Rokid co-founder: Meta's AI agent Muse topped US app stores within two weeks of launch

September 22 — The first Immersive Consumption Carnival and 2026 AI Terminal Innovation Development Conference kicked off in Beijing. During the event, a range of innovations from the AI terminal industry chain were showcased.

Red Star Capital Bureau observed on-site that numerous experts and industry professionals were engaged in discussions centered around "AI terminals." Coincidentally, an AI agent application developed by Meta has recently gained widespread attention.

AI Industry Shifts from "Making Models Smarter" to "Putting AI to Use"

The AI industry is transitioning from the first half—focused on making models smarter—to the second half, which emphasizes putting AI into practical use. AI agents have become the key link connecting large models, terminal hardware, and real-world scenarios.

As AI agents move from the cloud to terminals, the form of terminals themselves is also evolving. Zhao Qinping, an academician of the Chinese Academy of Engineering, introduced the concept of "Human-Centric Intelligence" (具人智能).

He explained that "Embodied Intelligence" (具身智能) addresses how to make machines perform tasks more intelligently, essentially centering on the machine. In contrast, "Human-Centric Intelligence" focuses on how to make intelligence better understand humans, placing people at the center and enabling AI to coexist with humans through dual-brain coupling and collaboration with the human brain.

Zhao believes that future terminals will no longer be cold tools but collaborative partners guided by human intent. They will take multiple forms, including AI glasses and AI smartphones.

"Over the next decade, large-scale and deeper AI普及 will rely not only on robots but also on these human-centric terminals," Zhao said.

AI Glasses Seen as a Key Form of Human-Centric Intelligence

Among the various terminal forms, AI glasses are regarded as one of the closest carriers to "Human-Centric Intelligence." Xiang Wenjie, co-founder of Rokid, said in an interview with Red Star Capital Bureau and other media outlets that the stronger AI becomes, the more it needs an optimal carrier to complete tasks such as shopping, recording, translation, and programming. He believes glasses could become the mainstream terminal for the next 10 to 20 years.

He noted that the key to the widespread adoption of AI glasses is not how impressive a single technology is, but how lightweight they are—light enough for users to wear all day. "When they are light enough and comfortable enough, even if the only functions are translation or taking photos, users will be willing to wear them."

During the interview, he revealed that the AI glasses industry is accelerating, with market volumes moving from early-stage tens of thousands of units to tens of millions. Technological iteration is also speeding up. Regarding the "iPhone moment" for AI glasses, he estimated it could arrive in three years under optimistic scenarios, or within five years under conservative ones. The key indicator, he said, is when users no longer frequently pull out their phones but instead rely on an on-device assistant to complete high-frequency tasks like shopping and communication.

Meta's "Muse" AI Agent Gains Traction

In the interview, Xiang Wenjie mentioned the recently popular "Muse," an AI agent developed by Meta. Within less than two weeks of its launch, it topped the free app charts on both the U.S. App Store and Google Play. Muse can perform various online tasks, including shopping, sending emails, and making phone calls.

"Technically, there is no difficulty—it is essentially a personal assistant. This aligns completely with our understanding. In the AI era, everyone needs a role that understands you and can help you with everything," Xiang explained. He added that its capabilities can be developed through continuous AI learning and improvement, and over time, it will come to know the user better. Unlike previous approaches that simply executed commands, it can proactively interact with users, analyze, and share. Xiang believes this will become a mainstream direction with enormous value in the future.

AMD: The Rise of the "Agent Computer"

"With the arrival of the AI era, everyone has the potential to become a 'super individual,' and the intelligent agent terminal is precisely the entry point for realizing this vision," said Ji Zhaohui, Vice President of Marketing for AMD Greater China. He also noted that in the three years since ChatGPT launched in 2022, major global large-model companies have focused on making models smarter. Now, the center of gravity in the AI industry is shifting from training to inference.

He told Red Star Capital Bureau that as generative AI accelerates its expansion from the cloud to terminals, on-device AI is evolving from applications primarily focused on interaction and assistance to an agent stage capable of autonomously executing complex tasks.

In March of this year, AMD introduced the computing category of "Agent Computer," emphasizing the ability for AI to continuously and autonomously complete tasks locally. "In the past, computers were used by people. An Agent Computer is designed for AI agents. Traditional computers need keyboards, mice, and monitors because we constantly interact with them. An Agent Computer does not require these peripherals because the agent runs directly on the device," he said. He added that in the future, intelligent agents will play a crucial role in enhancing enterprise capabilities.

Source: Red Star Capital Bureau