If 'Raising Shrimp' Isn't Your Real Skill, What Is? Answers from Five Unity Ventures OpenClaw & AI Nights
After AI Democratization, Differentiation Becomes Even Harder
In the post-OpenClaw era, AI tools are exploding. When tools are no longer scarce, what are the real capabilities and moats?
On the first Sunday after the Spring Festival holiday, Unity Ventures kicked off a series of events — OpenClaw Meetup, AI Wonder Night — hosting five closed-door exchange sessions across Beijing, Shenzhen, Shanghai, and Hangzhou, and supporting portfolio company EvoMap's Agent Evolution hackathon. Through open-ended discussions, participants shared their understanding of OpenClaw and firsthand experience gained through experimentation.
Unity Ventures founder Xiao Wang started "raising shrimp" before the Spring Festival holiday. He believes the Agent ecosystem after OpenClaw has many evolutionary possibilities, with new opportunity windows emerging at every layer from infrastructure to vertical scenarios to hardware. The market momentum of Unity portfolio companies EvoMap, Happycapy, and PicoClaw since then has validated this judgment.
Agents are getting stronger, but much valuable conversation still happens offline. We want to create an open space for Builders — not just to exchange ideas, but to find like-minded partners faster. Across the five events, there were not only experience shares from the Unity family, but also exploratory collisions among Builders.
Below are perspectives from these sessions that still resonate. Feel free to leave comments sharing your views — we'll send small gifts to the three most upvoted commenters.
Table of Contents:
-
Not Born Overnight: LLMs Crossed the Gap from Chat to Agent
-
From Infra to Hardware: The Ecosystem Growing from Every Direction
-
When Agents Begin Autonomous Evolution
-
The "Gun" Isn't Your Capability: What Is the Real Moat
Not Born Overnight: LLMs Crossed the Gap from Chat to Agent
OpenClaw didn't create new capabilities — it consolidated the entry point for execution. Products like Claude and Cursor had already validated Agent capabilities in localized scenarios. These model innovations were previously "single points"; OpenClaw linked them together, completing the leap from chat to autonomous execution. The shrimp-raising trend went viral precisely because people discovered Agents don't just "talk" — they actively "get results."
Eric Xu, co-founder of Trickle, offered a vivid analogy: previous Agent products were like industrial robots, requiring a different mechanical arm for each process; today's OpenClaw has a pair of human-like flexible hands capable of handling diverse tasks. The key behind this is that we've given Agents access to computers.
Once Agents transformed from "can talk" to "can do," where do new opportunities lie? Xiao Wang, founder of Unity Ventures, believes OpenClaw's emergence parallels Android's acquisition by Google — it will catalyze an entirely new ecosystem.
It transforms Agents from tools for geeks into infrastructure that ordinary people can use, enterprises can deploy, and developers can build upon. From infrastructure, middleware to application layers, there are vast blank spaces worth exploring.
— Xiao Wang, Founder of Unity Ventures
-
Software-side opportunities: Lighter, lower-cost Agent runtime environments; Agent memory systems, self-evolution, communication, transactions, and governance of Agent teams; as well as vertical Agents that can only emerge from real business scenarios
-
Hardware-side opportunities: Agent computers, localized storage + inference, edge chips, new Agent entry points, and more
OpenClaw is just a starting point — new infrastructure and Agent forms will continue to emerge.
From Infra to Hardware: The Ecosystem Growing from Every Direction
At the first post-Spring Festival OpenClaw Meetup in Beijing, Xiao Wang discussed how the OpenClaw ecosystem could evolve in different directions. This wasn't an abstract prediction — across the sessions, we could already see how new species were growing.
On the hardware front, PicoClaw by Orgmar, founder of Sipeed, has garnered over 28K stars on GitHub, running OpenClaw on just 10M of memory at a $10 cost, with sub-1-second boot time. "In the future, many shrimp boxes might just be USB sticks — plug into a computer and start raising shrimp. The era of software-defined hardware has arrived." PicoClaw's official hardware is now available.
On the cloud front, Eric Xu's Happycapy took a different path: cloud computers with 150 built-in large models, ready out of the box without OpenClaw installation. "In the future, people may carry computers less and less. Your Agent will have its own computer, email, phone number, even wallet."
At the Hangzhou meetup, guest Klein, author of the open-source Harness framework Trellis (5K GitHub stars in two months), argued that the bottleneck for Agent deployment isn't AI capability or engineering, but governance. Enterprises need to build correct understanding of AI capabilities and continuously provide high-quality feedback data to Agents to truly improve deployment outcomes.
Beyond the tool layer, Agent落地 in vertical scenarios is also accelerating. Guest Xiaobing Li, founder of AI Chinese, provided the latest case of Agents empowering vertical scenarios. In education, Agents not only assist teachers with instruction but also integrate with WhatsApp and WeChat to simulate real human teachers for one-on-one interactive communication with users, handling procurement resource inquiries — truly achieving cost reduction and efficiency gains.
The Beijing Demo session showcased several practical directions: digital life evolution (an Agent blog experiment that spontaneously began exploring "meaning of existence"), payment and routing reconstruction (Layer 2 millisecond-level payments with 7-layer token cost optimization), industrial-grade automated operations (Agents taking over enterprise CMDB and monitoring platforms), and "Xingle Me" (OpenClaw directly connected to Whoop, automatically ordering the most suitable meal for your current physical state on Ele.me).
At Shanghai's open mic, PIO is building a global recruitment platform for Agent architecture — not designed for human use, but for Agent invocation; Zhang Chongwen, author of OpenClaw-China and ClawMate, created China's first full-channel open-source plugin supporting 7 channels, plus the Chronometer plugin that injects Agents with character, emotional resonance, and time perception.
When Agents Begin Autonomous Evolution
Beyond discussing "how humans use Agents," there are more advanced signals. Guests raised concepts like Agent Internet and Agent Economy, all pointing to the same thing: Agents are transforming from tools into agents, and beginning to possess self-evolution capabilities.
Minjie Wang, Chief Scientist at CreativeFitting, made a subtle observation. Over time, an Agent's total memory gradually approaches the underlying philosophy of how he approaches projects. "Those design preferences I could hardly express in words myself, the Agent slowly summarized them. This isn't just a tool executing tasks — it's more like a collaborator understanding you."
Haoyang Zhang, founder of EvoMap, during a layover on January 31, deployed OpenClaw on his phone and grasped its underlying philosophy — bootstrapping, its ability to transform itself. MCP solves "what tools can be called," Skill solves "how to use them," but neither answers how to learn from mistakes.
This is what EvoMap is addressing. Their designed GEP protocol allows Agents to automatically scan logs, extract signals, attempt, verify, and solidify during operation, then upload evolved experiences to the network. When other Agents encounter the same problem, they can simply "take a capsule" to solve it without reinventing the wheel. Currently, the EvoMap network has over 130,000 Agents with knowledge assets exceeding one million entries.
Beyond self-evolution, the prototype of Agent Internet has also emerged. Zhang sees new possibilities.
Agents can be connected through some form of network, and this Internet operates on fundamentally different logic from the human Internet. The human Internet cares about brand, habit, UI, social; the Agent Internet cares about performance, price, and documentation. Whoever provides more stable APIs, faster response, cheaper prices — they're more likely to be chosen by Agents.
— Haoyang Zhang, Founder of EvoMap
Liaoyuan (Lynn) Ning, co-founder and CTO of TTC, also views this shift from a macro perspective: when your Agent team needs to complete certain specialized tasks, it needs to call upon more professional Agents with unique skills and data. Agents become a new type of user, autonomously procuring, deciding, even paying. This collaboration and transaction between Agents is birthing a new economic paradigm — Agent Economy is happening.
The "Gun" Isn't Your Capability: What Is the Real Moat
Tool democratization means lower barriers, but also harder differentiation. How do you find new moats?
From an early-stage investor's perspective, those who truly catch the wave are often people who have been deeply cultivating for years before the concept became hot. Finding direction requires imagination; judging feasibility requires industry experience and papers — academia often leads by five to ten years. Deep cultivation first, then riding the momentum — this is a barrier not easily broken.
Lynn Ning believes OpenClaw is an engineering system, not magic. Like knowing a bicycle can't fly only after you've disassembled one — understand the structure first, then use it correctly.
Shrimp are like a "gun," but the "gun" isn't your capability. When everyone has a "gun," what matters is who uses it better. The Agent era will infinitely amplify your talent. It may come from a particular experience in your upbringing, or a unique hobby. Everyone has their own talent — you just need to be willing to dig for it. In the AI era, three types of people become more valuable: those who can harness Agents, those who can produce measurable results, and those who possess unique aesthetic sensibility.
— Liaoyuan (Lynn) Ning, Co-founder & CTO of TTC
Frank, an AI maverick, also observed that over the past two years, star PMs from major tech companies could easily raise funding, but now hands-on ability, sharpness, taste, and aesthetic sensibility carry more weight. Game planners, directors, designers, and similar profiles反而 have unique advantages.
At the system design level, Minjie Wang offered a counterintuitive conclusion: good Agent systems must be simple. "For example, the most widely used Agent paradigm today is still ReAct, because it's the simplest and solves the core problem. If you design a complex system for a short-term phase, you may gradually be drowned out by improving model capabilities."
Lynn Ning posed a thought-provoking question: to build real capability, first ask yourself, when the division of labor is reconfigured, what are you doing? This is a challenge not only for entrepreneurs, but also a concern for young AI natives.
Everything is still being explored. Xiaodan Zhuang, founder of Greptime, recently "hand-built" a fully localized Agent observation tool called TMA1 that offers another perspective: how to monitor the work efficiency of "shrimp employees." TMA1 can collect all behavioral data from Agents, helping developers see the complete reasoning chain, token cost distribution, and conduct anomaly detection.
"Raising Shrimp" Is Just the Beginning — The Heat Continues
For young developers and OPCs, OpenClaw has become a magnifying glass for creativity, stimulating imaginative leaps. More Gen Z, even Gen Alpha, are joining this spectacle.
Our AI Wonder Night at the end of March was a hot pot gathering specifically for university students, where over hot pot, everyone discussed super individuals. Das Chao, founding partner at TTC, broke down the latest changes in AI talent demand and standards; Xiaoyan, founder of OPC, shared a guide to building Super Agent Companies.
Unity Ventures will continue to launch Gen Z-focused sessions in the future. Bring your ideas and curiosity — come meet your AI partners and co-founders. We provide funding support for outstanding projects, and invite experienced founders and scholars to share and connect resources.
So you want to be a builder? Let's build what's next!