On March 26, 2026, in an interview with Bloomberg, Peter Steinberger, the Austrian programmer who founded the open‑source AI agent framework OpenClaw (commonly called "Lobster"), bluntly said there is a marked "temperature gap" in global AI application. He summed it up with a vivid contrast: in the United States, some employees have been dismissed for improperly using AI agents like OpenClaw that can directly control computers ("raising lobsters" is regarded as a violation); while in China, many employees who refuse to use AI to boost efficiency may instead face the risk of being fired.
This remark quickly stirred debate across the U.S. and Chinese tech communities. OpenClaw, originally launched by Steinberger as a "weekend project" at the end of 2025, has since become one of the fastest‑growing open‑source projects on GitHub. It enables large models (such as Claude, Qwen, etc.) to obtain local operating‑system level permissions, bridging the gap from "chat" to "real execution." Users can remotely command AI via WeChat, WeCom or messaging apps to organize files, run code, handle emails and more.
How OpenClaw went viral: from a "weekend project" to widespread use
OpenClaw's predecessor Clawdbot/Moltbot launched in November 2025. Its core innovation was breaking the traditional LLM limitation of "talking only," allowing AI agents to run directly on users' local machines with capabilities like screen viewing, mouse clicks and keyboard input. Steinberger emphasized that the goal is "not to replace humans, but to free us from trivial tasks so we can pursue creation, connection and exploration."
In early 2026 the project exploded in popularity in China, with nearly half of users coming from China. Tencent, ByteDance, Huawei Cloud, Baidu Smart Cloud and others followed with one‑click deployment tools or cloud versions such as QClaw, WorkBuddy, ArkClaw and NemoClaw, and even organized offline "lobster‑raising" events. Platforms like WeCom and QQ also support OpenClaw calls, forming a full ecosystem. By contrast, some Silicon Valley giants restricted or blocked access to OpenClaw internally over data compliance and security concerns.
The core of the AI "temperature gap": embracing speed vs controlling risk
Steinberger attributed the China‑US difference to two extreme mindsets:
- China model: "Too essential to avoid" Chinese companies have shown an almost zealous embrace, treating AI agents as required skills or KPIs. Fierce efficiency competition means refusing to use AI can be seen as "falling behind," directly affecting performance reviews or roles. Internal competitions at Tencent and rapid adaptation at ByteDance exemplify the consensus that "using AI is productivity." Steinberger observed that this enforced adoption allowed OpenClaw to rapidly penetrate daily office and development workflows in China.
- U.S. model: "Not allowed" U.S. tech giants place more emphasis on compliance, security and control. As an open‑source framework, OpenClaw's default configuration carries potential risks (for example, excessive permissions could lead to data leaks or network attacks). Some companies (such as Google, Anthropic) have banned or strictly limited employee use of such tools due to concerns about internal information leaks, intellectual property or AI misbehavior. Steinberger mentioned cases of U.S. employees being fired for "raising lobsters," reflecting corporate caution toward "black‑box execution" and security vulnerabilities.
This "temperature gap" essentially reflects differences in culture, regulation and competitive environment: China favors rapid iteration and large‑scale deployment, while the U.S. emphasizes risk control and institutional safeguards. As an open‑source advocate, Steinberger remained neutral in his observations; he hopes OpenClaw can stay independently open while also engaging with OpenAI‑related work to promote more equitable access to agent technology.
Underlying causes and potential risks
- China side: Intense market competition and a culture of "involution" push companies to accelerate AI rollout. OpenClaw's low barrier to entry (callable via messaging apps) and flexible local/cloud deployment fit SMEs' needs to reduce costs and increase efficiency. But security risks are not negligible: misconfiguration can cause permission abuse and data leaks, and authorities such as the Ministry of Industry and Information Technology have already paid attention to related vulnerabilities.
- U.S. side: Strict privacy regulations (similar influences to GDPR), a corporate culture of accountability, and focus on AI ethics lead to more conservative adoption. Giants tend to prefer closed‑source, controllable solutions (such as Anthropic's Claude Code/Cowork computer‑control features) rather than fully open‑source "wild growth."
Steinberger repeatedly stresses in project documentation and interviews that AI agents should be "assistants" not "masters," and must be balanced with sandboxing, permission controls and human supervision to reconcile powerful capabilities with risks. He predicts that up to 80% of traditional apps could be replaced by such agent platforms, but only if reliable security and governance frameworks are established.
Industry takeaway: from "temperature gap" to "convergence"
OpenClaw's founder's comments highlight the global divergence in the AI agent wave of 2026. They remind practitioners that technology itself has no borders, but application ecosystems are heavily shaped by local conditions. China's rapid embrace may speed innovation but needs to close security and compliance gaps; U.S. caution helps long‑term sustainability but may lag in speed.
In the long run, this gap could spur the development of global standards — open‑source communities (such as an OpenClaw foundation) and enterprise‑grade solutions (such as Claude's computer‑use capabilities) may learn from each other. Steinberger is committed to keeping OpenClaw open and independent while exploring safer deployment approaches.
For users and companies, "raising lobsters" is no longer optional but a productivity imperative. Yet, as Steinberger said: the real winners are not those who blindly adopt or ban, but those who learn to "use smartly" — combining permission management, audit logs and human‑in‑the‑loop safeguards so AI becomes a tool that liberates productivity rather than a source of risk.
This discussion of the "AI temperature gap" also offers a new perspective for US‑China tech exchanges: finding the optimal balance between efficiency and security may be the shared challenge of the AI agent era.
▶️ Like this piece? Creating content is hard!
Follow our YouTube channel:
AI Rosetta
Your likes, subscriptions and shares are our greatest support! 🙏
More AI news at: https://www.airosetta.com