AI Reflections, Part Two: The Open Cards and Hidden Cards in AI | VOICE
Consensus and Non-Consensus in AI.
3,660 words, 9-minute read
By: Bingjian Wu
On AI, I've previously written A Framework for Thinking About AI Application Startups in the GPT Era, which included a section on consensus and non-consensus plays in AI. I'm continuing that thread here.
Open Hands and Hidden Hands
Any idea that comes to mind easily today is an open hand. You're laying your cards on the table from day one, competing on experience points, firepower, and health bars. Think of the most-discussed directions: AI assistants, virtual tutors, AI customer service, copywriting generators, and so on. In these obvious directions, there are too many masters and seasoned players. Simply playing an "early mover" card is rarely enough to win. Take AI education as an example. Education is a long-chain industry — from user acquisition to course operations, curriculum R&D, to sales — each link has established methodologies. In the end, it's a contest of "turning screws" efficiency. The industry is packed with veterans, many of whom have raised over $100 million and led thousand-person teams, just waiting for the right PMF to launch their next act. If a newcomer enters AI education, barely validates their PMF, and is just starting to learn how to run user acquisition or manage a hundred salespeople, just as they're about to upgrade from pistol to rifle, several masters pop out of nowhere and mow them down with Gatling guns.
What are hidden hands? Hidden hands tend to be non-consensus plays — completely zero-to-one product innovations that tap latent needs in markets with extremely fuzzy boundaries, ones that giants and masters either can't imagine or won't deign to look at. Snapchat, Airbnb, Kuaishou, ByteDance, Bilibili, Pinduoduo, Xiaohongshu — in their early days, all felt somewhat niche, with unclear potential for mainstream adoption. In 2013, if you'd asked whether the entire nation would watch 30-second short videos, there'd have been no clear answer. These hidden hands buy early-stage startups a multi-year window to grow quietly. By the time cards are on the table and it's a firepower contest, everyone has Gatling guns, and you've at least got a fighting chance in the shootout.
Open hands mean consensus. They typically target existing markets, replacing old solutions with new ones. Tenfold improvement is the ultimate weapon. Maybe it's a better copywriting tool, a better teaching tool — enhancing experience and efficiency.
Hidden hands mean non-consensus. They typically correspond to incremental user segments, addressing unmet needs. Zero-to-one innovation is the ultimate weapon. Before Kuaishou and Douyin, short video was barely validated — they conjured the short video paradigm out of thin air. Before Pinduoduo, lower-tier consumers barely shopped online.
If you're a veteran rich in experience points, skilled at winning through business acumen and efficiency, playing open hands can work. At a high enough level, it can even scare off and clear the field.
If you're an AI product manager with less experience but sharp user insight and a talent for innovation, you're better off thinking about hidden hands — entering through a unique, non-consensus angle. Leading with open hands is disrespectful to the Gatling gun.
The Fool's Window
Hidden hands become open hands once revealed. The fool's window — that period when a hidden hand remains hidden — is critical.
My judgment is that compared to mobile, the AI industry cycle will be much longer. Not one wave of opportunity, but multiple tides rising and falling — at least a decade or more. But for each specific project's entry point, the fool's window will be much shorter.
ChatGPT launched in November 2022. Three months later, major domestic and international tech companies had reached consensus — this was a must-fight battle, the AI foundation layer had to be won. They counted their chips and went all in. The moves were substantial: Microsoft launched New Bing and Copilot, Adobe launched Firefly, Google, Amazon, and Meta each entered from their angles, and BAT, ByteDance, and Huawei all joined the fray. For the platform layer, the fool's window was just two to three months — unprecedented.
Compared to mobile applications, AI applications' fool's window will also be much shorter. The dark forest principle will play out in AI applications with devastating intensity. In the mobile era, ByteDance, Kuaishou, and Pinduoduo had 3-5 year windows before the giants realized they'd been outflanked and began sniping and chasing. In mobile's dark forest, hunters fired when they saw torchlight. In AI's dark forest, they don't wait for torchlight — the moment a match is struck, a rocket launcher is already incoming.
Why do I predict the dark forest principle will intensify in AI? Three reasons:
First, ChatGPT and Midjourney were consumer hits from the start, with effects visible to the naked eye. AI possessing cognitive ability was unprecedented. Generating such photorealistic images was unprecedented. It was a wake-up call everyone instantly understood.
Second, the talent density in today's internet industry is unprecedented. A troupe of powerhouse actors had just finished a ten-year commercial epic called mobile. Then the director announced: "Folks, internet entrepreneurship is game over. From now on, we're only shooting hard tech. Go find your own way." Now the director has changed the script: PC was season one, mobile was season two, and seasons three through six are all AI. There's acting to be done again — going all in is non-negotiable!
This group includes both financially free Big Tech executives and mature founders who've scaled businesses to billion-dollar revenues or tens of millions of users, raised over $100 million, and led thousand-person teams. Their mentality is remarkably consistent — "The universe owes me a major win." These people are searching daily for the right entrepreneurial script.
Third, everyone's judgment has been "pre-trained" — they can anticipate your anticipation. Thirty-plus years of tech history, business warfare, and investment history constitute massive training data, effectively forging a tech-business large model. When we encounter an astonishing product or explosive news, our brains tell us: is this a big opportunity or small, how might the landscape unfold — yesterday once more.
Under the foreseeable dark forest principle, if you're playing a hidden hand, the right posture may be: less PR, head down turning screws, quietly making a fortune.
If you're playing an open hand, then make noise, unite all forces that can be united, build real strength — and see if you can play a few hidden hands on top of that open hand.
Where Are the Hidden Hands?
Let's look at mobile era's major hidden hands: Pinduoduo, ByteDance, Kuaishou, Bilibili, Xiaohongshu, miHoYo, Shein — all flanked their way from the periphery to platform-level main battlefields, all experiencing two phases: dismissed, and unstoppable.
They were dismissed because next-generation platforms initially appeared as features. At first, ByteDance was a sub-function of Baidu News — auto-recommendation. Pinduoduo was a group-buying game inside WeChat — much like an operational campaign. Kuaishou was a tool for converting video to GIF. A sharp enough feature can strike a vertical audience and generate high retention. But how large is this vertical audience? That's the second layer of fog. Statically, this audience might be 5% — will it become 50% in the future? Can it generalize successfully? This is usually controversial. Nearly all major products start with vertical audiences, then continuously generalize to become national-level products.
They're unstoppable not simply due to first-mover advantage, but because that advantage transforms into some structural advantage:
The first is scale-driven effects — network effects, two-sided effects, or economies of scale. Douyin, Kuaishou, Bilibili, Xiaohongshu are naturally community-oriented. Competitors can copy features pixel-for-pixel, can outspend on user acquisition, but beyond a certain scale, community culture forms — and that culture can't be copied.
The second is chain-as-asset — simply put, long business chains requiring both mouse and cement, with startups doing all the dirty work. Leaders in food delivery, ride-hailing, logistics, OTA, and education all built this way.
The third is natural barriers. Pinduoduo is the classic example — it leveraged WeChat traffic to grow, while WeChat was closed to Taobao. That's a natural barrier.
In other words, what's easily noticed? Innovating within existing large categories draws attention. Mobile browsers, for instance — hit a million DAU and you'd be 100% targeted. By 2012, all major companies had mobile browser projects, staffed with relatively elite teams. In the PC era, browsers were entry-level products that could intercept search traffic. The giants' nerves had already been sensitized. Cross the DAU threshold, and they'd send defensive teams immediately.
But feed streams weren't easily noticed. There was no feed stream category in the PC era, and feed stream KPIs were the opposite of search KPIs — feed streams wanted maximum time spent, kill time; search wanted minimum steps, getting users to results and out the door. So they didn't naturally connect.
For AI entrepreneurship, what are the hidden hands? This is extraordinarily hard to answer — if I could name them now, they wouldn't be hidden. Or put differently: what tables naturally produce hidden hands? A few examples to get a feel.
The first is broad entertainment. Fundamentally, ByteDance, Kuaishou, Pinduoduo, Bilibili, Xiaohongshu, and miHoYo are all in the entertainment business. Entertainment is an infinite game — water takes no fixed shape. Get the play pattern right, the approach right, and you can always find incremental audiences. And the biggest incremental audience is always the next generation. Many products carry community attributes; community culture is the hardest part to copy, and beyond a certain scale, they truly become unstoppable.
AI has three obvious characteristics: natural language interaction, generative capability, and multimodal perception. Based on any one characteristic, entirely new play patterns can emerge. You can imagine, for instance, that a game like Genshin Impact could become a truly open world inhabited by intelligent agents, where even random NPCs are AI-driven, and language interaction becomes a crucial plot-driving mechanism. Or consider virtual character chat apps — Character.AI, MiniMax — new entertainment play patterns driven by AI.
The second is long-chain, heavy-chain industries, typically service-related. Meituan's food delivery has 6 million riders. In 2012, the vast majority of internet companies would have chosen light models over heavy ones — a "dismissal" born of avoiding the heavy lifting. But these heavy-chain industries often come down to screw-turning efficiency; chain-as-asset can later become unstoppable.
AI will transform many industries, especially knowledge-selling services: law, medicine, education, programming, accounting, customer service, and so on. Software-as-a-service will become result-as-a-service. What's a result? For many industries, the result is AI plus human. Take legal services — perhaps the most valuable approach isn't providing software to legal practitioners, but building an AI-driven law firm yourself, with a fully AI-empowered middle office solving cost and non-scalability problems, while front-end client communication still happens through lawyers who provide deep service and emotional value. Privatized data and heavy service chains may make it unstoppable.
The third is serving "invisible" customers. Shein is the perfect example. Shein is a fast fashion brand backed by China's supply chain, selling overseas. It adopted the Shein brand in 2014 and stayed low-profile for six to seven years. By the time the general public heard of the company, annual sales were already in the hundreds of billions of RMB with a valuation in the tens of billions of dollars. This low profile was partly deliberate, but largely because it served "invisible" customers — entirely overseas, forming virtually no competitive dynamic with domestic players. This is almost the happiest startup state: not too much story, therefore not too much trouble.
Of course, there are many more directions suitable for hidden hands. Welcome to discuss — my thinking is limited, and these are just examples.
Now, the biggest open hand has already been revealed: the large language model foundation, and the ChatGPT application. Immense value, extremely high barriers. To say nothing else, tens of millions of dollars is the entry ticket; hundreds of millions gets you into the finals.
Both open hands and hidden hands offer paths to victory — the tactics just differ. Open hands basically come down to your cards, your chip count, absolute strength. Hidden hands offer the chance to grow quietly and punch above your weight, potentially hitting the jackpot, but also involving luck.
Meanwhile, open hands and hidden hands aren't opposites — they dynamically transform. A team building foundation models is playing an open hand with the model itself, but whether to build ChatGPT or some other application on top is a hidden hand. And all hidden hands, once targeted by giants and masters, become open hands.
Truth emerges from practice. Welcome to discuss more hidden hands together.
Heart Capital was founded in 2022 by Yan Han, formerly a founding partner at Lightspeed China. It is a China-based early-stage venture capital fund focused on technology and digitalization. The Heart Capital team consists primarily of Lightspeed China's founding partners, CFO, core investors, and senior industry investors from Cainiao and Baidu. The team's past investments include Full Truck Alliance (NYSE: YMM) and Xpeng Motors (NYSE: XPEV, HK: 09868) at Series A, as well as FinVolution (NYSE: FINV), 06810.HK, RoboSense, World Logistics, LandSpace, Lanhu, Micro-nano Star, Starfield, and others. Rooted in China with a global outlook, Heart Capital seeks to find world leaders who will disrupt the future. Heart Capital champions the value of "people" and "heart," and looks forward to accompanying more young Chinese entrepreneurs onto the world stage.
