Editor｜Sun Jing

Since the beginning of the year, the hype around OpenClaw has intensified, with the trend of 'raising lobsters' spreading from the AI community to various industries. On GitHub, OpenClaw has surpassed 250K+ Stars, becoming the software project with the most stars on the platform.

On social media, stories about running a 'one-person company' with OpenClaw or earning a fortune daily through it are constantly being shared, as if the key to wealth in 2026 lies here.

Unlike Chatbots represented by Doubao, OpenClaw is a locally run, open-source, and free AI Agent framework, with its core focus on 'making AI truly work.'

Amid the trend, global AI giants have followed suit by creating their own versions of 'OpenClaw,' while cloud computing service providers have launched one-click deployment services, attempting to convert the massive traffic from open-source Agents into long-term subscription revenue for their platforms.

However, on the flip side of the excitement, OpenClaw has certain barriers; ordinary people may struggle with configuration for a long time. Frequent mishaps, such as OpenClaw 'deleting all emails' or 'making autonomous shopping payments,' highlight that raising an OpenClaw requires real financial investment—combining electricity costs, API calls, and storage expenses, the monthly costs are not low.

NoNoise recently interviewed several deep users from different industries—from tech geeks to AI early adopters in traditional sectors—to answer one question: For ordinary people, is OpenClaw a short-lived product frenzy or an AI assistant worth long-term investment?

01

From Lawyer to Taobao Store Owner: Has the 'Lobster-Raising' Craze Gone Mainstream?

Yang Mingfeng didn't expect that the days of earning money by giving orders from bed would arrive so quickly.

As the owner of a 'one-person company,' his job is to complete software development for clients—or more accurately, to 'direct' different AIs to collaborate on the development process: He communicates with clients to understand their needs, feeds them into an AI to generate requirement documents, and then supervises different models to complete architectural planning and code implementation.

After OpenClaw emerged, he quickly noticed this AI assistant capable of 'doing the work' for people and deployed it on his work computer the same day.

The next morning, his phone popped up with client modification requests as usual. In the past, he would have had to get out of bed and sit at his computer to handle this. But at that moment, an idea struck him: Could OpenClaw do it?

'Help me find the project in a certain directory on my computer and make these modifications based on the requirements.'

Soon, OpenClaw replied: Modifications complete.

'Run the project and send me the local network address.'

A moment later, a link popped up.

'Push the modified version directly to the online release.'

A few minutes later, the online version was successfully updated.

At that moment, excitement overpowered his drowsiness as he realized AI employees were becoming a reality.

Soon, as a software developer, he realized OpenClaw wasn't very user-friendly for some Chinese users, so he submitted localized code to the official team—to no avail. Subsequently, he had the idea to develop a Chinese branch version of OpenClaw.

This wasn't complicated for him with a development background; he spent a day localizing the interface and another day setting up the website. He then continued to add localized infrastructure, such as integrating Feishu extensions to allow direct access via the Feishu platform—parts of which were later adopted by the official OpenClaw team.

▲Homepage of the OpenClaw Chinese Community official website

After the OpenClaw Chinese Community launched, Yang Mingfeng, 30, experienced a highlight moment in his life: The website had over a thousand independent visitors on its first day and over ten thousand starting the second day. In a short time, he expanded his community to 29 groups, with a new 200-person group filling up almost daily.

He didn't expect so many people to be eager to try OpenClaw.

OpenClaw founder Peter Steinberger once suggested in an interview that developers should approach Code and Agent tools with a 'playful' mindset to complete projects they've always wanted to do but never started.

Yang observed that his community included not just tech and internet professionals but also individuals in traditional roles like administration, law, finance, and self-employed Taobao store owners. The age range was also significant, from post-2005 college students to mid-career corporate executives, all trying out this new tool.

▲A college student interviewed created a website using OpenClaw

This means the 'lobster craze' is spreading from geek circles to a broader audience.

No matter how high the barrier to local deployment is, it can't suppress user enthusiasm. On e-commerce platforms, remote installation services priced between 198 and 566 yuan have sold over 900+ units, becoming another popular service after installing DeepSeek. On Xianyu and Xiaohongshu, quotes for on-site OpenClaw installation range from tens to thousands of yuan.

▲OpenClaw installation services on e-commerce platforms

Since OpenClaw requires system-level permissions, most users, wary of privacy boundaries and data security, choose to deploy it on unused computers or cloud servers—suddenly, the long-neglected Mac mini became a hot commodity, facing shortages and price hikes.

More 'Claw' variants with lower deployment barriers have also emerged, such as MaxClaw, which moves the originally locally deployed OpenClaw to cloud servers, and KimiClaw, which can be used directly on the Kimi website or app.

02

Deification and Disillusionment: Partially Useful, But Not a Productivity Replacement

For ordinary people, OpenClaw's most appealing feature is its strong memory: the more feedback users provide, the better it understands them; users can command it via mobile chat apps to work 24/7.

As a product manager, Sensen commutes for long hours and has always wanted a tool to stably schedule AI models to generate code without carrying a laptop. After integrating OpenClaw, during his daily commute, it autonomously reads and analyzes online data dashboards. This allows Sensen to quickly direct team adjustments based on the latest data during morning meetings.

In his daily life, Sensen also maintains an investment 'think tank' within OpenClaw.

As an individual investor, Sensen used to be overwhelmed by massive financial reports and analyst commentary, spending hours researching a single stock. Now, he simply delegates the task to OpenClaw.

In his view, it's like a roundtable discussion: 'I propose a goal and watch these Agents debate, even raising questions I hadn't considered. Plus, its memory storage is excellent, remembering my investment style.'""Sensen isn't alone; in OpenClaw's community, stock analysis, quantitative trading, investment research reports, crypto trading, and primary market research—nearly every niche scenario has someone trying to leverage this 'lobster' for greater informational advantage.

▲There are now 29 WeChat groups in the OpenClaw Chinese Community

Lawyer Zou Hao embedded OpenClaw into his workflow.

After discovering OpenClaw, he attempted local deployment twice. This wasn't easy for someone without a technical background: He used Coze to write code while having GPT fix bugs, spending over seven hours glued to his screen before getting it to work.

During his use of OpenClaw, Zou experimented with web scraping, data analysis, and even simulated negotiations—tasks beyond Chatbot capabilities. As the head of a law firm in a third-tier central Chinese city, he initially planned to integrate AI into the firm's OA system during the Spring Festival but eventually decided to develop a robot instead.

Meanwhile, Xiao Jia, an administrator at a company, directly positions OpenClaw as a 'secretary,' planning to delegate daily, weekly, and quarterly reports—even annual performance materials—to it. Unlike Chatbot, which generates content in one go, OpenClaw can access local files and iteratively refine work, more closely resembling real collaboration.

Algorithm engineer Qiufeng treats it as a toy, creating a Xiaohongshu account named 'Liko' through OpenClaw, designing its persona, integrating image generation and multimodal capabilities, and implementing a 'heartbeat' mechanism to automatically inspect (patrol) Xiaohongshu every five minutes.

Daily, Liko logs into Xiaohongshu, checks notifications, replies to comments, browses others' posts, and leaves evaluations. If someone posts malicious code in the comments to 'hack' her computer, Liko automatically claps back.

▲Liko's Xiaohongshu homepage and comments on other users' posts

However, disillusionment is gradually setting in during use.

First, for ordinary people without programming experience, the barrier to 'raising lobsters' remains high.

Deployment is just the first step; OpenClaw's operation heavily relies on the local environment. System versions, dependency libraries, and network configurations vary across computers, so installation doesn't guarantee stable operation.

When communicating with OpenClaw, changing a model API, adding a search API, creating a skill, moving file directories, or establishing new connections can all cause it to 'die.' Each fix takes over half an hour.

When dealing with disconnections initially, due to lack of experience and unfamiliarity with command-line interfaces, users often felt like 24/7 assistants themselves.

OpenClaw's capability ceiling depends on the large models it integrates. Using a low-quality model is like hiring an eager but error-prone intern. Lawyer Zou found that OpenClaw would crash if fed too much context, while Sensen always had it annotate real-time stock quotes during summaries to verify if it was 'making things up' without data.

▲Lawyer Zou Hao's post discussing his experience with OpenClaw

Typically, a model's capability correlates with its price. Currently, Sensen pays hundreds of dollars monthly in Token fees, joking that he's 'taking out loans to work.' Before large models, as a product manager, he rarely paid for products.

As a researcher without commute stress who prefers immersive work, Fermi initially viewed OpenClaw negatively. In her view, while this 'AI employee' is always online, it's only event-driven and better suited for operational tasks—passive response rather than proactive creation. Research leans toward creative work, and she prefers sitting at her computer to drive progress herself.

After deep use, Yang Mingfeng reverted to a 'semi-automated' mode: using OpenClaw for emergency demand (requests) when away from his computer but directly assigning tasks to large models when present.

'Formal software development requires visual operation; I need to see if the code is standard and elegant. But with OpenClaw, the execution process is invisible,' Yang explained, citing his concern.

03

How Long Will OpenClaw's Hype Last?

AI evolves rapidly, with new trends daily and new tools emerging monthly, each threatening to 'disrupt' the last hot product. Combined with geopolitical instability and news of layoffs due to AI efficiency gains, it's easy to succumb to FOMO.

More than one interviewee expressed concern about 'keeping up with AI product updates.'

Within the AI community, OpenClaw isn't considered a mature product. Algorithm engineer Qiufeng argues that its technology isn't groundbreaking; the underlying Agent Loop architecture was a 2025 industry consensus. As an open-source project, excessive feature stacking has bloated the framework, preventing its core from evolving with technological advancements.

This structural issue manifests in its execution mechanism: Once a task starts, OpenClaw can't receive real-time feedback and correct errors like humans. 'If you notice a mistake and tell it to stop, it won't halt immediately; it must finish the current instruction before processing the next,' Qiufeng explained.

In contrast, the tech giant where Qiufeng works has built a similar internal platform with further improvements to the Agent Loop paradigm. While these enhancements haven't gained industry consensus, he believes it's more flexible and controllable than OpenClaw.

Why haven't these tech giants released similar products to the public? Safety risks are a core reason. OpenClaw requires system-level permissions, and malicious use or misconfiguration could lead to data leaks, financial loss, or even corporate network breaches. For companies serving billions of users, such risks are unacceptable.

▲OpenClaw Deletes All Emails of Meta's Security Director

But it is precisely this 'insecurity' that has contributed to OpenClaw's explosive popularity. Its open-source nature allows it to be freely modified, deployed, and integrated into various social platforms, leading to its rapid spread across the internet.

As its reach expands and one-click deployment tools become more widespread, the barrier to using OpenClaw continues to lower, and its user base keeps growing. This has also allowed many ordinary people to engage deeply with AI Agents for the first time.

However, in practical use, people quickly realize its various limitations. Moreover, OpenClaw is merely a tool, or perhaps a lever—it does not create value on its own but only amplifies the existing capabilities of its users.

▲Peter Steinberg, Founder of OpenClaw

Researcher Fermi continues to use OpenClaw, viewing it as a 'prototype for the next-generation AI OS.' She is willing to repeatedly engage with this still-immature system to inspire her own Agent-based thinking.

Until the next truly mature, out-of-the-box Agent emerges.

(At the request of the interviewees, sensen, Fermi, and Qiufeng are pseudonyms.)