The topic of "raising crawfish" has recently become difficult to avoid. Social media feeds are filled with people posting their "crawfish-raising diaries," nearly a thousand people queued up outside the Tencent building, and some are even willing to pay 500 RMB for an on-site installation service, all in a rush to get their system operational.

For many, the temptation stems from a common vision: the idea of "one person equaling an entire team." The story of Fu Sheng during the Spring Festival holiday is often cited as the emotional trigger for this wave of enthusiasm. While bedridden with a fracture, he exchanged 1,157 messages totaling 220,000 words with OpenClaw over 14 days. He transformed a novice who "couldn't even check a contact list" into an automated team comprising eight Agents, even resulting in a "crawfish" autonomously publishing a tweet at 3 a.m. that garnered millions of reads.

It sounds like an automated money-printing machine. However, the narratives of viral success最容易忽略的恰恰是成本、门槛与风险。You might think you're installing the same "crawfish," but it could be entirely different. You might assume the cost lies in the software, but the real expense is in Token consumption. You might believe spending 500 RMB buys you an early-adopter advantage, but the value could depreciate to zero within three months.

Fu Sheng's "legend" is real, but it's very expensive. In a recent livestream, Fu Sheng admitted that his daily spending on OpenClaw exceeds $100 USD—approximately 700 RMB. And this is just for personal use. He integrated the crawfish into his company's Feishu platform, cultivating multiple Agents like advisors, copywriters, and operations officers; the daily Token consumption for the entire system is far greater. More crucially, these results aren't achieved simply by "installing it." Fu Sheng's output is built upon capital investment, top-tier model configuration, and dozens of days of intensive training. His core advantage is the ability to precisely deconstruct tasks, write Skill documentation, and design feedback loops—the real "secret weapon" is human capability, not a default, out-of-the-box "crawfish."

Treating such a case as a replicable template for the average person is like using F1 racing car acceleration data to convince a newly licensed driver to hit the track immediately. The outcome is often not "taking off" but "crashing."

Token Black Hole: The Unaffordable Crawfish What many fail to anticipate is that the cost of OpenClaw lies not in the software itself, but in the underlying model calls. It is inherently a "Token black hole"—every task execution consumes vast amounts of Tokens through interactions with backend large models. Once task chains lengthen, tool calls increase, or memory is enabled, consumption rapidly escalates. An ordinary chatbot might use only a few hundred Tokens per exchange, while OpenClaw executing a similar task could require millions of Tokens.

Some users report that searching for information and writing a 2,000-word document can burn through 7 million Tokens; running a simple crawler test consumed 29 million Tokens; cases of single-day Token consumption reaching 50 million are not uncommon. One SaaS company even provided a crawfish subsidy for all employees, with average staff consuming 150 RMB worth of Tokens daily, while the technical team's consumption reached 1,000 RMB. More insidious is OpenClaw's built-in "heartbeat mechanism"—even with no actual output, the system automatically incurs about 145 RMB in daily call fees, translating to over 5,000 RMB in monthly losses.

This sensation of "burning money" resonates strongly with overseas users as well. Some crawfish raisers have reported weekly Token costs as high as $1,500 USD, roughly 10,381 RMB. X user @Kekius Sage posted a screenshot lamenting, "This is my first day using OpenClaw, I'm really getting old." He merely had the crawfish read a few recent research papers, and the daily bill reached $22.1 USD. He mused in the post that continuing at this rate could lead to daily deductions of $1,000 by next week.

The "Crawfish" You Install Might Not Be the Same One Many are unaware that OpenClaw's capability ceiling depends heavily on the deployment path chosen. Its accumulation of nearly 250,000 stars on GitHub is indeed exciting. It's not just a chatbot but a digital executor "with hands": capable of opening browsers, reading/writing files, sending emails, and controlling social media accounts. These capabilities are real.

However, the reality is: your chosen deployment method determines just how much it can actually do. The mainstream deployment paths currently available roughly fall into four categories: 1. Local Dedicated Hardware (e.g., a dedicated Mac Mini): Offers the highest capability上限, able to read files, control browsers, manage calendars/email, with more complete context. The cost is also the most substantial: hardware investment starts at 4,000 RMB; configuring a high-performance workstation to run local large models and摆脱API dependency can easily exceed 100,000 RMB. Even using cloud models, the primary long-term variable cost remains API call fees. 2. Cloud Server Deployment (one-click solutions from Alibaba Cloud, Tencent Cloud, etc.): Monthly fees from tens to hundreds of RMB seem moderate, but the "crawfish" is更像被关在"空房间"—it lacks access to your files and authorized accounts, inherently severing much of its potential functionality, making it more like a supercharged chatbot. 3. Direct Installation on Personal Computer: Lowest barrier to entry, highest risk. Because it shares permissions with the operating system, misconfiguration or malicious skill injection can compromise not just the "crawfish" but the data and accounts on your entire computer. Security firm data shows over 40,000 OpenClaw instances are exposed on the public internet, with over 60% containing exploitable vulnerabilities. 4. Vendor-Hosted Solutions (e.g., KimiClaw, MaxClaw): Plug-and-play, with almost zero deployment barrier. However, "vision" and autonomy are constrained by platform rules, and both capability上限 and data autonomy are limited.

Major vendors' pricing strategies create a particularly convincing illusion of "low门槛": * Moonshot AI's KimiClaw Vivace membership costs 199 RMB per month. * Alibaba Cloud bundles top-tier models into subscription services, with a Pro version at 39.9 RMB for the first month and 100 RMB thereafter, providing 90,000 API requests monthly. * MiniMax's MaxClaw directly hosts OpenClaw in the cloud, claiming no need for personal servers, with basic functions accessible via the "MiniMax Agent Basic Subscription."

But a closer look at the math reveals these packages are essentially "limited buckets." 90,000 API requests might seem vast, but once the crawfish initiates multi-Agent collaboration or long task chains, the quota can be exhausted in days. Beyond the limit, pay-as-you-go pricing kicks in at market rates, often resulting in startling end-of-month bills.

Is This Crawfish Really a Qualified AI Assistant? From its origins as a weekend experiment in November 2025 to today, OpenClaw is less than four months old. It is a rapidly iterating but still粗糙的开源项目, with a significant gap remaining before it can be considered a true "mature product." Known major shortcomings include: simple tasks sometimes being over-complicated; tasks莫名中断 mid-execution; unstable memory功能, sometimes "forgetting" previous preferences; and a significant need for optimization in the efficiency ratio between Token consumption and actual output. Among the thousands of skills on ClawHub, hundreds have been found to contain malicious code.

Binance founder Changpeng Zhao (CZ) posted a brief but potent message on X: "Claimed that after installing the crawfish, you don't need to do anything. Then spent all your time adjusting the crawfish that does nothing." This精准命中s the experience of most average users—the promise of "freeing your hands" often first turns into "occupying your hands."

Security risks are a Sword of Damocles hanging over every user. Research by security agencies indicates over 40,000 OpenClaw instances are exposed publicly, with 63% containing exploitable vulnerabilities. Even companies like Google, Anthropic, and Meta have internally banned OpenClaw, not due to issues with the technology itself, but because current security防护 mechanisms have not kept pace with its capability expansion.

A particularly cautionary案例 comes from Summer Yue, AI Alignment Lead at Meta's Superintelligence Team. She gave the crawfish a seemingly simple instruction: "Check the inbox, suggest which emails can be archived or deleted." The crawfish proceeded to batch delete emails directly, with safety limits completely failing, only stopping after she physically powered down the machine. Elon Musk has also expressed concern about OpenClaw's security: "People are giving root access to their entire lives to OpenClaw."

OpenClaw's founder, Peter Steinberger, himself offered a blunt truth: "If you don't understand the command line, this project is too risky for you." This statement is worth careful consideration for anyone犹豫要不要装龙虾.

Don't Let AI Anxiety Become the Scythe That Cuts You Returning to the fundamental question: Why were nearly a thousand people willing to queue outside the Tencent building? Why would someone pay 500 RMB for an installation? One observer hit the nail on the head: "It's primarily selling an information gap, targeting those fearful of 'doing it themselves' and 'code.'" Put simply, many aren't buying a crawfish; they're buying a sense of certainty against AI panic—they fear missing out on the next big thing, akin to the internet 20 years ago or Bitcoin a decade ago.

This anxiety has its rationale. OpenClaw确实验证了 the exciting possibility that "AI is no longer just a chat window, but a true executor that can work for you." But possibility is not reality, a prototype is not a product, and others' success stories do not equate to your use case.

However, consumption driven by anxiety has always been the favorite prey of "shovel sellers"—from the 500 RMB installation fee to one-click deployment packages, skill packs, API price differentials, corporate AI training courses... an entire industry chain has清晰地 formed around your insecurity.

A crawfish won't automatically become a good employee, just as a good computer won't automatically make you an excellent programmer. AI is an amplifier; it magnifies inherent human capability. Fu Sheng used the crawfish to achieve the output of a team because he already possessed the underlying ability to build a team, not merely because he wanted to "install it first and see."

Amid the noise, maintaining independent judgment and refusing to be swept into irrational decisions by the "fear of being left behind"—this is the core asset that every irreplaceable human individual should prioritize protecting in this technological wave.