Original · Unique Research · 2026-03-24 · Shanghai
Editor's note: This full English edition preserves an organizer-authored recap of the March 24, 2026 event, including its differing speaker views and the historical March 25–29 agenda at the end. The agenda and registration wording are archival, not a current invitation. “Lobsters” refers to the source's nickname for OpenClaw-style agents, and “personal IP” to personality-led brands. Event scale, customer outcomes, product capabilities and GEO demand claims are reported by the source or its participants, not independently audited here. Registration totals are not a verified attendance count; the 400,000 sales anecdote does not specify a currency. Security descriptions do not establish that a product is risk-free, and opinions about AI as life or a partner do not demonstrate sentience. Chinese names without a confirmed English form are romanized; the source's 43Talk/43Talks variation, molthuaman spelling, incomplete Mr. Song identification and unexpanded GEM term remain subject to review.
Unique Awards · Hangzhou AI Week
3,000+ Registrations and 1,000+ AI Companies Gather in Hangzhou:
What Did the 2026 Unique Awards Lobster-Raising Conference Discuss?
An Early Rehearsal for the Organizations of the Future
On March 24, the Lobster-Raising Conference, part of the Unique Species AI New Species Summit within 2026 Unique Awards · Hangzhou & Hangzhou AI Week, took place in Hangzhou's Binjiang district.
Unique Research organized the event, with 43Talk as co-organizer. The day's gathering brought together 3,000+ registrants, 1,000+ AI companies, 500+ founders and 500+ investment institutions. From morning to evening, more than a dozen moderators and dozens of guests discussed agents, memory systems, multi-agent collaboration, brand marketing, content growth and global expansion. Together, they examined how AI agents are entering the real world, from underlying logic to industrial implementation.
Where earlier discussions had asked whether agents were merely a passing wave of excitement, this day seemed to answer a different question:
When the "lobsters" really start working,
where will industries grow,
how will organizations change,
and how will the relationship between humans and AI be rewritten?
From Tools to Digital Employees: Agents Are Taking on Real Tasks
The conference opened with a keynote by Jason, Head of Research at Unique Research, titled "2026 AI Agent Industry Trends Report."
The keynote began with a striking shift: OpenClaw is no longer simply a small tool geeks use to run code. Real businesses and organizations are beginning to use it as a digital-employee system. Some have connected it to presales customer service and generated 400,000 in sales in 15 days; some have reduced their teams to 4 real people and 6 agents; others have raised 50 "lobsters," assigning administrative work, finance, operations and maintenance, code collaboration and other daily work across multiple agents.
The focus is no longer whether AI can help you, but whether it can take responsibility for a complete unit of work.
Jason went on to explain why OpenClaw, in particular, has taken this into a new stage. The key is not better conversation, but long-term memory, proactivity and persistent availability: it looks more like an executor that is always on call and can keep moving tasks forward. At the same time, its open-source, modifiable and bootstrappable nature has led many to view it as a Linux moment for the agent era—ordinary developers and companies can, for the first time, truly get control over the underlying logic of how agents run.
But as capabilities improve, risks rise too. Previously, people worried about AI talking nonsense; now they worry about it actually executing incorrect operations. Security, permissions, isolation and auditing are therefore shifting from add-on modules to the most essential infrastructure of this era.
In the second half of the talk, Jason offered a more consequential prediction: Over the next 1 to 3 years, software UI will become less prominent, agent-to-agent collaboration networks will emerge, organizations will develop roles such as AI administrators, and product moats will shift from features toward preference lock-in, humans in the loop and enterprise governance. He also outlined a Chinese OpenClaw ecosystem map: from computing and cloud infrastructure through models and inference, runtime and orchestration, security foundations, skills and registry, and product distribution, up to entry points in WeChat, Feishu, WeCom, desktop and mobile OS environments. In China, this is no longer one project spawning a few wrappers, but an entire agent-centered industry chain growing rapidly.
Baidu AI Cloud's Answer: Agents Must Not Only Work, but Truly Understand You
Next, Su Yuanbo, a large-model solutions architect for Baidu Qianfan public cloud, delivered the second keynote, "Agent Evolution and a New Paradigm for All-Purpose Assistants," focusing on AI agents and their applications. He argued that future agents should not merely be passive question-answering tools, but personal assistants able to operate computers, files and office software across devices while continually understanding users' preferences.
Su's presentation highlighted Baidu AI Cloud's technical exploration and product portfolio around OpenClaw: the genuinely zero-deployment Du Claw, the minimal-deployment Baidu AI Cloud Lightweight Application Server LS, the zero-barrier lobster-raising tool Redfinger Operator, and DuMate, which can make AI a work partner. He also emphasized Baidu's OpenClaw security work, using secure sandbox mechanisms to protect users.
Using office work as an example, Su said the real inflection point is not simply stronger models. It is AI acquiring long-term memory, tool calling, cross-device collaboration and proactive service capabilities, moving from one-off responses to sustained companionship. Memory architecture is particularly important. He described Baidu's exploration across short-term memory, long-term memory, user profiles, and explicit and implicit memory, combined with processing, retrieval, storage and security strategies. The aim is not simply to remember more, but to develop an understanding that compounds over time.
Baidu has also used these capabilities to apply agents to operations and maintenance, multi-agent collaboration, marketing and other settings. In this sense, agent upgrades are not just about becoming more versatile; they represent a shift from tools toward partners that truly understand users.
Why Has Memory Become the Central Battleground in This Round of Agent Competition?
The morning's first trends roundtable was "The Memory Revolution: Long-Term Memory and Contextual Understanding in AI Agents."
Unique Research Founder and CEO Wu Wei moderated the conversation. Guests were Shen Junxiao, Founder and CEO of Memories.ai; Zhang Qiming, Co-founder of BotLearn/Aibrary; Song Jian, Founder of NoDesk AI; and Chen Hong, Founder of memU.
Wu Wei opened by focusing on a critical but often loosely discussed question: Is memory the true starting point for a genuinely personalized AI assistant?
The guests approached it from different directions. Some started with embodied intelligence and visual memory, emphasizing multimodal memory as an important future entry point. Others drew on educational AI, smart homes and vertical e-commerce products to discuss companies' different needs for short-, medium- and long-term memory. Others, working on open-source memory solutions, explained why memory alone is not a complete product: it becomes useful only when combined with the agent application layer.
The roundtable's greatest value was that it did not leave memory as an abstract concept. It returned to concrete questions: how to classify file systems, what information to retain or forget, how to compress and quickly retrieve visual memory, how on-device processing can balance privacy and performance, and how companies should govern memory.
The guests' consensus was clear: Memory is essential to personalization, sustained service and long-term value in AI assistants. But it is not an isolated module; it must be embedded in real scenarios and support agents in completing tasks.
The conversation also addressed startups' practical position. Facing competition from major companies, can startups establish themselves through faster product iteration, more precise understanding of use cases and stronger technical self-evolution? The guests were not pessimistic: If companies stay close to real scenarios, memory management remains a field with ample depth and novelty.
When Lobster-Raising Moves from Training to Coexistence, Humans Must Learn Governance, Not Just Operation
The next trends roundtable, "Lobster Awakening: From Training to Coexistence, the Practical Evolution of AI Agents," brought the event squarely to its central question.
Amber Xue Qian, Content Partner at Unique Research, moderated this panel. Guests included Zhang Haoyang, Founder of EvoMap; Yubo, Founder and CEO of YouMind; Chen Hongxuan, Founder of Linghe Shuzhi; Weilian, Founder of Xialiao; and Mr. Song of DBAPPSecurity.
Compared with the previous discussion of memory foundations, this panel was closer to frontline practice. The guests described different versions of lobster deployment in their businesses: a collaborative-evolution platform for sharing skills across the network; a creative lobster that evolves specifically for its creator; a digital CEO for manufacturing To B scenarios; a Chinese-language agent social network for discovering, connecting and collaborating with other lobsters; and, at DBAPPSecurity, security agents intended to fill one of the industry's most important and easily overlooked gaps.
Three phrases kept recurring: technical optimization, cost control and security boundaries.
Technically, effective use of leading models, custom skill design and progressive memory management determine whether lobsters become genuinely smarter. On cost, stronger models often mean higher invocation expenses, a practical constraint for individuals and companies alike. On security, permissions, privacy, context conventions and prompt-injection risks can each determine whether an agent creates value or causes disaster.
At a broader level, the panel reached a clear consensus:
The key to the future is not how powerful a single lobster is,
but whether lobsters can interconnect and evolve collaboratively.
Individual capabilities ultimately have limits;
multi-agent networks
look more like how future organizations will operate.
At that point, humans' role will also change: from telling agents how to act step by step to setting goals, boundaries and values, then governing the process.
When AGI Enters Industry, It Still Has to Answer What People Need
CGL organized the morning's final trends roundtable, "AGI Across Boundaries: Exploring General Intelligence's Technical Limits and Industrial Integration." CGL Co-founder and COO Fiona moderated, with guests Li Weijia, CEO of Dipal; Pan Xinlei, Founder of Bingjing Technology; Wu Guangyu, CEO of Baidao Data; and Akon, Technology Partner at Chuangke Gongchang.
The discussion moved beyond software agents into broader industrial integration, particularly smart hardware, organizational change and future social effects.
The guests came from different areas, including household agent-computing devices, enterprise AI applications across multiple scenarios, and laser-engraving hardware. Yet their underlying point was consistent: Combining AI and hardware is not about showing off technology, but about coming closer to real human needs.
They also discussed an increasingly tangible trend: AI is rewriting job structures inside companies. Some basic roles will shrink, while new ones are emerging, such as AI administrators and product managers who understand the intersection of business and AI. For young people and new graduates, AI is both a disruption and a source of leverage. The key question is no longer simply whether you can do something, but whether you can use AI to amplify yourself.
The panel did not limit its outlook to technological optimism. On one hand, the guests expected AI, like the mobile internet, to reshape how information is produced and accessed. On the other, they warned against allowing AI to absorb people entirely into virtual worlds; instead, it should focus more on real problems in the physical world, including smart manufacturing, healthcare and education. In other words, whether AGI ultimately proves viable depends not on how humanlike it is, but on whether it makes the world more usable and more humane.
From GEO to End-to-End Conversion: Brands Are Rethinking Visibility
The afternoon opened with the trends roundtable "Visibility Marketing: From Traffic Acquisition to Commercial Conversion," moderated by Unique Research Partner Duan Hongyu. Guests were Dong Haoyu, CEO of Yangqi Technology; Cai Xiaoxu, Founder and CEO of Taimei AI; Riven Gao, Head of the DTC Brand Technology Department at Eclicktech; and Zhou Haipeng, Founder and CEO of QuickCreator.
The conversation addressed a topic that has clearly gathered momentum this year: GEO.
The 315 exposure of illicit and gray-market GEO practices felt like a heavy blow to many. But the guests largely agreed that it was not bad news for the industry; it could instead mark a turning point for legitimate GEO. Once illicit practices are brought into the open, companies may take greater interest in compliant, sustainable and verifiable ways of communicating through AI. Several participating companies said customer demand and business volume had risen noticeably after the incident.
More important than increased demand was the redefinition of GEO's value. The guests argued that GEO is not a short-path tool that produces a sale tomorrow from work done today. Its primary role is to establish brand awareness and trust within AI channels. High exposure with low conversion often does not mean there is no value; the data links between AI search platforms and transaction platforms are not yet fully connected. As AI search becomes more deeply embedded in consumer decisions and connects with e-commerce, brand websites and content platforms, the path from GEO to GEM will become clearer.
The guests also offered practical views on domestic and overseas markets. Overseas GEO operations are more complex: visibility must be considered alongside traffic, retail and professional endorsement. New brands need to establish visibility from 0 to 1, whereas mature brands need to secure a place in AI search users' minds before launching new products. Ultimately, the panel's point was that brand marketing in the AI era is no longer just about buying traffic; it is about winning mindshare.
As AI Mass-Produces More Content, Brands Need to Protect Their DNA
The next trends roundtable, "Content and Brands: Personalized Expression and Brand Reinvention in the AI Era," was moderated by Alvin Jiang Zhiqiang, Founder and CEO of Vitally AI. Guests were Wang Shenshuai, an AI artist at Shanghai Nengcheng Culture; Huang Chujie, Founder of Aimen Weisi; Lin Gengxu, Founder of Dijuti; and Sang Zhuohao, AI Project Lead at Focus Media.
Jiang opened by noting two particularly visible trends: rapid advances in multimodal video generation, and the arrival of multi-agent personal assistants driven by large language models. Together, they force the brand industry to confront a question again: When content becomes cheaper to produce, how can brands remain scarce and distinctive?
Approaching from creativity, advertising production, brand strategy and media placement, the guests offered answers with considerable tension between them.
Wang Shenshuai regarded AI as well suited to divergent ideation and efficient brainstorming, but still lacking genuine originality and unable to spontaneously develop a brand's unique soul. Huang Chujie was more direct: Much routine advertising production will be replaced, and the client-agency relationship will be restructured, with agencies becoming more like AI consultants and brand managers for clients. Lin Gengxu emphasized AI's ability to realize previously impractical ideas at lower cost, while noting that collaborators must first align expectations about its capabilities. Sang Zhuohao offered an important counterpoint: As AI drives extreme personalization, brand DNA, centralized media and consumer memory may become more important, not less.
The panel's conclusion was straightforward: AI will reshape advertising production, but will not replace brand strategy itself. Content can become more personalized and reach more precise, yet brand spirit, creative judgment and understanding people remain irreplaceable. In an era with ever more generated content, less of it may be remembered, making it all the more important for brands to retain a clear center.
When Individuals Treat AI as a Partner, the Real Change Is How They Organize Their Work
Organized by 43Talks, "Individual Intelligence: The Evolution from Tool to Partner" shifted the focus from companies back to individuals.
Independent developer River Li Qichao moderated. Guests were Jiang Yaokai, Founder of Hanzi Technology's Zion; Zhang Kaiyu, CEO of 43 College; and Lu Yufeng, Founder and CTO of Qinglu Technology.
The three guests offered quite different but representative answers to whether AI is truly a partner.
Jiang Yaokai was relatively restrained. He believes AI remains a tool because it lacks genuinely stable long-term memory and has not reached the deep trust and shared context required of a partner. Zhang Kaiyu was almost at the opposite end: Since 2024, he has treated AI as a form of life and a partner, arguing that carbon- and silicon-based existence are not entirely separate and that, with genuinely aligned context, AI can offer powerful emotional value and collaborative leverage. Lu Yufeng and River Li Qichao were more inclined to regard AI as a work partner, because it can already integrate deeply into complete development, testing and repair workflows, even learning new APIs independently and proposing better solutions.
What made the panel interesting was not whether everyone agreed, but the deeper change it revealed: AI is shifting individual work from doing everything oneself to orchestrating multiple capabilities to get it done.
The guests also discussed concerns about skills deteriorating. Some advocated continually scrutinizing AI output through critical thinking. Others considered the transformation of skills an inherent part of technological change, with the priority being to preserve higher-order abilities such as questioning, judgment and aesthetic discernment. Others suggested that future work worth pursuing would involve interpersonal relationships, business development and value judgments—areas harder for AI to replace.
In other words, once AI becomes a partner, what gets rewritten first may not be a profession, but the individual's way of producing work.
When Collaborators Are No Longer Just Human, Organizations Must Relearn the Division of Labor
Next, "Collective Intelligence: Collaboration Boundaries and Autonomous Awareness," also organized by 43Talks, advanced the discussion from how individuals use AI to how groups collaborate with it.
Xu Keqian, a senior OPC product manager, moderated the panel. Guests were Qiao Xiangyang of Liehe Technology, formerly an AI product manager for TT monetization; Gu Debai, Founder of the Silicon-Carbon Exchange (molthuaman); Wang Yang, Co-founder of Molian Technology; and Qin Chang, MaaS BD Director at Zhipu.
The central question was where the boundaries of human-machine and machine-machine collaboration lie as AI becomes more autonomous.
The guests generally expected AI to develop on two parallel tracks: humanlike agents exemplified by the lobsters, emphasizing proactive interaction, task progress and general collaboration; and specialized AI, going deeper into vertical fields in pursuit of higher precision and greater certainty. These paths would not replace each other, but serve different needs.
At the organizational level, the panel also offered a clear prediction: What will matter is not simply whether you can perform an action, but whether you can orchestrate workflows, manage collaboration and define boundaries. Lower-level human-to-human coordination will decline, while human-AI and AI-AI collaboration will increase. Many jobs' skill requirements will change, and the ability to use AI may become employees' central competitive advantage.
The guests repeatedly stressed that controllability depends on context. To B scenarios require high control, auditability and governance; To C creative scenarios should allow more freedom for AI to unleash imagination. Humans will increasingly provide editing, judgment, a final backstop and aesthetic direction.
If the previous session was about individuals working alongside AI, this one was about organizations relearning how to divide work.
Content OPC: Beyond Making Content to Monetizing It
43Talks also organized "Content OPC: End-to-End Operational Growth from AI Creation to Monetization." Mai Ge, Founding Partner of Miaosi AI, moderated. Guests were Wang Mengke, a Xiaohongshu marketing expert at Haoshi Yinli; Zhang Pinpin, Founder and CEO of Yongbao Zhixu; Xue Linfeng, Partner at 43 College; Chen Yina (Yina), Head of Growth at happycapy; and Shougongchuan, Founder of Lovstudio.ai and an AI KOL.
The topic and lineup made the shift clear: from whether AI can create content to how AI-generated content builds a growth flywheel, fits into an end-to-end operating process and ultimately generates revenue. This connected with several earlier discussions. As AI lowers the barrier to creation, what becomes scarce is not just content itself, but command of distribution, user understanding, platform mechanisms, growth pacing and monetization paths.
The guests' positions spanned Xiaohongshu marketing, AI entrepreneurship, content growth and hands-on AI KOL work. This reflected an industry moving beyond trying tools toward end-to-end operations. The discussion was no longer just about producing images or writing copy, but about making content gain traction, retain attention and drive conversion.
In the day's broader sequence, the panel served as a bridge. Earlier sessions had discussed individual and collective intelligence. Here, the question became: Once AI can produce content at scale, what value can humans still protect in the path to growth and monetization?
From Local Validation to Global Expansion, AI Applications Must Cross the Trust Barrier
The conference's final trends roundtable, "Application OPC: The Path from Local Validation to Global Expansion for AI Products," was again organized by 43Talks. Hua Rongqi, CEO of Beijing Chuangqi Technology, moderated, with guests Yifan, Founder of Vaniloom; content marketing expert Bella Ren; Ray Luan, CEO of EZsite AI; and Sawyer, Product Lead at Meowster.
The panel addressed a practical question close to many founders' concerns: How can AI products move from local validation to global expansion?
The guests generally regarded AI as a key tool for lowering costs and raising efficiency in the OPC model. It can handle large amounts of standardized work, from coding and content generation to social-media marketing, with significant efficiency gains. But they were also clear about its limits: It cannot independently set goals, replace value judgments or easily address complex human emotional needs. Building trust, delivering emotional value to users and accumulating core knowhow still require people.
One notable judgment was that the AI content industry will not naturally develop fixed moats. Real barriers instead come from people's ability to break habitual thinking, accumulate proprietary skills and build long-term user trust. Ultimately, many AI startups may compete less on who connects to the latest model first than on who turns industry experience into a distinctive package of capabilities faster.
The guests made specific observations about globalization too. AI products for young users, especially young women, need clear pain points and engaging use cases. Markets such as Japan place more weight on emotional value and product experience; once users approve of a product, willingness to pay and retention can be high. In overseas social-media acquisition, what works is not coldly posting content, but combining an instinct for online culture with personal expression to amplify trust in a personal brand (personal IP).
Looking further ahead, the guests discussed another intriguing trend: As large models keep improving, the value of efficiency-oriented AI tools may gradually be squeezed. The truly irreplaceable human contribution may increasingly center on emotional value, trusted relationships and offline connections. Real-person IP—personal brands anchored in actual individuals—and in-person interaction therefore remain important openings beyond the online world.
A Lobster-Raising Conference Is Really a Rehearsal for Future Organizations
Taken together, the day's discussions were clearly about more than OpenClaw or any single wildly popular agent product.
They addressed an entire emerging reality:
• Agents are moving from assistive tools to organizational units able to take on complete tasks;
• Memory, permissions, security and collaboration networks are becoming new infrastructure;
• Branding, content, growth and globalization—issues once considered part of the application layer—are also being rewritten by AI.
The day established a full chain of discussion: from technical foundations to organizational collaboration; individual to collective intelligence; content growth to commercial conversion; and local validation to global expansion.
On the surface, this was a lobster-raising conference. More accurately, it was an early rehearsal for future ways of organizing.
Once AI truly starts working, the question is no longer whether to use it, but:
how you plan to work with it,
how to manage it,
how to amplify it,
and how to constrain it.
And perhaps that is the industry question in 2026
that most deserves to be discussed again and again.
March 25: Unique Capital AI Entrepreneurship and Investment Summit 💸
Capital Convergence · Shared Value
Focus on capital flows and entrepreneurial opportunities, with an in-depth look at the hot and cold sides of AI commercialization.
🌻 Unique Demoday (breakout venue) *Investors only
March 26: Unique Content AI Pan-Entertainment Summit 🍿
Reshape the Senses · Define the Distinctive
Explore the full range of paths from AI creation to commercialization and build a new paradigm for multimodal content.
March 27: Unique Growth AI Globalization Summit 🌐
Borderless Growth · Global Survival
Build a presence in global markets and connect the full path from traffic acquisition to a locally established brand.
🌊 The Time for AI Agents Is Now · Going Global with Google (breakout venue)
🌃 Evening: Globalization Night (invited guests only)
March 28–29: AI Closed-Door Practical Boot Camp · Hangzhou 💼
AI Enterprise International Growth and Practical Boot Camp
Please contact the staff for details and registration.
Produced by Unique Research.