Original · Unique Research · 2026-03-21 · Shanghai
English edition note: This complete translation preserves the source's forceful metaphors, Ren Bobing's views and Wu Wei's remarks. Lobster refers to OpenClaw and related Agent systems, not a literal animal. Popularity comparisons, software-history claims, drug-development timelines and the forecasts attributed to METR are reported source assertions, not independently verified findings. The pharmaceutical passage does not establish clinical success or regulatory approval, and future Agent capabilities are predictions rather than demonstrated outcomes. Descriptions of organizational information access do not imply permission to access private information. The virtual-character anecdote describes a reported portrayal and audience response, not evidence of an AI's feelings. No investment, medical or security recommendation is being made by this English edition.
Unique Research · Lobster Night Talks
Find Three Entry Points and Position Yourself Early for Business Between Lobsters
Your Once-Valuable Experience Can Explode into Your Most Dangerous Burden at Any Moment
Just last month, countless people were arguing heatedly over which large model wrote better code. In the blink of an eye, Agent systems represented by OpenClaw made its open-source project's popularity eclipse that of the once-unassailable Linux in just four months. Ren Bobing, an investor who has spent more than a decade in technology venture capital and overseen numerous emerging AI projects, offered a diagnosis: in this operating-system-level migration, your once-valuable experience can explode into your most dangerous burden at any moment.
The Linux of This Era—and Tools That Are No Longer Tame
When most people first install their "little lobster," they use it as little more than a helpful search engine or a translator for long documents.
But in Ren Bobing's eyes, it is much more than a useful tool. It is the operating-system foundation of a new era.
"First there was Unix. Then Linux conquered territory through being free and having an exceptionally strong community ecosystem, eventually squeezing Windows down to the desktop alone." Ren Bobing compared OpenClaw to the Linux kernel of an internet of intelligence.
Why? Because earlier AI was entirely reactive: give it a push, and it moves once.
In this new framework, however, it has its own vital sign: a Heartbeat. It can operate your terminal devices autonomously, quietly execute background programs in your system, and even modify memory and set up long-term memory files itself. "You can even change its heartbeat cadence—from once every few dozen minutes to a few times a day—to maintain persistent system connections."
The shift is dramatic. Tools have finally grown hands and feet and are trying to operate to their own rhythm. Software in the GUI era was designed to reduce the cost of humans clicking screens. Amid this upheaval, we are startled to discover:
A large number of software products are not intended for humans at all.
If you build an Agent assistant for humans, its ceiling remains trapped in the old system: user numbers multiplied by per-customer spending or subscription fees. Products for Agents—underlying protocols, interaction middleware, and automation sandboxes—follow an entirely different logic. They draw money through tolls on extremely frequent API calls and charges for communication environments. This ecosystem is highly transparent and need not cater to the supposed UI experience; it could produce oligopolies with unprecedentedly concentrated monopoly power.
In this transparent world where people and machines intermingle, an even stranger organizational inversion is unfolding.
The Disappearance of the Front End—and Bosses Who No Longer Assign Work
While people worry about the right incantation to make AI draw an attractive image and boost click-through rates, a quiet meat-grinder effect has begun among workers.
"From conversations with several projects, we found severe attrition among engineers, especially front-end and back-end developers," Ren Bobing said bluntly. "Some teams now rely entirely on product managers using AI to hammer out front-end code themselves and complete development."
Hiring costs time and money. As AI's understanding bridges the steep divide between natural language and machine code, building software engines by expanding armies of programmers is becoming a poor investment. During the conversation, Wu Wei joked: "People used to hire more troops once they made money. Now, as soon as they make a little, they pour it all into buying TOKEN."
The whirlwind has even reached biomedicine, regarded as an especially asset-heavy field.
Ren Bobing gave an example that went to the heart of the issue. Previously, even after a biotech company raised a large sum, its money might stretch only to pushing a handful of pipelines forward, enduring five to eight years before reaching a milestone. With highly concurrent Agents today, that block of ice has suddenly broken apart.
"With the same pool of funding, you can now advance dozens of drug-research combinations concurrently."
With AI handling vast amounts of trial-and-error validation in parallel, the once-lengthy cycle of pharmaceutical success is compressed into one or two years. "Even while building such a large network for validating assets, the pharmaceutical team no longer needs to be as sprawling and bloated as before."
Alongside the widespread removal of work, the direction of managerial power is reversing.
In a decentralized team full of Agent collaboration nodes, for example, the hierarchy barriers of traditional enterprise software collapse. Every employee is now chatting with AI that moves across business lines; a determined frontline employee may find it quite possible to extract another department's—or even management's—strategies and spending.
Some especially alert companies have therefore begun a drastic restructuring: the boss stops issuing work instructions downward altogether. The top-level goal goes directly to a single AI enterprise hub. The workflow is then broken down to the extreme: AI instantly divides up the entire month's schedule and separately assigns specific digital workers and human employees to do the work.
Subordinates' reports no longer need to climb the hierarchy layer by layer. AI connects information horizontally, aggregates it, and checks it. The boss sits at the end and receives only the final deliverables.
At that computing node, people are effectively becoming peripheral units of external computing power attached to a vast AI grid.
This demands exceptional adaptability from founders. It is no longer an era of competing on empathy or emotional value. "Agents do not need unnecessary empathy," Ren Bobing said. "What is required now is an extremely strong mental capacity to absorb dense, contrary signals and dynamically adjust course in an instant."
Three Entry Points—and the Treasury Rules for Super-Individuals
Amid a boom where large models and underlying protocols repeatedly squeeze them, where should entrepreneurs position themselves?
After extensive exploration, Ren Bobing laid out a penetrating framework of three entry points for the first time in this exchange:
1. The Data Entry Point
This is more than scraping and repackaging public data. Can you provide Agents with a highly substantive, structured foundation of private-domain data that can evolve over time? It might be human biological-probe data or an exclusive, privately held stream of everyday operations. If you secure that territory at the source, the system has to come to you.
2. The Physical-Carrier Entry Point
AI cannot remain floating in the cloud. It has to land. Smart speakers and headphones with a lobster foundation, intelligent in-car systems, and physical robots at the endpoint: whoever can build a low-cost container that touches the physical world directly controls the user's ultimate experience.
3. The Interaction Entry Point
Something like Xianyu, the secondhand marketplace, between Agents! A hub for credit transfers, token settlement, or matching data transactions among different systems is precisely the kind of interaction center the OpenClaw ecosystem desperately needs.
In the rougher, wilder infrastructure underneath, enormous companies could also emerge in communications—handling public-network ID assignment for globally distributed Agents and resilient connection points—or in deep enterprise collaboration tools.
But if two or three people can hold tightly to an entry point, work with a group of so-called lobsters as super-individuals in an OPC model, and already generate extremely high revenue per person, do such scrappy teams really need investment firms to come in and tell them how to run things?
It is a painfully blunt question. Ren Bobing did not evade it.
If the aim is merely for a few friends to get together and make a quick living, taking VC is optional. But if the ambition is to raise a flag and become a dominant force amid the chaos, then even with the strongest tools, capital remains critically important.
Simply because competition is so intense.
"You used to be able to earn steadily from something for two years. Now that cycle of survival and turnover has compressed to weeks or months."
Fail to secure your territory within two months, and a wildfire of free alternatives or an underlying update may wipe you out.
In this feverishly fast-moving sector, investors bring more than rescue money when death is near. They bring a pass to traffic and visibility, industry-wide momentum, and the broad integration of shared social resources. That is precisely the strongest boost that a grassroots OPC trying to conquer the world behind closed doors with a few bundles of compute cannot obtain.
An Agent Moore's Law—and the Feast of Who Ultimately Eats Whom
METR proposed a concept that has made many Silicon Valley elites break into a cold sweat: an Agent Moore's law. Roughly speaking, the time horizon over which Agents can reason coherently and solve complex tasks without interruption is increasing at a rate of approximately one doubling every 7 months.
Some forecasts even suggest that around 2028, a single Agent could be given a task and complete a white-collar specialist's full day of continuous core work without intervention. Three years later, that could become a full year's work. This is tantamount to pronouncing a practical verdict on the arrival of AGI, artificial general intelligence.
But does that mean we can simply fine-tune our AI armies at leisure and comfortably wait for money to arrive?
As the conversation approached its end, Wu Wei posed its most unsettling question with blade-like precision: "Software once ate the world, and AI is now eating all software. As we use large numbers of Agents to stake out territory at the outer layer, while large models rapidly absorb it all behind us… will this thick lobster shell ultimately be swallowed whole by large models, unable to resist?"
Ren Bobing replied crisply: "Possibly. But new lobster shells will certainly emerge too."
That is the thread of fate hidden in all our feverish calling today. The instructions humans and entrepreneurs eagerly issue, the repeated error correction, and the exploration of use cases all leave data trails that ultimately feed the hungry pools of post-training data deep inside large models.
At a certain stage, once a large model has absorbed these patterns and mastered reasoning through world-scale dynamic environmental simulation, the one effortlessly coordinating across threads and issuing commands to underlying infrastructure throughout the system will no longer be you watching the screen.
"When that day really comes, Agents will no longer be something we operate and use. The large models themselves will be using them."
— Ren Bobing
Selected Q&A
Q: If established experience and practical expertise can be instantly outclassed by AI, does the yardstick for assessing an outstanding host or founder change entirely?
Ren Bobing
You are right. It is now difficult to assess someone fully through a few conversations, but the feedback mechanism we watch most closely is the speed at which outside information prompts them to update themselves and adjust their main roadmap. Whenever we meet, good founders can break out of limitations and present fresh perspectives that change the terms of the problem. Whether or not the eventual solution works, someone with that capacity for rapid evolution far outpaces peers who cling to their supposed credentials and fight a war of attrition.
Q: How do you view AI virtual humans and social companionship today? For example, is the open-source girlfriend that recently became popular in South Korea a good way to make quick money?
Ren Bobing
There is undoubtedly room. Once an Agent with deep memory and strong feedback-driven interaction is operating, humans can unconsciously project intense emotions onto it. In this field over the previous two years, there was even a virtual AI character voted out at the bottom of the rankings that shed tears during its farewell, prompting widespread sympathy and calls to send razor blades to the studio. If we can maximize stickiness and interaction in this lobster ecosystem, it will certainly be a lasting business. But the central challenge is whether the emotional relationship you have built up still counts if the underlying engine model updates or a Skill is copied and displaced. The difficulty is building a moat.
Q: If the OPC model is so efficient and powerful, will well-funded venture capital develop new forms of lending or follow-on investment that overturn conventional practice?
Ren Bobing
If extraordinary growth or the enormous leverage of a future stock-market listing is not the ambition, many tiny, attractive, comfortably profitable individual businesses do not need the squeeze of traditional venture capital. But if the ecosystem produces many alliances of small teams whose profits have a ceiling yet whose current breakthroughs urgently require vast resources and momentum, the industry will not ignore them. It will probably develop context-specific, tiered capital-recovery instruments somewhere between today's heavy VC equity investment and lighter commercial-bank lending.
Q: With the rules changing every month, do you still make long-term strategic plans as someone watching developments on the front line?
Wu Wei
I feel many changes have already left behind the rules we treated as gospel in the previous cycle. The change is not linear but discontinuous. So I have simply cut the grand long-term strategic visions and focused all my energy on a single month. I think only about what we can do with new tools in these three weeks or this month, and implement what can be implemented. What about the year after that? Let it go; leave it for later!
This article was compiled from an in-depth interview with Ren Bobing, executive director at Sinovation Ventures.