OpenClaw reached the top of GitHub in just four months, surpassing Linux and React, and became the fastest-growing open-source project in history. However, most users discovered after installation: the API costs are burning, and the lobster is idle.
Who is actually making money? Can on-chain transactions be handed over to the Agent? What if they're attacked? What are the differences between domestic and overseas approaches? Will it be a minority phone or WeChat in a year? In this episode, we invite five shrimp farmers to find answers to these questions together.
The following is the timeline directory of this episode's content; friends in need can jump directly:
00:04:42 - Shrimp farming experience sharing (self-introduction, shrimp usage experience, pitfalls)
00:28:46 - Money-making issue (Can OpenClaw help users make money in the crypto market, AI + CRYPTO new scenarios)
00:53:58 - Security issues (permission boundaries, which operations can be handed over to the Agent)
01:02:31 - AI agents for on-chain transactions (safety, differences from quantitative robots)
01:13:38 - Domestic vs. overseas ecosystems (Xianyu installation services, Tencent/government subsidies, opportunities for Chinese players)
How does it feel to run for the first time?
The first experiences of the four guests almost all went through the process of "higher expectations lead to harsher falls."
0xTodd: I hit two big pitfalls in just two days
It was deployed two days after release, and I fell into two big pits—
The first pit: The lobster committed suicide. I allowed it to configure the API itself, ended up deleting all core files like my soul.md, and had no backup. After tweeting, I found that many users had the same experience.
The second pit: Costs exploded. Charged 50 dollars for Claude API, burned it all in one night, with each conversation costing around 1 dollar. Later switched to domestic models (MiniMax/Kimi), and the price dropped by 90%, achieving excellent cost-effectiveness.
DeFi Teddy: A typical failure of expectation management
Got started at the end of January. Originally expected it to control MetaMask for auto-signing, but the browser's operational capabilities were far below expectations, and two core scenarios couldn't be executed. Later adjusted expectations, and found a truly usable direction: digital employees assist in coding, deploying GitHub, and releasing products; digital companions maintain AI boy/girlfriends locally on Mac Mini, with consistent faces and scenes that switch.
The biggest cognitive shift: no longer treating it as a tool but as "another kind of sentient lifeform."
Lisa: Security instincts raised alerts immediately
The first time it ran was indeed shocking—AI finally moved from the chat box to real control of the computer.
But security instincts immediately raised alarms: the more powerful the lobster, the greater the permissions needed; the greater the permissions, the larger the attack surface. The core advice: you can play boldly, but must use isolated devices, strictly separating personal computers, work computers, and "shrimp machines."
Danny: From uninstalling to getting back into it
I uninstalled it after two hours of playing the first time. After getting back into it, I discovered a rule: use it in a lower dimension—let the AI capable of calculus do simple addition, subtraction, multiplication, and division; it will be much more useful. Once you let it do investment research analysis, the illusions immediately appear.
The worst pitfall: I asked the lobster to generate a wallet and manage the private key, and ended up with the private key overwritten, and the money was gone. The hash it returned simply did not exist.
Can you make money in the crypto market with the lobster?
The answers from the four guests were highly consistent: making money directly from the lobster is nearly impossible.
Todd stated it most directly—The brain of the lobster is essentially still Claude/GPT, and its intelligence has not changed. Last year’s AI crypto trading competition, where each GPT/Claude/Gemini participant used 10,000 U to trade crypto, ultimately resulted in losses for all, with DeepSeek barely left with a few thousand dollars, and Doubao "winning" because they didn’t open an account. Putting the same brain into the lobster would yield no different results.
A more fundamental logic: large language models are essentially "commentators," not "players." Similar to AlphaGo and today's large models—AlphaGo is designed specifically for chess and can crush Ke Jie; however, letting Claude play against AlphaGo would result in the same dismal defeat. The algorithms used by top quantitative firms are the AlphaGo of the crypto industry; large language models are suitable for explaining how good these algorithms are, rather than replacing them in quantitative tasks.
So what can the lobster do?
✅ Organize news, follow hot topics, gather information
✅ Assist coding, deployment, and automate transactional tasks
✅ On-chain data analysis, risk address identification
✅ Smart contract vulnerability detection (improves efficiency, does not replace human work)
❌ Trade decisions
❌ Manage private keys
❌ Quantitative arbitrage
Danny's summary was the most practical: while it can help you reduce costs and increase efficiency, helping you go open source is nearly impossible.
How serious are security issues?
Slow Fog’s Lisa provided the most systematic analysis:
Why is there doubt about OpenClaw's stability?
The iteration speed is too fast, with a new version every day or two, and each update can have dozens or even hundreds of fixes, completely overturning traditional software engineering rhythms. At this speed, it is impossible to complete full testing across devices and scenarios.
Main risk points:
- Skills poisoning: Malicious plugins can steal account passwords, API keys, and Tokens, leading to fund theft
- Supply chain attacks: The lobster automatically updates Skills, and new versions do not guarantee security
- Permission abuse: Users with encrypted assets on their computers face the risk of funds permissions being abused
Danny added painful lessons: absolutely do not let the lobster generate wallets and manage private keys, as the returned private keys may be fabricated. Skills updates must be manually reviewed; do not let it install automatically.
Teddy's reminder: When using third-party relays, data passing through their servers poses risks of leaking sensitive information like API Keys. Someone put in a Google API Key and ended up racking up hundreds of thousands of dollars in charges.
Reference for the principle of minimum privilege
✅ Can be handed over to the Agent: Writing code, organizing documents, pulling data, information gathering
❌ Must be confirmed manually: Involving funds, private keys, core server permissions
When connecting wallets, it is recommended to use Skills from Coinbase Wallet, requiring manual secondary confirmation for each transfer on the wallet side for multi-layer isolation.
Major exchanges are giving the lobster “skills trees”; is AI trading reliable?
Binance and OKX have successively launched OpenClaw-related Skills, but the practical attitude is generally cautious.
Danny: Only give the lobster read-only APIs for backtesting, and never let it place orders. It’s fine for up to five orders; beyond that, illusions are bound to appear.
Todd: The essential difference between AI trading agents and quantitative robots is—quantitative algorithms are specifically trained "AlphaGo," while large language models are just "commentators." Letting the lobster handle quantitative trading is like letting a commentator play in a professional match; it can't win.
Teddy: The lobster can serve as an interactive entry point, but the underlying execution logic must be an Agent you have trained specifically, not the lobster making decisions directly.
Conclusion: High-frequency quant— the lobster's response speed is insufficient; trading decisions— the lobster's intelligence is inadequate.
Domestic Lobster Ecology vs. Overseas: Who Has More Imagination Space?
Danny’s judgment is the sharpest: OpenClaw is essentially "a button spirit with a brain," extremely unfriendly to ordinary people, like Linux rather than Windows. The ones who can truly use it well are one in a thousand.
His forecast: Two months later, the popularity of OpenClaw will wane; the truly accessible product will be those developed by big companies like Tencent and Byte, which will create "Windows-level" products. The Personal Computer form released by Perplexity may be the real mass entry point.
Todd’s observation: The domestic market is hotter than overseas because the government is rapidly intervening to promote (Shenzhen and Wuxi have taken the lead in subsidies), and domestic models are priced extremely low, making the "gambling cost" far lower than that for overseas users. In overseas markets, running a task with Claude could cost several dollars; domestically, using Kimi/MiniMax could cost only a few cents, resulting in a completely different experience.
Where are the opportunities for domestic players?
- Sell courses/install: A few days of installation services on Xianyu brought in 260,000, but this is an information gap profit and not sustainable
- Model stocks: After MiniMax went public in Hong Kong, its stock rose from 200 HKD to 1,000 HKD; such opportunities are worth paying attention to (not investment advice)
- Crypto payment infrastructure: AI agents naturally require cross-border, no KYC, and micro-payment supporting settlement methods, USDC micro-payments and native crypto payments are worth continuous attention
- Infrastructure for one-person companies: The lobster makes it feasible for "one-person companies to have digital employees" for the first time, creating potential tools and services around this scenario
Finally, A Few Tips for All Shrimp Farmers
- Manage expectations well: The course sellers may hype the lobster to 150 points, but it may actually only score 65 points
- Use in a lower dimension: Letting AI capable of calculus do simple operations yields the best results
- Device isolation: Shrimp machines ≠ work machines ≠ personal machines
- Fund independence: Any operations involving private keys and funds must have manual double confirmation
- Don't blindly trust Skills: Review before installation, pay attention to updates, and be aware of supply chain poisoning
- It is an intern, not a fund manager: Using it to organize information and assist in decision-making is fine, but letting it manage money independently is not acceptable
Note: This article is compiled from the transcript of PANews Space "Shrimp Farmers Alliance: Tencent Steps In, Government Subsidies, Xianyu Installation Services—How Does the Crypto Market Deal with "Shrimp" Anxiety?" The guests' opinions do not constitute investment advice.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
