Show Notes
When James Peng founded Pony.ai in 2016, many in Silicon Valley believed autonomous driving was only three to five years away. But he expected it would take at least a decade, because the challenge was never just teaching a car to drive. Commercialization also required regulatory approval, public trust, reliable fleet operations, and a cost structure that could support large-scale deployment. Ten years later, his vision is becoming reality. In this conversation, we start with his founding journey, the milestones and how Pony.ai became a leader in the autonomous driving space. We also discuss the gap between assisted driving, Level 4 autonomy, and the longer-term goal of Level 5, as well as how Pony.ai uses simulation, real-world driving data, and increasingly capable AI models to improve safety. James explains that the hardest problems are often not the obvious ones but interpreting unpredictable human behavior and handling rare edge cases consistently. The conversation also explores China’s cost advantage in robotaxis. A mature automotive and electronics supply chain, close collaboration with automakers, and faster iteration can materially lower vehicle and system costs. But moving into new markets still requires Pony.ai to adapt to different road conditions, regulations and driving cultures, from trams and roundabouts to local pickup behavior. James’s broader point is that the industry has focused too heavily on the initial technological breakthrough. Getting a car to drive itself is only the beginning. The next phase is about deployment density, utilization, maintenance, charging, remote support, and economics. At the end of the conversation, I asked what he believes is underrated. James, an experienced operator, replied - scaling. Pony.ai may have crossed the zero-to-one threshold, but the harder task is scaling from one to ten, and eventually from ten to one hundred. Check out this insightful conversation. For more interesting conversations with people who are charting the way of the future of AI, check out the podcast tab or follow us on Spotify! Chapters 02:36 Why James Peng founded Pony.ai06:36 The milestone that proved robotaxis could work09:27 How passengers learned to trust driverless cars11:41 Level 2, Level 4 and Level 5 autonomy18:59 How AI and simulation improve self-driving24:13 Teaching cars to understand human behavior29:11 China’s cost advantage and global competition35:03 Expanding robotaxis into international markets41:13 Why Pony.ai is also building autonomous trucks48:40 Adapting to new cities, roads and driving cultures53:45 Why scaling is often harder than reaching zero to one Transcript Grace Shao: Hi everyone, welcome back to another episode of AI Proem Differentiated Understanding. This is your host, Grace Shao. Look where I am, the back seat of a car. Doesn’t look that exciting, does it? Let me flip this around. Look at that. There is no driver. I’m in the back seat of a Pony.ai robotaxi. Today joining me is James Peng, co-founder and CEO of the leading robotaxi company. It’s expanded its footprint across the globe, in Asia, in Europe, in the Middle East. But obviously, today we’re in its leading home market, China and Shenzhen, where it has a fleet of a couple hundred vehicles deployed on the streets already. Hi James, thank you so much for sitting down with me. I’m really excited to be having this conversation with you. So you left Baidu in 2016 to found Pony.ai when many in Silicon Valley were saying self-driving cars are only three years away. But obviously that wasn’t the case. Grace Shao: So, what did you believe then that this consensus was getting wrong? Tell us about your journey from 2016 until now. James Peng: Yeah, sure. We were founded in 2016, about 10 years ago. But even at that time, I didn’t believe that autonomous driving can be solved in three to five years. Just from a technical point of view, because even back then, 10 years ago, even a demo for autonomous driving was already v
More from AI Proem Podcast
This episode was a special collaboration project with Tencent’s Behind the Penguin podcast. AI Proem remains editorially independent but is open to collaborating with companies and individuals if that means they can bring better resources/ guests/ insights to the AI Proem audience. I speak to Serrino Liu, the AI Product Expert at Miora, Tencent. He is responsible for product innovation across film, 3D, and the interactive formats built on top of them, bringing AI agents into real content-creation and workflows. He tells me that Miora is not another image or video generator. Rather, it is a creative agent studio that runs entire workflows for filmmaking, e-commerce, game and 3D professionals. What makes the conversation unusual is Miora’s stance on models: it deliberately routes across mainstream models from different providers: DeepSeek for documentation, Seedance for video, Hunyuan Image for graphics, and models from direct competitors like ByteDance and Kling. Serrino is refreshingly candid about why: Miora is not a foundation-model company, and every improvement in the underlying models flows back into the product. He also unpacks Miora’s relationship with Tencent’s office agent WorkBuddy (”two pieces on the same board” where one manages the workspace, the other the creative work), and how the newly launched Hy Image3.5 preview is “professional-grade” in the sense that matters: usable, consistent, and able to survive revision, not just prettier. The heart of the episode is a nuanced take on taste and judgment. Serrino splits creative work into what can be verified, such as frame format, dialogue, and character consistency — and what only a human can judge: whether the emotion lands, whether the camera movement serves the story. Features like the new StoryWeave and side-by-side version comparison are built to keep the human in the loop, “a partner in making, not replacing.” The same philosophy extends to memory: Miora’s visible, editable memory system exists to kill tedious repetition — remembering your color palette, your story decisions, your style — so the next revision respects everything you’ve already decided, instead of forcing you to re-explain yourself. The episode closes with two big-picture exchanges. On geographies of preference in AI adoption, Serrino argues the difference between Hollywood and Asia is less about values than stakes: Hollywood’s AI debate arrived as a contract question (consent, credit, compensation), while Asia’s arrived as a capacity question — enormous demand meeting thin mid-tier production, with both anxieties eventually converging. And in his contrarian take on a fully automatic creative AGI, he lands the episode’s most quotable line: in creative work, the most expensive thing was never execution — it’s knowing who to pass the ball to. Craft will always change, but the author and the user never do. The AI Proem Podcast is part of the AI Proem newsletter, which has ~13k followers globally. To learn more about China AI, the business of AI, and how AI is impacting society, please check out the newsletter here and more insightful conversations here. Chapters 00:00 Welcome from Macau: Meet Serrino Liu 03:35 What Is Miora? Agents, Memory & the Professional Harness 04:53 One Product, Many Models: Routing, Rivals & Token Economics 11:14 Building for Real Creators: The StoryWeave Approach 18:12 Taste, Judgment & Keeping the Human in the Loop 28:09 Memory as a Creative Asset & the Hy Image3.5 preview Launch 37:20 Asia vs Hollywood: Two Different AI Stories 44:23 Misconceptions, Copyright & a Contrarian View on AGI AI generated transcript, for reference only Grace Shao: hello everyone welcome back to AI Proem Podcast this is your host Grace Shao. Today’s episode is very special because we’re partning up with Behind the Penguin a newly launched Tencent backed podcast. We’re in Macau today because we are at the 2026 International Film Camp and joining us today is Serrino Liu, Tencent’s Miora AI product
In this episode, I speak with Kristy Loke, an AI governance researcher focused on US-China relations, about what the latest Trump-Xi meeting could mean for AI governance, safety, and cooperation between the two leading AI powers. We discuss the possibility of an incident reporting mechanism for malicious AI-enabled cyber activity, the growing risks around frontier models, and whether the US and China may have more shared interests than the current competition narrative suggests. One part of our conversation I found particularly interesting was Kristy’s framework for understanding China’s AI strategy. Rather than simply racing toward the frontier, she describes a four-pronged approach focused on staying close to the technological frontier, translating AI into productivity gains, securing core technology supply chains, and protecting people’s welfare. This helps explain some of China’s decisions around AI diffusion, chips, and regulation of emerging technologies such as AI companions. Kristy argues that the American approach is more closely tied to technological leadership and power, while China’s approach is more pragmatic and focused on longer-term economic and social goals. That distinction also raises questions about how AI safety is framed, particularly when scientific concerns about frontier models become intertwined with geopolitical and commercial interests. We ended by looking at whether meaningful US-China coordination is actually possible. Kristy sees a potential path from continued dialogue to sharing safety practices and, eventually, common standards, even as competition remains intense. We also talked about what middle powers such as Singapore, the UK, and countries in Europe can do as AI governance increasingly becomes a conversation between the world’s two leading AI powers. Relevant articles by Kristy: * https://www.theinformation.com/articles/americas-ai-dream-failing-launch?rc=sqpcn6 * https://www.wired.com/story/china-isnt-buying-silicon-valley-call-for-ai-slowdown/ The AI Proem Podcast is part of the AI Proem newsletter, which has ~13k followers globally. To learn more about China AI, the business of AI, and how AI is impacting society, please check out the newsletter here and more insightful conversations here. Chapters 00:00 AI Governance and International Relations 02:48 The Stakes of AI Safety 05:42 China’s Pragmatic Approach to AI 08:05 The Dynamics of US-China AI Competition 10:56 The Future of AI Cooperation 16:05 The Four Prongs of China’s AI Strategy 24:36 Contrasting AI Visions: China vs. America 27:43 Personal Journey into AI Governance 36:09 The Intersection of Governance and Development 38:59 Non-Consensus Views on AI Leadership 40:24 Middle Powers in AI Governance AI- Generated Transcript (for reference only) Grace Shao (00:00) Hi Kristy, thank you so much for joining us today. Kristy Loke (00:02) Hi Grace, great to be here. Grace Shao (00:03) Yeah, so you’re studying in Toronto, but you are watching the Trump-Xi meeting very closely. I I believe you were just on like all kinds of media throughout the day commenting on what happened. Help us understand the high level takeaways. Xi and Trump just met in DC, I believe it was yesterday, and AI governance, AI safety was very high on the agenda. Kristy Loke (00:24) Yeah, I think a good way to think about it is good vibes actually matter in this kind of summit where the actors play a major role. Xi and Trump clearly have a rapport. And I thought it was really good to see a continuation, if not a formalization, of the AI dialogue that started in May. so it’s good to see that. And then there were some pre-summit discussions by Bessent and He Lifeng, the Chinese and US representatives. about potentially setting up an incident reporting mechanism which would meet some of the growing needs for them to either de-escalate or warn each other of like let’s say malicious actor activity around models. Grace Shao (01:03) So what is the de escalation plan lookin
In this episode, I speak with Aoni Wang, product designer at Harvey, and Yiyang Hibner (Silicon & Spice with Yiyang), product manager at Airbnb, about how AI is changing the way they work day-to-day. This conversation is a bit more light-hearted than the usual conversations on AI Proem, but I think it is equally insightful because these two professionals work in the heart of Silicon Valley, and they give us a glimpse into how people are really adopting AI. Both are already using AI as part of their regular workflows, from thinking through ideas and understanding complex systems to research, writing and building interactive prototypes. We explore the tension between making execution faster and deciding what is actually worth building. As AI makes it easier to produce prototypes and generate work, Aoni and Yiyang discuss the risk of confusing speed of output with speed of progress. More work can be produced, but that also creates a greater need for filtering, judgment, and accountability. They also discuss why AI can be too agreeable, and why users still need to challenge its outputs, verify information and bring their own expertise to the process. The conversation then looks at how AI could change the boundaries between product, design and engineering. Then they discuss the growing importance of skills such as systems thinking, prioritization, communication, taste and the ability to define what a good product or experience should look like. For Aoni, this includes thinking about visual language and coherent systems rather than simply producing individual screens or prototypes. Finally, we get into the more practical questions around working with AI, including how teams maintain shared context when everyone has their own AI tools, how people decide which products are actually useful, and where human judgment remains difficult to replace. The conversation offers a view of AI adoption from inside the workflow rather than from the perspective of the technology itself, raising a broader question about the future of work: as more of the execution becomes easier, how will people decide where to focus their time, attention and expertise? The AI Proem Podcast is part of the AI Proem newsletter, which has ~13k followers globally. To learn more about China AI, the business of AI, and how AI is impacting society, please check out the newsletter here and more insightful conversations here. Chapters * 00:00 The Evolution of Design Work with AI * 05:14 AI’s Impact on Product Management * 07:59 Efficiency vs. Overwhelm in AI Utilization * 11:26 Navigating AI Slop and Quality Control * 14:19 The Future of Roles in Design and Product Management * 17:59 The Negative Impacts of AI on Work * 21:22 The Changing Landscape of Collaboration with AI * 28:03 Shifting Workflows: AI vs. Traditional Software * 30:17 Creative Tools: Evaluating Effectiveness and Usability * 33:38 The Importance of Human Skills in an AI-Driven World * 36:49 AI’s Role in Decision Making and Workflow Management * 40:57 Personal Use Cases: AI in Everyday Life * 46:28 Differentiated Views: Perspectives on Work Culture Transcript (AI-generated, for reference only) Grace Shao (00:00) Hi guys, thank you so much for joining today. Really really excited to have you on. Yiyang Hibner (00:03) Thanks for having us. Aoni Wang (00:04) Thanks for having me, Grace. Grace Shao (00:04) Yeah, so before getting started, I know you guys both have some disclaimers from the company you have to say. So go ahead, do your thing. Yiyang Hibner (00:12) All right, I’ll go first. I’m Yiyang, I’m based in Bay Area and currently I work at Airbnb. And opinions are my own. So you know, Airbnb is one of the companies I’ve been working at in the past few years. There are a lot of other tech companies, so like some experiences are not unique to the current company I’m at. Yeah. Aoni Wang (00:30) Hi, I’m Aoni. Similar to Yiyang, I’m based in the SF Bay. I now work at Harvey as a product designer. But today all vi
In this episode, I speak with Ang Li, co-founder and CEO of Simular AI, about the rise of computer use agents and his vision for computers that can increasingly do work on behalf of humans. We discuss why computer use matters beyond the current wave of API-based agents, particularly for the vast amount of enterprise work still carried out through legacy desktop software that was never designed to be accessed programmatically. We explore where these agents could have the most immediate impact, from processing invoices and extracting information from unstructured documents to navigating financial, healthcare and other enterprise systems. Ang argues that the goal is not necessarily to remove humans from the workflow, but to shift the balance between execution and judgment, with agents handling repetitive tasks while people remain responsible for decisions and sign-off. The conversation also gets into the economics and technical challenges of computer use agents. Ang explains why relying on a frontier model for every click can be expensive, slow and difficult to control, and how Simular’s neurosymbolic approach turns repeated workflows into executable playbooks. We discuss the role of smaller and open-weight models, the changing economics of AI agents, and why Ang sees computer use as a way to make automation more accessible beyond the highest-value technical work. Finally, we look at what this shift could mean for SaaS and the future of work. Ang distinguishes between systems of record and software that primarily serves as a portal or interface, and argues that these categories may face very different levels of disruption from agents. We end with a broader question about human agency: if AI increasingly handles execution, will the more valuable skill become knowing how to identify the right problems to solve in the first place? The AI Proem Podcast is part of the AI Proem newsletter, which has ~13k followers globally. To learn more about China AI, the business of AI, and how AI is impacting society, please check out the newsletter here and more insightful conversations here. Chapters 00:00 Introduction to Simular and Its Vision 05:14 The Future of Work and Human-AI Collaboration 10:21 Technological Landscape and Market Demand 15:31 Real-World Applications and Use Cases 20:40 Challenges and Competitive Landscape 29:55 The Dual Nature of Work: Content Creation vs. Execution 33:50 The Competitive Landscape: Frontier Labs vs. Open Weight Models 37:17 The Rise of Computer Use Agents 43:30 Empowering the Average Person: High Agency Through Technology 48:46 Understanding SaaS: Infrastructure vs. Portals 52:32 The Future of AI: High-End Jobs vs. Repetitive Tasks AI-generated Transcript (for reference only) Grace Shao (00:00) Ang, thank you so much for joining us today. Really excited to have you. To start with, could you tell us a bit more about yourself and why you built Simular? What is your long-term vision for the company? And what really brought you here along your academic journey? Ang Li (00:13) Yeah, thank you so much, Grace, and thank you for having me here. My name is Ang Li. I’m the co-founder and CEO of Simular. We’ve been working on this company for almost three years. In short, we call ourselves the autonomous computer company, meaning we are building autonomous computers. The goal of Simular is basically this: everyone has computers right now, but we have to work on them manually by moving the mouse, typing on the keyboard, looking at a screen, and understanding what’s going on ourselves. We envision a future where computers will do the work on behalf of humans, on their own. That’s really the technology that we’re building towards. Nowadays people also call them computer-use agents. It’s basically a general-purpose agent that can use the computer just like a human. A bit about my background: I’ve spent almost 20 years researching AI. My personal research direction has basically been trying to figure out what people someti
Hello everyone, I've been on the road, so I'll be delaying this week’s written post and next week’s podcast episode. Apologies in advance. In this episode, I’m joined by Asia Bureau Chief Jing Yang and Senior Reporter Juro Osawa from The Information to unpack DeepSeek’s surprising shift toward external capital, its unique investor structure, and what its rapid revenue growth reveals about the economics of AI. We also explore why China’s AI labs remain fiercely competitive despite a far smaller capital market than the US. The conversation moves from DeepSeek and Huawei to accusations of distillation, ByteDance, Alibaba, and Tencent, and the growing race to turn AI models into real businesses. We also look at China’s emerging advantage in robotics, from its dense hardware supply chain to the vast amounts of real-world data being generated through deployment. Finally, we discuss the next frontier: world models. As the race moves beyond language models, can China’s hardware and data advantages translate into an edge in embodied AI? And in a provocative final take, Jing argues that Chinese frontier models may never fully overtake their US counterparts. The AI Proem Podcast is part of the AI Proem newsletter, which has ~13k followers globally. To learn more about China AI, the business of AI, and how AI is impacting businesses, please check out the newsletter here and more insightful conversations here. Chapters 00:52 The Rise of Chinese AI Models 04:06 DeepSeek’s Unique Position in the Market 12:31 Capital Structure and Investor Dynamics 16:12 The Landscape of Chinese AI Labs 19:17 DeepSeek and Huawei: A Strategic Partnership 25:43 The Competitive Landscape of AI Labs 33:51 Challenges in the Chinese Capital Market 36:44 ByteDance’s Unique Approach to Model Training 41:32 The Future of BAT Companies in AI 48:22 China’s Robotics Landscape and Advantages 53:51 Impact of US Regulations on Robotics 54:43 Unitree’s IPO and Market Dynamics Grace Shao (00:00) Hey Jing, hey Juro thank you so much for joining AI Proem today. Juro (00:03) Hey Jing Yang (00:03) Yes. Grace Shao (00:04) to start, you know, let’s start with Ox alpha. That was such a teaser. It turns out to be ZAI again, but I think it was very much expected by people who do watch space closely. It was no surprise in that sense. But I think what’s shocked a lot of people is that, you know, the inference was running on Chinese domestic ships and potentially GLM five point three flash. The price, it’s priced at about like one fortieth of Opus 4.8. That’s quite significant of a gap, right? Tell us about what you think about that, the implication on these continued Chinese open weight models that are coming out strong but cheap, and how that’s impacting US frontier models. Juro (00:43) Yeah, maybe I can start on this one, but I think those flash models are coming out, like Deep Seek had the V4 flash, which is a smaller size one, and that became very popular. And then this one comes out. Alibaba just had a three point eight flash as well. So those are smaller size, you know, lower cost models that can still handle a lot of like AI Asian type of tasks really well. So know, this is like a sort of sweet spot for demand, right? So a lot of Chinese companies with open source models are coming out. So that’s one, you know, one thing about this, you know, that’s part of that broader trend. But I think as you pointed out, the chip part, you know, it kind of shows how far the Chinese chips have come in terms of inference, right? And They have become more capable, especially Huawei chips, being used, you know, more and more for inference. And you know, Deep Seek came out with V4 and you know it was also optimized for you know to run on Huawei chips as well. And we’re gonna see more and more of this for sure. But I mean to what to keep in mind that when it comes to training, it’s not the same story yet, not quite yet. So, you know, a lot of Chinese companies are using, still using the most advanc