OpenClaw 3.28 Launched — Over 100 Major Updates in Four Days, Sparking Open‑Source Agent Debate

OpenClaw update floods community

It has been reported that OpenClaw — nicknamed the "red crayfish" (红色小龙虾) in Chinese tech circles — pushed a 3.28 release that delivered more than 100 major updates within four days, igniting a wave of adoption and scrutiny. The rapid release tempo landed at the heart of this year’s Zhongguancun Forum, where founders, academics and cloud vendors converged to ask: is open‑source the shortcut to practical, self‑evolving agents, or a scramble to scale before the compute bottleneck bites?

From scaffold to self‑evolving agents: forum debate

At the forum, heavyweight voices framed the moment. Zhang Yaqin (张亚勤) of Tsinghua University’s AIR argued we are shifting "from generative AI to agent AI"; Yang Zhilin (杨植麟) and Zhang Peng (张鹏) of Zhipu Huazhang (智谱华章) described OpenClaw alternately as a "possibility" and a “scaffold” that lets researchers and firms assemble agent stacks. Luo Fuli (罗福莉) and Huang Chao (黄超) stressed a core tension — do we build an all‑in‑one super agent or a lightweight, composable helper? The consensus: open source lowers entry barriers and accelerates ecosystem contribution, but engineering, safety and deployment remain hard.

Tokens, compute and national strategy

One practical consequence is explosive token consumption. It has been reported that some infrastructure firms saw token use double every two weeks since late January and grow tenfold overall — a surge compared by executives to the mobile‑data boom in the 3G era. That raises a policy and industrial question: can China convert its energy and manufacturing strengths into a sustainable “token factory” that supplies global AI demand? The issue is acute given wider geopolitical pressures — including U.S. export controls on advanced chips — that are already pushing Chinese players toward localised stacks, open‑source models and tighter data‑sovereignty solutions.

Risks and roadmap

Speakers also flagged risks: information‑integrity, physical and bio safety, and longer‑term existential concerns. Vendors such as Baidu (百度) warned of privacy and design pitfalls, while NetEase Youdao (网易有道) unveiled LobsterAI (有道龙虾), reportedly the first domestic fully open‑source desktop agent emphasizing local execution and auditable privacy sandboxes. The path ahead blends technical scale‑up with governance: can the industry turn rapid innovation into robust, secure products that serve science, industry and everyday offices without repeating past mistakes?