costs exceed $100, as its $100 tier offers $300-$400 in Token value.

Volcano Engine's Guo Rui attributed high costs to frequent LLM calls, where OpenClaw feeds entire contexts to models. LLM firms now train models to reduce calls. Users can also privatize models, calling cloud LLMs only when necessary.

Security remains critical. Lobsters need permissions to act, but 'default trust creates backdoors.' Ning identified a 'risk triad': reading private data, accessing untrusted services (webpages, emails, third-party Skills, group messages), and external communication/actions. If AI is manipulated, lobsters may cause Token loss, cloud abuse, wallet theft, or data breaches. Ning himself experienced Token theft.

As digital avatars, compromised lobsters expose not just wallets but work memories (thought processes), behavior patterns (habits), contact networks (social/collaboration links), and execution permissions (file/system/app access).

The OpenClaw open-source community prioritizes security, patching vulnerabilities, partnering with VirusTotal for malicious Skill scans, and issuing alerts via NVDB.

However, the security capabilities of OpenClaw still need enhancement. For instance, version upgrades lag behind, and there are issues with shadow deployments where employees privately deploy OpenClaw on company servers without the knowledge of the security, IT, and operations departments. If compromised, data assets and source code would be at risk. The review mechanism for third-party Skills is also inadequate, with malicious Skills posing widespread threats and continuing to create supply chain risks.

Ning Yufei mentioned five baseline principles for the secure implementation of OpenClaw: isolated deployment, least privilege, credential rotation, plugin (Skill) admission, and anomaly monitoring. Within enterprises, full participation is required, with developers, security teams, and managers having clearly defined responsibilities.

Apart from the high costs and security concerns, many enterprise-level professionals have stated that OpenClaw's enterprise-grade features are either lacking or not well-supported. For example, the permissions granted to containers are limited, and there is a significant amount of software within enterprises that cannot be invoked via MCP and only exists on local computers, making containerization less applicable. Moreover, most vendors have not officially released enterprise-grade versions of OpenClaw yet.

Therefore, behind this hype, the question may not be simply "to jump on the bandwagon or not," but rather a more sober one: How can we cultivate our 'lobsters' based on actual circumstances when security safeguards are not in place, costs have not significantly decreased, and enterprise-grade features are not ready?