A conversation about "lobsters" reveals the key variables in AI development

Securities Times reporter Han Zhongnan

“Actually, when I first started using OpenClaw (commonly known as ‘Lobster’), I wasn’t very adapted.”

The response from Xia Lixue, co-founder and CEO of Wu Wen Xin Qiong, brought knowing laughter to the audience at the 2026 Zhongguancun Forum Annual AI Open Source Frontier Forum. On March 27, during the roundtable discussion, Yang Zhilin, founder of Dark Side of the Moon, posed a core question to four peers in the AI field: What is your deepest impression of using OpenClaw? How do you view its evolution compared to other related agents?

Xia Lixue admitted that he initially found the “slow response” of this “lobster” difficult to adapt to, but he quickly changed his tone: “Later, I realized that it is not just an agent responsible for chatting; it’s more like a ‘person’ that can help me complete large tasks.”

This statement encapsulated the core consensus of the roundtable discussion—AI is moving from “chatting” to “really getting things done.”

Zhang Peng, CEO of Zhipu, sees OpenClaw as the “scaffolding” of AI, capable of building a sufficiently stable, convenient, and flexible framework on top of model capabilities, allowing ordinary people to easily use top models without mastering code.

Huang Chao, assistant professor at the University of Hong Kong and head of the Nanobot team, summarized the unique value of OpenClaw with a popular internet term—“human-like feel.” He stated that many past agents felt more like “tools,” while OpenClaw, through its interaction method embedded in IM (instant messaging) software, brings AI closer to the image of an intelligent assistant in people’s minds.

As the discussion deepened, the future landscape of AI agents gradually became clearer. Yang Zhilin pointed out that open-source models and inference computing power are forming a new ecosystem. With the explosive growth of Token, the entire industry may gradually shift from the training era to the inference era.

This judgment was strongly supported by Xia Lixue. “Since the end of January, our Token volume has doubled every two weeks, and it has grown tenfold so far,” he said. “The last time I saw such a growth rate was during the proliferation of mobile data in the 3G era.” In his view, this is a signal of a transformative era; only by optimizing and integrating existing resources can AI truly serve everyone.

However, the explosive growth in Token demand also brings real challenges. Zhang Peng admitted that having smarter models perform more complex tasks consumes vast resources, and the required Token volume could be ten or even a hundred times that of answering simple questions, which is also the reason for Zhipu’s recent price increase for the GLM-5-Turbo model.

Luo Fuli, head of Xiaomi Mimo’s large model team, provided a perspective from the angle of technological evolution. She believes that “self-evolution” will be a key trend in the AGI (Artificial General Intelligence) field in the coming year. “With the support of a powerful self-evolving agent framework, large models will bring exponential acceleration to scientific research.” She revealed that her team, leveraging top models and agent frameworks, has improved research efficiency nearly tenfold.

At the end of the dialogue, Yang Zhilin asked each guest to use one word to forecast the development trend of large models in the coming year. Huang Chao chose “ecosystem,” Luo Fuli emphasized “evolution,” Zhang Peng focused on “computing power,” and Xia Lixue placed greater importance on the industry’s “sustainability.”

“As a Token factory, whether we can continuously, stably, and on a large scale output usable Tokens to enable top models to serve more downstream scenarios is what I care about the most,” Xia Lixue expressed with practical enthusiasm. He further elaborated on his vision: “In the past, we talked about Chinese manufacturing, transforming cost advantages into high-quality products for global output; today, we want to make AI Made in China, through high-quality Tokens, making China the global Token factory.”

In this roundtable dialogue, the “lobster” seems to have transcended its significance as an AI open-source agent, triggering a deep reflection on the future development of AI in China. From “chatting” to “really getting things done,” from a manufacturing powerhouse to a global Token exporter, every step of exploration hides the codes of industry evolution. How to walk steadily and far down this path from technological breakthroughs to global breakthroughs is indeed the key variable worth observing in the AI industry in the coming year.

(Edited by Dong Pingping)

     【Disclaimer】This article only represents the author's personal views and is not related to Hexun.com. Hexun.com maintains neutrality regarding the statements and viewpoints contained in the article and does not provide any explicit or implicit guarantees regarding the accuracy, reliability, or completeness of the content. Readers should only use it for reference and assume all responsibilities themselves. Email: news_center@staff.hexun.com