On August 6, 36Kr learned that Agibot’s Lingxi division president Wei Qiang and co-founder Yan Weixin had both left the company. Their responsibilities are now being handled by Peng Zhihui on a rotating basis.
Yan is an associate professor and doctoral advisor at Shanghai Jiao Tong University’s School of Mechanical Engineering, chief scientist at the Shanghai Artificial Intelligence Laboratory, and a co-founder of Agibot. Wei, formerly of Panasonic, JD.com, Huawei, and Anker Innovations, played key roles in launching products such as Panasonic’s first high-definition set-top box in China, JD’s Dingdong smart speaker, and Huawei’s smart earbuds.
Nearly a month ago, news surfaced that Agibot planned to acquire at least a 63.62% stake in Swancor Advanced Materials. The initial block trade pegged shares at RMB 7.78 apiece. By August 5, Swancor’s stock had soared to RMB 110.48, a surge of over 1,300% across 20 trading days.
Market enthusiasm around robotics was already running high. But on August 5, the Shanghai Stock Exchange flagged irregular trading activity around Swancor, citing investor behaviors that disrupted market order or misled others, prompting account suspensions.
So what’s happening inside Agibot, the company behind the frenzy?
36Kr sought to understand the reasons behind Yan’s departure amid speculation of a “lack of capability.” A company insider, using the alias Yang Kexin, said Yan was better suited to academic environments and that his exit reflected a misalignment with Agibot’s current business priorities. Wei, by contrast, was described as “genuinely focused on building a great product.”
In fact, Agibot has seen a string of executive exits this year. “Given where the robotics industry is today, personnel change isn’t unusual. Plus, Agibot is still a young startup that’s barely two years old,” said Tina (pseudonym), a veteran headhunter.
Still, something about Agibot feels different.
Restructuring overnight
Last October, CEO Deng Taihua announced a sweeping reorganization of Agibot’s R&D division during an internal meeting, shifting from a centralized structure to business units organized by product line. The move may have been spurred by a new policy tailwind: national guidelines issued in late September 2024 aimed at accelerating mergers and acquisitions among listed companies, part of a broader push to develop China’s “new productive forces.”
According to insiders, the October meeting stretched late into the night.
“We were still working overtime the night before, and the next morning we were suddenly told the structure had changed,” several Agibot employees told 36Kr. The reorg appeared to be a spontaneous decision during that meeting.
Agibot’s R&D division was abruptly split into four parallel business units:
- Expedition, for general-purpose humanoid robotics.
- Genie, for wheeled dual-arm robots.
- Lingxi, for compact humanoid robotics.
- Commercial cleaning solutions, which spun off into its own subsidiary by year’s end.
The Genie division is led by former Nio engineering director Yao Maoqing, who also heads Agibot’s research institute for embodied intelligence. “He’s smart, emotionally intelligent, and knows how to lead without micromanaging. He’s really the soul of Agibot,” one colleague said. “He’s the ‘Cao Cao type,’ if you were casting a Romance of the Three Kingdoms remake.”
Yao brought on chief scientist Luo Jianlan, who helped establish a partnership with Google DeepMind’s Physical Intelligence (PI) team. Luo holds a PhD in robotics from UC Berkeley under PI co-founder Sergey Levine.
The Expedition division is led by Wang Chuang, who joined DJI in 2017 and later served as CTO at Anker’s robotics subsidiary. Known for empowering his team and decisive leadership, Wang is seen as a steady hand.
Wei, the former Lingxi division head, brought deep hardware manufacturing expertise from his time at Panasonic, JD, and Huawei, and was widely respected internally for his product-centric approach.
All three division leads were selected for their experience in mass production and commercial success, an indication that CEO Deng’s restructuring was designed to make each unit financially accountable.
Traditionally, robotics firms organize R&D vertically by function across hardware, software, algorithm, and full-system design. This allows tight control and efficient resource allocation. Unitree Robotics, for example, still adopts this model.
Agibot broke with that convention, assigning entire product pipelines to independent business units, each responsible for its own revenue and delivery targets. What had once been a unified team was suddenly split into sibling units, each carrying new performance burdens.
The company now publicly targets RMB 800 million (USD 112 million) in revenue, with a stretch goal of RMB 1 billion (USD 140 million).
The pressure builds
As commercialization took precedence, engineers found themselves under immense pressure, tasked with near-impossible deadlines while still accountable to tight internal benchmarks. The strain triggered a wave of high-profile departures, including motion control lead Li Zhen, embodiment team head Zhu Guangfei, and core developers Xiong Zhilin and Liu Jifan.
“Our work is packed with demos. We’re mentally overloaded, working 14-hour days with no time to actually refine our tech,” Yang Kexin said.
According to public data reviewed by 36Kr, more than 60% of Agibot’s R&D staff are postgraduates from top schools, including Shanghai Jiao Tong University, the University of Chinese Academy of Sciences, and Tsinghua University.
“Agibot’s people are some of the most hardworking in the field,” headhunter Tina (pseudonym) said. “They are incredibly driven and capable,” R&D lead Wang Luchuan (pseudonym) added.
Yet some industry veterans, like Lin Qingmu (pseudonym), argue that the sector should prioritize stability over speed when pushing products to market.
At this year’s World Artificial Intelligence Conference (WAIC), Wang Chuang boasted that Agibot’s robots could walk for over 360 hours without falling. Still, during the same event, one of Agibot’s Expedition robots, weighing 70–80 kilograms, toppled midway through the demo. The company attributed the fall to a low battery and has since promised a 24-hour live stream to demonstrate stability.
Still, the incident raised safety concerns. In industrial settings, a falling robot of that size could pose real hazards.
Other demos were also met with skepticism. Expedition A2-W’s WAIC logistics demo used identical soft clothing items, far from the messy variability of real-world warehouse environments. “This kind of task was already doable in 2018,” Wang Luchuan said. By comparison, Figure AI’s demo appeared significantly more adaptive and refined.
Agibot’s viral Lingxi X2 bike-riding demo also lost some luster upon closer inspection. The robot was mounted on the seat, relying on center-of-gravity physics rather than actual balance algorithms.
Wang Luchuan assessed the company’s progress bluntly:
“If Tesla’s at 30–40 out of 100, Agibot is at about 5. There’s still room to grow.”
The burden of brilliance
A company growing too fast can leave people behind.
“Zhihui is brilliant, full of ideas, and works really hard. But he’s never launched a complete product before, and now that he’s in the driver’s seat, it’s tough on the team,” said Qin Yi, a former colleague.
Peng Zhihui rose to prominence after winning nearly 40 innovation awards as a student. He later joined Huawei through its “Genius Youth” program and built a following as a tech influencer on Bilibili. In February 2023, he co-founded Agibot and took on the role of CTO.
In August 2024, Peng introduced the fully open-source Lingxi X1 robot. Two months later, he released the complete set of blueprints and source code. Priced at RMB 108,000 (USD 15,120), the robot was developed by Peng’s X-Lab team in just over two months with fewer than ten people. Reviews, however, were mixed. Customers reported issues such as uneven weight distribution, instability, and a lack of after-sales support.
“Zhihui meant well, he wanted to revive the open-source spirit of the early internet era,” Qin said.
By January, Agibot claimed to have shipped 1,000 robots: 731 bipedal units (Expedition A2 and Lingxi X1) and 269 wheeled models (Expedition A2-D and A2-W). But turning a lab prototype into a commercial product proved premature. Without dedicated support infrastructure, users faced usability and integration problems.
According to insiders, the Lingxi X2 is far more refined and better suited for mass production. It’s also expected to outsell the Expedition series this year.
“The X1 was built by X-Lab as a DIY kit, while X2 is a fully assembled product meant for real deployment,” Wang Chuang said.
Still, Peng’s dual role as CTO and head of X-Lab comes with challenges. His ideas are reportedly often bold but not always grounded in practicality. “You need hard data to convince him of anything,” one employee said.
The clash of visions and misalignment with team dynamics has led to further turnover. Notable departures include former algorithm director Gao We and former manufacturing general manager Zhang Shaozheng.
“There were directions Gao wanted to pursue. If the company had followed his lead, we might have made more progress,” said employee Yin Kuan (pseudonym).
RELATED ARTICLE
Hedges and ambitions
Unitree Robotics, founded by Wang Xingxing in 2016, only began IPO preparations in mid-2024, nearly a decade later. Now backed by CITIC Securities, it stands as one of the industry’s most established players.
By contrast, rumors in July suggested Agibot was pursuing a Hong Kong listing alongside its acquisition of Swancor. The company denied the reports. “We’re eyeing a domestic IPO by year’s end, but that’s still under discussion,” said Yang Kexin.
If Agibot does file for IPO by December, it will have reached the IPO stage in under three years, accomplishing in a fraction of the time what took Unitree ten.
Some insiders view the Swancor deal as a contingency plan for CEO Deng, perhaps a way to keep listing ambitions alive, even if Agibot doesn’t claim the title of China’s first publicly traded humanoid robotics firm. The logic: if a direct listing falters, a strategic merger could still offer a viable path.
For now, Agibot’s divisional model supports rapid, production-oriented growth. But without a coordinating layer, internal tensions could rise, potentially forcing a future reorganization to reintroduce horizontal platform teams and restore strategic cohesion.
So what has Agibot truly gained from the Swancor episode?
Beyond the headlines, perhaps only temporary relief. Investor sentiment has been reassured. Internal pressures may have eased, for now. But if Agibot does go public, it will have to answer to a more demanding audience: the open market.
The robotics boom has drawn its share of opportunists. But the industry’s long-term promise, and the talent driving it, deserve more than hype.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Chen Jiahui for 36Kr.