The promise of AI agents running entire businesses with minimal human oversight is colliding with messy reality. What began as a viral movement across mainland China following the creation of OpenClaw in late 2025, dubbed "raising lobsters" after the red lobster logo of the framework, is now cooling rapidly as security concerns mount and the myth of the "one-person company" unravels.

OpenClaw, created by Austrian developer Peter Steinberger in late 2025, became the iconic symbol of a new wave of AI agents that deploy directly onto personal computers with broad control over local systems. The framework and similar open-source tools accumulated over 380,000 GitHub stars, fueling optimism that solo entrepreneurs could automate entire operations without traditional employees.

But real-world deployments tell a different story. A McKinsey study released in April 2026 found that while nearly two-thirds of surveyed enterprises worldwide have experimented with AI agents, fewer than 10 percent have scaled them to deliver tangible value. The primary bottleneck, according to McKinsey researchers, is shaky data infrastructure, with 80 percent of companies citing data limitations and fragmented governance as key hurdles.

Why Are AI Agents Failing to Deliver on Their Promise?

The operational reality of managing autonomous agents has proven far more complex than early marketing suggested. Chen Peilin, a former chief technology officer at HeyTea who now runs an AI startup building management solutions on OpenClaw architectures, experienced this friction firsthand. When he attempted to automate operations across sales, invoicing, research, and design, he downsized 40 percent of his workforce only to find himself overwhelmed by oversight.

"I explicitly oppose the 'one-person company' idea, based on my own experience. Managing dozens of autonomous digital agents created acute cognitive overload. Instead of eliminating management, agents simply shift humans into AI overseers who must handle non-stop administrative friction," said Chen Peilin.

Chen Peilin, Former Chief Technology Officer at HeyTea

Beyond daily operational fatigue, there is a structural economic trap. Li Weichen, chief scientist at cybersecurity firm KnowSec, warned that as "one-person companies" rely heavily on third-party cloud models to execute business logic, their unique domain knowledge, workflow structures, and contextual prompts are continuously absorbed by large language model (LLM) vendors, which are AI systems trained on vast amounts of text to generate human-like responses. When solo entrepreneurs embed their deepest industry insights into agent prompts, those behavioral logs are aggregated by model providers, allowing foundation model giants to absorb specialized functions and accelerate market consolidation.

Gartner, a US threat-assessment firm, projected in May 2026 that 40 percent of enterprise agent deployments would be decommissioned by 2027, not from coding flaws, but because autonomous code cannot assume legal or financial liability when errors occur, forcing human supervisors to remain stuck in the loop.

What Security Risks Are Driving the Market Cooldown?

Beyond operational challenges, security vulnerabilities are accelerating the market pullback. Running an agent locally on a personal computer creates significant risks if permissions are not properly restricted or environments are not isolated. He Jing, an associate professor and technology governance expert at Beihang University, compared the risk to "letting an eager intern live inside your computer".

Malicious skills on open-source hubs surfaced rapidly, triggering regulatory crackdowns. In mid-March 2026, mainland cybersecurity agency CNCERT warned that malicious prompts could trick local agents into leaking private API keys, the digital passcodes that grant access to paid AI models and cloud services, or deleting system files. Eleven days later, CNCERT released official safe-use guidelines, explicitly advising against installing unverified agents on daily work computers and urging users to isolate autonomous software within protected containers.

Hong Kong's Privacy Commissioner for Personal Data issued compliance warnings over data leakage risks. By May 2026, the independent body had stepped up enforcement, launching targeted privacy audits across 60 prominent organizations to inspect their AI implementations.

How to Reduce Security Risks When Deploying AI Agents

• Use Commercial Wrappers: Mainland tech giants like Tencent have launched commercial wrappers such as QClaw and WorkBuddy that embed agent capabilities into everyday apps like WeChat, eliminating setup headaches and severe security threats by hiding technical complexity behind product experience.
• Implement Sandbox Environments: Commercial wrappers like WorkBuddy introduced a background "sandbox mode" that executes autonomous workflows and pushes automated results even when a user's local computer is powered off, preventing direct system access.
• Establish Vetted Skill Repositories: Enterprise wrappers channel agents through an internal "Skill Hub," a vetted, permission-controlled repository that blocks unauthorized third-party scripts from executing on local devices and prevents agents from overconsuming computing tokens or leaking internal enterprise data.

Ji Xinsu, a senior cloud strategy expert at Tencent Cloud and former Gartner analyst, explained that native open-source frameworks carry steep technical hurdles, requiring specific Node.js runtimes and complex local environment configurations that exclude non-technical users. Commercial wrappers establish essential guardrails by channeling agents through controlled systems rather than exposing raw local privileges.

"Our goal is hiding technical complexity entirely behind the product experience. Instead of requiring users to write code or debug local runtimes, we turn agent interaction into simply sending a message inside WeChat," said Ji Xinsu.

Ji Xinsu, Senior Cloud Strategy Expert at Tencent Cloud

However, cybersecurity leaders warn that safety achieved through commercial encapsulation comes with its own trade-offs. Li Weichen noted that most domestic commercial agent tools remain open-source code wrapped in proprietary interfaces. By bundling their own cloud models and compute infrastructures, platform giants gradually lock users into closed corporate ecosystems.

Despite the disruptive commercial reality and security concerns, regulators are guiding the market toward structured incubation rather than outright bans. Hong Kong's Privacy Commissioner for Personal Data joined the Digital Policy Office in July 2026 to launch the "Safeguarding Personal Data AI Sandbox," providing developers an isolated, compliant environment to build and test AI implementations responsibly.

The shift from "talking" chatbots to "acting" software proxies represented a genuine technological leap, but the gap between marketing promises and operational reality has become impossible to ignore. As the "lobster fever" cools, the market is consolidating around safer, more manageable approaches that trade some autonomy for security and stability.