Ⅰ. The Shift: From Authoring to Directing (The Strategic Signal)
The software industry is currently facing a paradoxical phenomenon: developers are spending less time writing code, yet software iteration speed is accelerating faster than ever before. This change is driven not by a new programming language, but by the AI-native paradigm known as Vibe Coding.
As former Tesla AI Director Andrej Karpathy coined the term, it provided a name for the era: humans no longer write code line by line; we transmit business intent via natural language to an AI, which automatically translates it into executable logic.
This is not merely a development tool; it’s a paradigm shift.
This transformation presents two opposed strategic pathways in the global market:
The US Path: Efficiency Maximization
The US Strategy is Efficiency-Driven, a manifestation of the Silicon Valley mandate: maximize speed and scalability. The focus is on Core Logic and Customization as the ultimate lever for competitive advantage. For instance, a top CRM platform successfully utilized AI-generated customized logic to reduce enterprise system implementation costs by over 40%. This is not a cost-cutting measure; it is a capital acceleration strategy.
The overriding goal is the strategic pursuit of efficiency maximization to secure a competitive edge through market velocity, rapid product iteration, and organizational reinvention.
The China Path: Risk-Controllability Priority
In contrast, China’s approach emphasizes system stability, auditability, and governance. The focus is on Compliance and Safety. A major state-owned bank, for example, embedded regulatory checks directly into AI-generated code, resulting in audit pass rates exceeding 98%.
The foundational goal is ensuring risk and compliance alignment so that the speed of AI does not introduce fragility into critical sectors.
This difference—speed-focused optimization versus stability-focused governance—is an important signal for global tech executives.
II. The Evolution of Code: Where U.S. and China Diverge Strategically
Vibe Coding did not appear overnight. It is the natural continuation of decades of shifts in how software is produced. Yet beneath this evolution lies a deeper truth: U.S. and Chinese enterprises made fundamentally different choices at each stage, and these choices now shape their Vibe Coding trajectories.
1. The Syntax Era (pre-2020): Repetition Reveals Two Philosophies
For over a decade, developers in both markets were weighed down by repetitive, low-value tasks — from Java interface plumbing to iOS compatibility debugging. But how each ecosystem responded foreshadowed today’s strategic split.
The U.S. doubled down on individual leverage.
Developers relied more on personal expertise and community-driven resources such as cheat sheets, “developer roadmap” repositories, and decentralized best-practice guides. The design assumption leaned toward individual autonomy in workflow decisions.
China prioritized collective efficiency.
As early as 2018, Alibaba open-sourced the P3C plugin to enforce unified coding conventions across teams. By eliminating style inconsistencies and common anti-patterns, P3C dramatically reduced avoidable defects. It became an early symbol of China’s preference for organizational consistency over individual preference.
These early choices — individual ability vs. collective efficiency — laid the strategic foundation for the next decade.
2. The Pattern Era (2020–2023): General Intelligence vs. Vertical Precision
As AI-assisted coding entered the mainstream, both markets again made opposite bets.
The U.S. pursued generalized innovation.
GitHub Copilot scaled to tens of millions of developers by 2023, accelerating coding through pattern prediction across the global open-source corpus. GitHub’s research showed an average 18% productivity gain, with a significantly higher impact for tasks that involve boilerplate code. YC mentors even called Copilot an “MVP accelerator” for early-stage teams.
The underlying belief: If a tool enhances individual capability, innovation will propagate organically.
China focused on deep vertical optimization.
Chinese companies built domain-specific templates tied tightly to real business logic—such as complex “618 promotion rules + inventory deduction” workflows—allowing developers to input intent and instantly generate compliant code. Many teams reported that generic AI outputs required heavy rework, while business-bound templates delivered predictable, ready-to-ship results.
This produced a second divergence:
The U.S. bets on general-purpose intelligence spreading across industries.
China bets on execution certainty through domain-specific depth.
3. The Intent Era (2023–2025): Capital Acceleration vs. Compliance Firewalls
As natural-language development became feasible, the two markets moved even further apart.
The U.S.: Innovation accelerated by capital.
Sequoia’s State of AI 2024 highlighted a structural shift: “Developers are moving from writing code to describing behaviour.” Tools like Cursor—capable of generating multi-tenant SaaS systems from natural language—became new VC darlings. The investment thesis was simple:
Chinese enterprises reinforced governance frameworks to manage uncertainty in AI-generated code. Alibaba, Huawei, and major financial institutions embedded code-security reviews, audit trails, and permissions controls into the development pipeline. Startups entering large enterprise supply chains reported significant compliance investment as table stakes.
Here, the divergence becomes structural』: The U.S. chases the speed threshold where innovation compounds. 『China builds a compliance framework where innovation aligns with system stability. These differing priorities reflect variations in regulatory environments, industry structures, and investment incentives.
4. The Vibe Era (2025– ): Organizational Reinvention vs. Execution Intensification
In 2025, Vibe Coding enters its fourth stage.
The U.S. path: “Small teams, larger output.”
Sub-5-person teams now ship products that previously required dozens of engineers and months of work.
The definition of “technical leverage” is being rewritten:
The unit of innovation is no longer the team, but the individual amplified by AI.
The China path: Scaling operational throughput.
Large enterprises use Vibe Coding workflows to enhance existing systems at scale.
Examples include:
Alibaba’s “Luban,” which reduces design cycles by 70%.
SHEIN’s AI-optimized supply-chain logic is improving turnover speed.
The logic:
The U.S. uses AI for new-market creation and product velocity.
China uses AI for system-level refinement and execution density.
Both strategies are rational, and both create competitive advantages.
III. Closing: The Strategic Commandment
Vibe Coding is accelerating a global shift in software development, where the U.S. focuses on universal development paradigms while China emphasizes domain-specific precision and governance. These differences stem from structural market conditions rather than cultural characterization.
A classical strategic principle applies here:
Sun Tzu’s The Art of War states:
“势者，因利而制权。” In modern business terms: “True leverage is not inherent; it is manufactured by recognizing and exploiting current market advantages.”
While Artificial Intelligence has defined the last decade, it remains constrained by classical computation — unable to fully capture the complex probability spaces that govern natural phenomena, such as molecular quantum behaviour.
Quantum systems, by contrast, compute through superposition and entanglement — the very mathematics of uncertainty.
When AI merges with quantum algorithms, the outcome is more than faster processing — it marks the dawn of a new paradigm for reasoning, creativity, and discovery.
II. Case — Insilico Medicine: Quantum + AI, Redefining Drug Discovery
Drug development is notoriously slow and costly — often taking over 10 years and costing around $2 billion per approved drug, with high failure rates due to the limitations of classical simulation.
Quantum computing aims to change that.
According to McKinsey, by 2035, quantum applications in life sciences could create up to $2 trillion in value, driven by advances in molecular simulation and protein folding.
In my book Strategic Development of Technology in China, I introduced Insilico Medicine, a pioneer in AI-driven drug discovery and among the first to explore integrating quantum computing with generative AI.
Book cover:
Its Variational Quantum Eigensolver (VQE) — a leading hybrid quantum-classical algorithm — has demonstrated improved molecular energy estimation accuracy, potentially enhancing drug target modeling.
By 2025, Insilico used a quantum-enhanced generative model to design novel molecules targeting KRAS mutations, one of the toughest challenges in oncology, achieving verified biological activity in early testing.
By combining quantum’s computational power with AI’s predictive intelligence, Insilico exemplifies how the two technologies can create a powerful synergy — accelerating discovery and redefining what’s possible in drug development.
III. Turing Test 2.0 — From Industrial Tools to Cognitive Engines
McKinsey’s long-term outlook (2040+) envisions quantum computing not merely as a faster processor, but as a cognitive amplifier — enabling humanity to simulate aspects of reality once unreachable by classical computation.
The true test of intelligence may ultimately lie at the intersection of information theory and the laws of nature.
Redefining the Test of Intelligence
The Classic Turing Test Evaluates whether a machine can mimic human conversation well enough to appear intelligent — rooted in a deterministic, logic-based framework.
The Quantum Turing Test (QTT) Extends this idea to ask whether an artificial system can model and reason within a quantum world of uncertainty and indefinite causality — a challenge that touches the frontiers of quantum gravity and computational theory.
Though still theoretical, the QTT expands our understanding of what “intelligence” could mean in a universe that resists deterministic explanation.
The New Definition of Intelligence
This shift reframes intelligence itself — from seeking absolute, predictable answers, to building self-consistent, adaptive models capable of operating amid fundamental uncertainty.
IV. Conclusion: Embracing the “Superposition Mindset”
The story of quantum is the story of humanity learning to master uncertainty.
From Einstein’s doubt — “God does not play dice” — to Insilico Medicine accelerating drug discovery with quantum-inspired algorithms, science has transformed uncertainty into strategy.
For decision makers, the key is adopting a Superposition Mindset:
In an age of rapid technological change, the future isn’t binary — it’s entangled, with multiple possibilities unfolding at once.
To lead in the quantum-AI era, we must invest for the long term and cultivate the ability to think, decide, and build amid uncertainty — not despite it.
I. Introduction: The Quantum Inflection Point — A Long-Term Consensus Between Capital and Science
Quantum mechanics is no longer an academic pursuit in the lab; it is a trillion-dollar variable actively reshaping industrial landscapes.
According to McKinsey’s Quantum Technology Monitor 2024, the global quantum technology (QT) ecosystem has attracted $42 billion in public investment. Private investment fell 27% in 2023 to $1.71 billion — but that’s far smaller than the 38% global average decline in startup funding. In other words, long-term capital confidence in quantum remains intact.
McKinsey estimates that by 2035, quantum technologies could unlock $2 trillion in economic value, with life sciences, finance, chemicals, and transportation as early beneficiaries. By 2040, the combined market size of Quantum Computing, Communication, and Sensing is expected to hit $173 billion.
The 2025 Nobel Prize in Physics has effectively pressed the “accelerator” on this global transformation.
II. The Nobel Prize Decoded: The “Tunnelling Signal” for Industrial Scale
The 2025 Nobel Prize in Physics honoured three pioneers — John Clarke, Michel Devoret, and John Martinis — for their foundational work in “Macroscopic Quantum Effects.”
Their work, at its core, involved creating a 1-centimetre-sized “Josephson Junction” using superconductors and insulators. They proved that billions of electron pairs could act as a single, giant particle, performing a seemingly impossible “collective tunnelling” (or “quantum wall-piercing”) — a quantum effect observed at a visible, macroscopic scale.
This discovery laid the foundation for today’s superconducting quantum computers, bridging the gap between theoretical physics and scalable hardware.
Notably, Martinis later led Google’s quantum team, achieving “quantum supremacy” in 2019, marking a historic moment where quantum systems outperformed classical computers on specific tasks.
The core message behind the honour is significant:
Removing Conceptual Barriers to Scalability: Demonstrating that large systems can show stable quantum behaviour provides the necessary theoretical and experimental basis for developing larger, more complex fault-tolerant quantum computers.
Strong Alignment with Investment Priorities: The Nobel recognition confirms the industry’s shift towards fault-tolerant quantum computing (FTQC). McKinsey reports that in 2023–2024, IBM, Microsoft, and IonQ have made key advances in “quantum error correction.” For instance, IBM utilized 288 physical qubits to preserve 12 logical qubits, resulting in a 90% reduction in error rate compared to traditional methods.
III. Case: IonQ AQ64 — The Commercial Quantum Computer You Can Access in the Cloud
IonQ follows the trapped-ion approach — using electromagnetic fields to capture charged atoms and encode data in their internal vibrations. This design achieves an operational error rate below 0.1%, the most precise among all mainstream quantum computing architectures.
Image source: IonQ
Hardware Leap and Commercial Scale: Its fifth-generation system, “Tempo,” launched in 2025, achieved a milestone of 64 Algorithmic Qubits (AQ). This means the system can simultaneously explore over 18 quintillion computational possibilities — a 268 million-fold increase in processing power over last year’s AQ36, significantly outperforming comparable published IBM systems.
Application and Ecosystem: Tempo has already demonstrated real-world impact across industries:
Energy & materials: simulating lithium-ion battery reactions with Hyundai;
Drug discovery: accelerating molecule screening with AstraZeneca;
Finance & optimization: modelling complex portfolios with Multiverse Computing.
Crucially, IonQ is the only quantum system simultaneously available on Amazon Braket, Microsoft Azure, and Google Cloud, dramatically lowering the entry barrier for enterprises and accelerating commercial adoption.
IonQ is not building prototypes — it’s engineering the backbone of the post-classical computing era.
IV. Closing: Why Investors Should Care Now
Quantum is no longer a far-future bet.
It’s entering the engineering and early commercial phase, where hardware performance translates directly into enterprise applications.
For founders, this means a new layer of computational advantage; for investors, it’s a once-in-a-generation inflection point similar to the semiconductor industry in the 1970s.
The Nobel Prize told us quantum coherence is real. IonQ proved it can be built, scaled, and rented by the hour.
The Mid-Autumn Festival* is here again. Moonlight spills across the eaves, carrying the faint sweetness of osmanthus on the wind. Centuries ago, the famous Chinese poet Li Bai once wrote:
“The people of today do not see the moon of old, But the same moon once shone on those before us.”
A founder friend recently asked me: “After a year of chasing projects and partnerships, what’s your most honest feeling this Mid-Autumn?”
My answer is simple: cherish what’s in front of you.
Not “someday.” Not “later.” Not after product-market fit or the next funding round. Right now.
The project you’ve been meaning to start with someone—do it together today.
The gratitude you’ve been holding back—say it now, directly.
The people you love—don’t save it for “after I’ve made it.” Offer your heart while you still can.
These aren’t lines from a self-help book. They’re truths learned the hard way—through regrets that cannot be undone.
I. A Missed Goodbye
The summer before I went to university, my grandmother came to visit. Afternoon light poured like warm amber across the room as she asked softly, “Shall I take you to register at university?”
Generated by Sora 2
Eager for the future, I declined gently, worrying she was too old for the journey. I promised I’d bring her once I had settled in. She smiled, said no more. That moment—sunlight in her hair, her quiet voice—turned out to be our last real goodbye.
There is no “later” for some promises.
II. A Mentor Without a Photo
In my darkest professional valley, a mentor stood by me. He saw a “blooming flower” and a “rising sun,” when I saw only data points of failure. Our most pivotal conversations happened on walks, a moving boardroom for strategy and soul.
When I began my startup, he fell gravely ill. To avoid distracting me, he chose radio silence—a final, devastating act of support.
I arrived at the hospital to find a man, once a towering figure, reduced to a whisper. I had to stand at the foot of his bed for our eyes to meet.
Generated by Sora 2
We knew each other for over a decade. We never took a single photo. Always waiting for a better moment, a less rushed day, a time when we weren’t so focused on the work itself.
III. The Dog Who Walked Closest to the Street
And then there was my dog, Bubble—forever cheerful, forever protective. On walks, she always positioned herself closer to the road, nudging me gently toward safety. When I called my parents on video, she would push her big head into the camera, competing to see whose face was bigger.
Generated by Sora 2
Through every triumph and every setback—funding wins, product failures—she was there, offering joy or quiet companionship. Her short life taught me the purest definition of love and loyalty.
Technology has given me a way to preserve fragments of those moments. Photos, videos, and even AI reconstructions keep traces alive. But they are no substitute for the living presence we too often take for granted.
What This Means for Us as Founders and Leaders
Entrepreneurs often live in the future—thinking in terms of scale, milestones, and exits. But life is happening now. And leadership is not only about building the next platform, product, or company; it’s also about how we honour the relationships and moments that give meaning to what we build.
The truth is: what softens us, sustains us, and makes us human is rarely the grand narrative. It’s the small things—the afternoon sunlight, a word of encouragement, a wagging tail.
As another Mid-Autumn moon rises, I return to an ancient line:
“Ancient and modern, like a flowing stream— We gaze upon the same bright moon.”
Generated by Sora 2
Time carries us all forward. But tonight, under the same moon as countless generations before, we still hold the power to pause, reflect, and treasure the present.
So I leave you with one question:
Who—or what—deserves your attention right now, before “later” becomes too late?
Final Note:
As founders, investors, and leaders, may we learn not only to build companies that last—but also to live moments that matter.
Note: The Mid-Autumn Festival is a traditional Chinese holiday celebrating family reunion, observed on the 15th day of the 8th lunar month, coinciding with the brightest full moon. A key tradition is eating mooncakes—dense, round pastries with a variety of sweet or savoury fillings that symbolize the full moon.
For years, AI has lived on flat ground — processing text, classifying images, and predicting numbers. But the world isn’t flat. The real test of intelligence is moving through a messy, unpredictable, 3D physical world.
This is the next frontier: spatial intelligence. And it’s not just a technical race — it’s a test of vision, strategy, and execution.
1. The Two Strategic Paths to the 3D Future
When it comes to building AI for 3D environments, two strategies are emerging:
1. Marble — Persistent, High-Fidelity Worlds Fei-Fei Li’s World Labs is behind Marble, a system that generates vivid, stable 3D spaces from text or images. Think gaming, metaverse design, or architecture — anywhere quality and persistence matter more than real-time change.
World Lab’s Marble Testing
2. Genie — Real-Time, Physics-Driven Worlds DeepMind’s Genie focuses on dynamic interaction and physical simulation. It generates environments that follow physics rules — ideal for robotics training, disaster response drills, and scientific simulation.
These aren’t rivals. They’re two sides of the same coin: one optimizes for creativity and permanence, the other for interaction and adaptability. Both point to the same core challenge: teaching AI not just to generate 3D content, but to understand 3D space.
2. World Labs’ “Large World Model” — Cracking the Code of Spatial Intelligence
Dr. Fei-Fei Li and the World Labs team are betting on the latter with their Large World Model (LWM). Their thesis is simple, yet profound: If AI is to become truly intelligent, it must master space before language.
Biologically, animals mastered spatial awareness (recognizing paths, finding food) hundreds of millions of years before humans developed complex language. Spatial intelligence is the “source code” for general intelligence.
Trilobite fossil specimen
World Labs’ bold move is an attempt to give AI three key abilities:
From 2D to 3D: Reconstruct objects and spaces from flat images using geometry and reasoning.
Generation + Reconstruction: Not just dream up virtual spaces but also digitize real ones with physical rules intact.
Scarce Data, Rich Reasoning: Shift from brute-force data collection to efficient spatial reasoning, overcoming the lack of labelled 3D training data.
3. The Fei-Fei Li Playbook: From ImageNet to World Labs
For every Founder and Investor, the trajectory from ImageNet (2009) to World Labs (2024) reveals Fei-Fei Li’s methodology:
Start from first principles: In 2009, ImageNet was dismissed as impossible. Her insight? If recognition requires data, build the dataset first.
“Do it, then prove it.”: She didn’t wait for consensus. Fei-Fei Li created the dataset first (ImageNet) by mobilizing 48,000 people to label 15 million images, betting that value creation beats theory.
Stay on the core logic: Just as ImageNet unlocked vision, World Labs is betting spatial intelligence will unlock robotics, AR, and embodied AI.
The entrepreneurial takeaway: when the logic holds and value is real, act before it’s obvious.
4. Ancient Wisdom for Modern Tech Cycles
The strategic risk of this shift is immense. The philosophy to navigate it comes from the ancient text, the I Ching (Book of Changes), specifically the Kan Gua (坎卦), representing Peril/The Abyss:
I Ching Kan Gua
“Xi Kan” (习坎) Challenges are normal: Treat challenge as the normal state of exploration. Innovation isn’t a smooth road; it’s a series of checkpoints.
“You Fu” (有孚) Hold your conviction: Maintain inner conviction and sincerity. In a capital market driven by hype, sincerity to the core problem is the source of resilience.
“Xing You Shang” (行有尚) Keep moving – like water, flow around barriers instead of forcing through them. World Labs embodies this: when 3D data proved scarce, they didn’t quit. They pivoted to reasoning-driven models—same goal, different path.
5. Why This Matters
For investors and executives, the message is clear:
Spatial intelligence is the missing link between today’s “flat” AI and tomorrow’s embodied, useful agents.
This is infrastructure, not hype — the foundation for robotics, industrial automation, metaverse, disaster response, and beyond.
The winners will combine deep tech with resilience — the courage to commit before the market consensus, and the adaptability to change tactics without losing direction.
AGI won’t arrive with another chatbot. It will arrive the moment AI can move through the world as confidently as it can talk about it.
And that journey, like all great ventures, requires both cutting-edge science and the ancient wisdom of how to cross numerous challenges.
Google DeepMind的Genie则代表了另一条路径：“实时交互与物理模拟”。作为世界模型，它专注于生成可根据指令实时修改、遵循物理规则的动态环境。其核心应用场景在于机器人智能体的训练（模拟现实物理规则以降低实体测试成本）、防灾应急演练模拟（复现地震废墟、火灾蔓延等动态场景）等，为科研与功能性训练提供了一个低成本、高效率的沙盒环境。
两类路径并非竞争关系，而是 “需求匹配” 的体现：若需落地商业创意，Marble 的 “高质量持久世界” 更高效；若需支撑 AI 科研或功能性训练，Genie 的 “实时动态交互” 更关键 —— 但它们共同指向一个核心问题：AI 的核心价值不仅是 “生成 3D 内容”，更在于 “理解 3D 空间逻辑”，这也是 World Labs 探索的核心方向。
二、World Labs 的 “大世界模型（LWM）”：让 AI 真正理解 3D 世界
2024 年 2 月，李飞飞团队带着 World Labs 敲开了空间智能的大门 —— 他们要做的 “大世界模型（LWM）”，核心目标是让 AI 像人类一样理解 3D 空间逻辑，实现 “感知、生成、交互” 三位一体的空间智能。这一决策并非偶然，而是基于对 AI 进化本质的深刻判断。
