Original · Unique Research · 2026-03-15
Editor's note: This is a complete historical livestream account; first-person observations and conclusions belong to the original author. Ray Luan is named in the source; Luan Xiaorui and Qinglang are romanizations. Career history, customer counts, sales figures and staffing changes are source or speaker claims, not independently audited results. The figures of 500 and 300 refer to earlier teams, while the reported current change is from 10 people to 4; this is not a documented reduction from 500 to 4 in one company. The source does not state a currency for “fifty or sixty thousand” in daily sales, or establish that sales are profit. “L5” is the author's driving analogy, not a technical certification or an instruction to remove safeguards. The account explicitly retains gradual delegation, self-checking, human acceptance and correction. References to customer chat histories, undisclosed AI conversations and stock-account research do not establish consent, privacy compliance, authorization to access data or trading authority, and are not recommendations for concealed outreach or investment decisions. Revenue anecdotes do not by themselves prove the author's broader conclusions about AI capability.
Unique Research · Late-Night Lobster Conversations
Why Did He Start Treating OpenClaw as an “Employee” Rather Than a Tool?
“He is trying to reinvent the company itself.”
Imagine someone telling you:
He spent 7 days over the Spring Festival installing a lobster;
A month later, that lobster helped him take a product from 0 to launch, while also reducing what had been a 10-person team to 4 real people + 6 agents;
Your first reaction would probably be skepticism: is AI genuinely doing the work, or is this another story manufactured for attention?
"
But after watching the entire Late-Night Lobster Conversations livestream that day, I kept thinking: if some entrepreneurs have already started managing AI as employees, are we still understanding it through the previous generation's tool-based mindset?
That is the real reason I wanted to write this article. Ray is not talking about a productivity tool. He is trying to reinvent the company itself.
What Interested Me Most Was Not That He Installed the Lobster, but That He Began Giving It Goals, Not Processes
Ray Luan, also known as Luan Xiaorui, has a varied background. He was a front-end engineer and is a serial entrepreneur; he served as a VP at the U.S.-listed Autohome, and his most recent job was building ByteDance's used-car e-commerce business from 0 to 1. Earlier, he built marketing and customer-service tools for small and medium-sized businesses as well as an AI website builder, gaining experience in overseas markets with tens of thousands of registered users and paying users in the high thousands. That background made me pay closer attention when he discussed OpenClaw: what he saw was not an entertaining AI tool but a system that could reorganize production and rewrite the division of labor.
Before the Spring Festival, he bought a secondhand Mac mini for about US$300.
He did not know how to use a Mac, had not touched a command line in more than a decade, and all his employees were on holiday. There was nobody to ask. By repeatedly asking AI, he finally got OpenClaw installed. It took a full 7 days from start to finish.
Many people would read this as a story of an entrepreneur willing to tinker relentlessly. But that was not the point I heard in the livestream. The point was that, after installing it, he did not treat OpenClaw as a more sophisticated chatbox or a plugin that could save him a little time.
He treated it as an employee.
He put it directly: his relationship with OpenClaw is collaborative, not that of a person and a tool. He consults it before every decision. He gives it goals rather than detailed procedures: build a new product portfolio around Cloud Lite; find 1000 paying users in three months; or provide research and strategy recommendations for his U.S. stock account. It then breaks down the OKR, divides the work and executes on its own.
This may sound like a difference in phrasing, but it represents two entirely different philosophies of work. In the tool era, the logic is: I know what to do, and you do part of it for me. In the employee era, it is: I know only the desired result, and you find the path yourself. That was what I most wanted to press him on in the livestream, because most people would not dare work that way.
His Most Extreme Step Is “Fully Autonomous Driving”
During the livestream, I said that, using autonomous driving as an analogy, Ray was effectively attempting L5 full autonomy. He delegates the vast majority of decisions to the lobster, specifying goals rather than processes.
Why is this worth discussing? Because it is counterintuitive. Most people do not dare do it for a simple reason: everyone has seen AI go wrong. It makes mistakes, hallucinates, says it has finished when it has not, and can even damage files. The question I kept pressing him on boiled down to this: why do you dare?
Ray's answer was interesting. His first reason was that there were many things he genuinely did not know how to do. When installing OpenClaw, he did not know Macs or the command line, and nobody could help. He could only ask AI. Through repeated questions, he developed a new habit: when a problem arises, ask AI first, not another person.
The second reason mattered more. He said that if you always assume “I am smarter than AI,” you inherently constrain it. You think you are teaching it, but you are actually pushing it back into the position of an underpowered tool. People with fewer fixed technical ideas may find it easier to put AI to real use. Only with sufficient trust and authorization can it deliver its best performance.
I strongly agree with that statement. Many people struggle to use AI today, and the problem is not the model, the prompt, the workflow or even the token. The problem is that they say they trust AI while constantly keeping it on a tight leash.
But This Is Not Blind Faith: His Approach Is Closer to “Freedom + Acceptance Checks”
This is the easiest point to misread. Many people assume Ray's approach is to let everything run completely free. It is not. I also pressed him on another crucial question: if you give only goals and do not watch the process, what happens when it makes a mistake? How do you verify it?
His answer was practical. He does not grant unlimited authority; he delegates gradually. He describes this as a more free-range approach to raising a lobster: first provide room to act, but do not hand over everything at once; let the agent check its own work before submitting it for his acceptance; if the result has a problem, send it back to be redone.
In other words, he is not abandoning management. He has shifted the act of managing from controlling the process to verifying the outcome. Nor does he manage by intuition alone. He built a dedicated Mission Control dashboard, using logic much like a company's OKR process:
1. Start with an overarching goal;
2. Have the agent break it down into an objective and key results;
3. Divide those among different agents;
4. Finally, check whether each day's deliverables have drifted off course.
He has now created 6 agents: Muddy, the overall coordinator; Hunter, the researcher; a social-media agent running Twitter, LinkedIn and Facebook; Tommy, responsible for GEO/SEO; Peter, responsible for UI and conversion optimization; and Jenny, responsible for growth metrics and task breakdown. These 6 agents collaborate in a daily production pipeline.
From 500 People to 4: I Am Less Interested in Downsizing Than in Why Companies Still Need So Many People
During the livestream, I also pressed him on another uncomfortable topic: “optimization.” Ray had previously managed as many as 500 people and had also built a startup team of 300. After adopting OpenClaw, he reduced his former 10-person team to just 4 people: himself, his partner, an architect and one person combining HR, finance and administration.
I asked: if you cut so many people and the business later grows, will recruiting again not be more trouble? Or are you already thinking that you will no longer hire so many human employees?
His answer was equally direct: having more people and being more efficient are completely different things. He said that, whether in startups or large companies, the greatest pain was never too much work. It was the enormous amount of time consumed by communication, alignment, meetings and management. Very little time remained for producing actual results. To him, the greatest cost of human organizations is coordination, not salaries.
That statement is more worth writing about than layoffs themselves, because it touches something deeper: AI may be challenging not a particular job but the organizational form of the company. A small team that once needed 10 people is starting to compress into 1 leader + a few key humans + a group of agents. This is not simply cutting costs and increasing efficiency. It is breaking apart the units from which an organization is built.
Guest Qinglang's Remote Contribution Was Also Particularly Interesting
He connected OpenClaw to sales in a paid-knowledge business, and it was “working pretty well.” But he encountered a subtle problem: the more explicitly he stated his requirements, the less intelligent the lobster seemed.
Qinglang's practical experience ran in the opposite direction. He said that simply giving the lobster customer chat histories and letting it converse could sometimes be more natural than loading it with a complete sales SOP beforehand. The real difficulty was not getting it to answer, but keeping its performance at a consistent, controllable level of quality.
Qinglang said he did not really understand the business very well himself. Yet he used the lobster to create the content to be delivered and acted as the human forwarding it. In less than a week, he was unexpectedly selling fifty or sixty thousand a day. Customers did not know AI was behind the conversation, but people were nevertheless willing to pay. This passage demonstrates something many people are reluctant to admit: in many businesses, what holds AI back is not its capability but humans' own unease.
The Best Audience Questions Brought OpenClaw Back from Showing Off Technology to Concrete Use Cases
Besides Qinglang, another online audience question struck me as especially worth writing about. Someone working in logistics asked a practical question: in a vertical industry such as logistics, should one use OpenClaw directly to improve efficiency and speed, or build a dedicated vertical-agent system?
The question brought the discussion back from knowing how to raise a lobster to something with genuine commercial value: not everyone needs to build a grand system first. Start with the problem you actually want to solve. Ray's answer went to the heart of it: choosing between OpenClaw and a vertical agent comes after identifying the part of the workflow you want to address.
Ultimately, OpenClaw Is Not a Cheat Code but a Lever Amplifying the Capabilities You Already Have
There is one line from Ray that I felt had to remain. He said OpenClaw is fundamentally an amplifier: leverage, a lever. It can amplify your existing ability to make money from 1 to 10, or even 100. But it is not a silver bullet. It will not instantly turn someone who does not know how to make money into someone who does.
It is neither magic nor a toy. It is closer to a new form of organizational leverage. If you already have a business, goals and a way of making money, it can amplify you to a startling degree.
My Honest Reaction After Watching the Whole Livestream
On the surface, this livestream was about OpenClaw. But by the end, what stayed with me was not lobster-raising tips but a larger shift: AI is separating people who use tools from people who lead AI teams.
The most valuable people in the future may not be those who are best at doing the work, but those who are best at defining goals, breaking down tasks, checking results and managing AI teams. That may be what really makes these Late-Night Lobster Conversations keep people awake.
Selected Q&A
Q1: Why does Ray dare to use OpenClaw like a fully autonomous vehicle?
A: He found that the issue often is not that AI lacks capability, but that humans are too eager to control the route. Ray gives goals rather than processes while retaining mechanisms for verification and correction. It is not blind faith but gradual delegation.
Q2: If he only gives goals, is he not afraid it will make mistakes?
A: He is. That is why he first has the agent check itself, then submit the result for his acceptance. If there is a problem, it goes back to be redone. He also uses Mission Control to break overarching goals into OKR and progressively verify each day's deliverables.
Q3: Why can AI become harder to use for some people who are better at writing instructions?
A: Qinglang's practical experience is typical: in paid-knowledge sales, he found that the more detailed his requirements became, the more “stupid” the lobster seemed. Giving it appropriate freedom produced more natural results. Ray's explanation was that more knowledgeable people are more prone to distrust AI; the more rigidly they prescribe the route, the less likely they are to get its best results.
Q4: Can OpenClaw really generate revenue directly?
A: At least in this livestream, the answer was that people had already done it. Ray described agents covering many stages, from product development to customer acquisition, GEO/SEO and social-media operations. Two weeks after launch, the product had more than 100 registered users and fewer than 10 paying customers. Qinglang also said that, after connecting the lobster to sales, he was selling fifty or sixty thousand a day in less than a week.
Q5: Should traditional industries use OpenClaw directly or build their own vertical agents?
A: The logistics question got to the heart of the issue. Before choosing a large system or a small tool, clarify the specific problem: acquisition, quotations, outreach emails, customer service or delivery. Once the scenario is clear, decide whether to connect OpenClaw directly or build a more specialized system.
Q6: Why does Ray naturally understand OpenClaw from an organizational perspective?
A: His background is not purely technical. He has worked in front-end engineering, founded several businesses, led large teams and experienced the entire journey from building businesses to managing organizations at companies such as Autohome and ByteDance. He therefore looks beyond OpenClaw's features to whether it can change coordination, the division of labor and organizational efficiency.