After OpenClaw's Viral Success: A New Form of Company Is Emerging | Yunqi Capital Attent!on Podcast
Interview with Happycapy Founder
Over the past few weeks, discussion around OpenClaw has shifted from "technological excitement" toward "social anxiety."
Initially, people mostly talked about Agents, skills, and one-person companies. Recently, though, conversations about account security, misoperations, data leaks, and even risks to industrial control systems have clearly heated up.
At the same time, the tech giants are accelerating their entry. This week, Alibaba's CEO personally took charge of reorganizing its AI architecture, attempting to integrate models, Agents, and enterprise workflows. When cloud vendors start building "digital employee systems," this "lobster craze" is no longer just a battle over approaches.
This shift isn't surprising. Nearly every technological leap follows a similar arc: optimistic imagination first, then risk awareness, then systematic implementation. And it's precisely at this stage that we need to return to more fundamental questions: What exactly are Agent systems like OpenClaw changing?
In the latest episode of Yunqi Capital's brand podcast, Attent!on, we spoke with Xu Ming, founder of Happycapy — the earliest "lobster-like product" to go viral.
As a heavy user of Claude Code, Xu Ming began advocating for vibe coding within his team back in 2025. Even non-developers started using AI to participate in code and tool building, significantly boosting work efficiency. Inspired by this, he and his team developed Happycapy, a low-threshold, out-of-the-box "Claude Code" product that quickly topped Product Hunt after launch.
This episode synthesizes multiple perspectives from AI investors, entrepreneurs, and organizational evolution participants. Following the "lobster" phenomenon, we discussed the Agent ecosystem, entrepreneurial opportunities, and the many possibilities of future organizations. We've distilled the highlights into the notes below to share with you.
Conversation Guests
Xu Ming (Jarod) — Founder, Happycapy
He Zeqing (Cobo) — Frontier Tech Investor, Yunqi Capital
Host
Li Na (Linda) — Managing Director, Yunqi Capital
01 Will Companies Evolve into Precisely Billable "Asset Units"?
When a technology wave first appears, people typically start by discussing efficiency.
OpenClaw is no different. The conclusion that "Agents can replace some repetitive labor" is obvious. In the short term, the most immediate change for many enterprises will likely be improved human efficiency and released profit. But if we extend the timeline, this isn't merely an efficiency issue.
When one person can mobilize a swarm of Agents to work, and large volumes of execution are modularized and automated, will the company as the smallest unit become increasingly lightweight? It may no longer need to be a constantly expanding team, but rather a highly compressed work unit.
Following this logic further leads to an even more interesting question: as a company's cost structure becomes increasingly transparent — with token consumption, compute costs, and automation execution costs all precisely calculable — will evaluating a company for investors increasingly resemble judging an "asset unit"? Whoever consumes fewer tokens while generating higher revenue has the better business model.
This may sound like a distant thought experiment, but OpenClaw makes such questions feel less remote. What it truly pries open isn't just a tools赛道, but even the very concept of the "company."
Discussion Excerpts
Linda: "Could this mean that in future companies, because there are only a few people and the accounting is so clear, with profits and revenue both worry-free, they might not need an IPO at all? Or even subsequent funding rounds?"
Cobo: "Future company forms might be one or two people, calling upon dozens or hundreds of Agents below them."
Jarod: "I actually started by transforming myself into this way of working. Later I discovered that work really became easier, yet what I could accomplish multiplied. My management radius, my decision-making radius, were both enormously expanded."
02 Are Agents Starting to Grow Like an Ecosystem?
Compared to being an "Agent tool," OpenClaw's greater value may lie in letting many people see, for the first time, the ecosystem structure behind Agents.
The most obvious layer is the rapid emergence of various "lobster-like" projects: some copying, some modding, some making lighter versions, others trying to shrink Agents into different devices and environments. The technical community has begun continuously forking around a single framework.
But two other layers of change are more noteworthy. One is Agent-to-Agent collaboration. People have begun experimenting with collaboration networks, transaction systems, and even social relationships between Agents. Agents are no longer just tools, but more like participants within a system.
The other layer is Agent-human collaboration. In real work, many tasks inherently require humans and systems to complete together. Humans handle goals, judgment, and acceptance; Agents handle execution and exploration. Thus, product forms around this kind of "mixed workflow" are also gradually becoming clear.
In other words, OpenClaw has let more people see: Agents aren't single-point capabilities, but a system that is forming structure.
Discussion Excerpts
Jarod: "OpenClaw's origin actually comes from Claude Code — Claude Code let Agents run directly in your computer, essentially giving them a pair of 'hands.'
OpenClaw made two key modifications: first,打通通信层, letting you communicate directly with Agents through chat tools like Lark; second, turning the original tools into 7×24 online services, so Agents reside permanently in the computer.
Of course it's still very early, the code structure needs refactoring, and new alternatives may emerge in the future, but the overall evolution will continue in this direction."
Cobo: "Lobster actually gave Agents the concept of being ecosystem-ized for the first time. We're seeing many Agent-to-Agent directions, like Agent social networks, Agent exchanges, Agent collaboration systems, and so on."
03 Skills Matter, But Are Hard to Be the Endgame?
In this wave of discussion, one concept easily overestimated is skill.
The imagination around skills is actually quite natural: will everyone make skills in the future? Will a massive skill marketplace emerge? But looking down from product logic, skills are more like underlying components than end products.
The reason is simple: ordinary users rarely care how many skills you have connected behind the scenes. They only care whether the task gets done.
Configuring skills is fun for developers, but for most people, it's more like a barrier. So the form more likely to be widely accepted in the future is, conversely, the agent packaged further up. Not a pile of scattered capabilities, but "digital employees" that already come with memory, settings, and role assignments.
There's a vivid formulation from the show: "What Agents target isn't just a SaaS, but a SaaS plus an employee."
In the past, enterprises bought SaaS but still needed someone to operate it; in the future, what many people actually want to buy is an execution unit that can directly deliver results.
From this perspective, skills are more like building blocks than battlegrounds.
Discussion Excerpts
Linda: "Could there be a possibility that current SaaS software, as this productivity enters serious enterprise scenarios, slowly transforms into a kind of skill?"
Cobo: "I don't think skills will be the battleground. Skills are essentially patches for delivering results — they're process, not purpose. For end users, what truly matters is that final result."
Jarod: "Skills are an important component, but they're more like a core building block. What's more likely to become the standard product in the future is actually agents that package different skills, with basic memory and settings."
04 After the Giants Enter, Do Opportunities Lie at the Infrastructure Layer?
Almost every new wave encounters the same question: when the giants enter, do startups still have opportunity?
This is already beginning to happen. Startups are exploring product forms, cloud vendors are undertaking deployment capabilities, and giants are beginning to compete for system-level entry points. Competition looks set to intensify quickly.
But another view is: this market may only be just beginning. Many people see the direction, not the landscape. To use an imprecise but vivid formulation, the current completion level may only be 10%.
This means startup opportunities may not lie in the most bustling places. Much of what truly accumulates value is instead hidden at deeper layers:
Communication layers, scheduling layers, sandboxes, security mechanisms, long-running operations, Agent collaboration systems... These things look insufficiently fancy, yet determine whether Agents can evolve from demos into systems.
Simply put: don't just patch models. The problems that truly persist long-term are often at the systems layer above the model.
Discussion Excerpts
Linda: "Every technology wave encounters one question: when the giants enter, do startups still have opportunity?"
Cobo: "After LLMs emerged, product form changes driven by technical paradigms generally operate on a six-month cycle. Whether giants or startups, everyone has a relatively short time window. If you don't generate market buzz in the first two or three months, it becomes increasingly difficult for startup teams."
Jarod: "There actually isn't real competitive landscape yet. Everyone has just seen a direction and is moving toward it."
05 Ultimately What's Changing Isn't Just Software, But Organization
If we pull the perspective even wider, what truly gets pried open may not be products, but organization itself.
When Agents enter workflows, many people's work structures will change. One clear trend: work focus will gradually shift from execution toward both ends.
One end is pre-event thinking and decision-making, the other is post-event acceptance and feedback. That middle stretch of repetitive execution may increasingly be handed to agents. This sounds like an abstract judgment, but it's actually quite concrete.
Much information can first be organized by Agents, then passed to humans for judgment; many processes will be compressed; the rationale for many hierarchical levels will also be re-examined. Organizations may become flatter. Decision radii may expand. The frequency of human-Agent dialogue may even exceed that between human colleagues.
Of course, this process won't be smooth. In the short term, some inefficient, repetitive work will be displaced, bringing chaos and anxiety. But if we extend the timeline, perhaps what's more worth considering is this: What Agents displace may not be human value, but those processes that sit between values yet consume enormous time.
Discussion Excerpts
Cobo: "The future world will certainly amplify the capability boundaries of the decision-making layer. Bold prediction: within perhaps 18 months, many of our reporting and collaboration counterparts will become AI."
Jarod: "Attention in work will migrate from execution toward both ends — more time spent on pre-event thinking and discussion, and post-event acceptance."
Jarod: "There will be some short-term chaos, but in the long term, people will increasingly do things that are truly valuable and meaningful."
More in the Podcast
If these thought experiments about Agents, Skills, and organizational structure have sparked new questions, welcome to scan the QR code below and access the full discussion on Attent!on.
Timeline
Part 1 Why Has This "Little Lobster" Stayed Hot for So Long?
04:03 Cognition Gap: OpenClaw Isn't Just a Tool, But a Leap in Agent Architecture
05:50 Key Inflection Point: From "Mechanical Arms" to "Dexterous Hands" — Agents Finally Have "Hands"
09:56 Why It's Hotter Domestically: The Triple Drivers of Subsidies, Claude Alternatives, and "Raising" Appeal
Part 2 From "Raising Lobsters" to Product: How Do Agents Enter Real Workflows?
13:27 Happy Capy Founder: How I Let Agents Work for Me While I Sleep
15:50 Biggest Challenge: Capability Boundaries Expanding, But Execution Paths Extremely Unstable
20:02 Turning Point: Hadn't Coded in Five Years, Leveraged Vibe Coding for 10x Efficiency Gains
24:07 Happycapy: Turning "Cloud Sandbox" Into an Out-of-the-Box Agent Operating System
Part 3 Thought Experiment: When "Companies" Are Compressed, What Do Investing and Entrepreneurship Become?
35:30 When Long-tail Demands Are Decomposed by Agents, Does SaaS Still Hold?
39:36 From "Human Scale" to "Asset Unit": Companies Are Becoming Calculable Structures
41:29 Changing Due Diligence Logic: Will Future VCs Look at Token Consumption and Output?
43:37 A More Extreme Question: If a One-Person Company Runs a Proven Model, Is an IPO Still Needed?
Part 4 Prediction: Will Human Work Be Rewritten in 18 Months?
01:01:39 Entrepreneurial Moat: Don't Patch Models, Look at the Harder "Systems Layer"
01:12:09 Management Boundaries Broken: When Your Reporting Line Becomes AI
01:14:50 Work Migrates to Both Ends: The Middle Execution Layer Is Being Eaten by AI
01:18:18 A Counterintuitive Conclusion: Programmers Might Be Able to Code Until 60
🎧 How to Listen: Open "Xiaoyuzhou" and search for Attent!on, or click "Read Original" at the end of this article to find this episode.
We also have a small gift for dedicated listeners:
Leave a comment on the Xiaoyuzhou episode page sharing your Happycapy experience, or your thoughts on Agents / OpenClaw / one-person companies. We'll select 5 listeners to receive a 1-month Happycapy Max experience. Experience at: happycapy.ai**