The First AI Entrepreneur to Leave the ICU | A Conversation with Binson, CEO of Xinying Suixing
Summary
A rear-end collision at roughly 100 km/h left Binson in the ICU for 6 days and transformed his entrepreneurial mindset from “unable to let anything go” into calm resolve. The taxi stalled in the fast lane without hazard lights; seconds later, it was hit from behind and spun roughly 720 degrees. More than a dozen ribs on his right side fractured, as did ribs on the left; displaced bones punctured his lung. All he could think was: “For my wife and my child, I can’t die here.” Surviving did not make him give up entrepreneurship. Instead, it convinced him: “If I’ve chosen to keep building a company, then I have to make it work. Otherwise, I’d be letting down this second life.”
Being forced away from the company for 3 months unexpectedly proved that Duoduo Game Partner had the organizational foundation to operate without founder heroics. The accident came as financing agreements were being finalized, yet the team completed the financing work during his hospitalization while the business and user growth continued moving forward. Binson came away convinced that “the best management is when the company runs the same way when you’re absent.” After returning, he delegated more aggressively while tightening performance reviews and actively letting go of people who could not keep pace. Delegation is not “autopilot”; it requires a group willing to move the vehicle rather than hitch a ride.
Duoduo Game Partner’s core product insight was not “build a desktop pet,” but that AI should not compete with games and short video for users’ time. Binson believes AI-generated content cannot be better than human content or games. “AI can use productivity tools to improve creative output and lower costs, but it cannot replace the director.” The company therefore chose ambient interaction: narrowing from young people to anime and gaming audiences, then entering through the shared gaming context to turn danmaku into a character that understands the screen, offers攻略, and interacts. The desktop pet was a high-retention format discovered through iteration, not the day-one end state.
Duoduo Game Partner has accumulated roughly 6 million registered users, but the key asset is not the user count itself; it is the game-by-game data and UGC being built underneath. The team has roughly 40 people; PC has reached break-even, while mobile has not fully broken even. Revenue comes from skins, gifts, and memberships, with memberships accounting for roughly one-third; there is no gacha. Competitors can copy the desktop-pet appearance, but would still need to rebuild the game knowledge,攻略, and visual recognition layer. Duoduo Game Partner has standardized support for new games: Black Myth: Wukong and Infinity Nikki could be integrated within 1–2 days of launch, and users are now beginning to contribute game knowledge and visual data.
Binson’s view of AI applications in 2025 is “user education accelerates and the startup淘汰赛 begins,” not an immediate killer-app breakout. Products such as DeepSeek are giving users a more intuitive understanding of AI, but on the company side there will be more shutdowns, bankruptcies, and pivots. At the same time, people leaving foundation-model companies bring practical operating knowledge that could improve the quality of the second wave of application startups. He believes the industry follows an objective cadence: “When the killer app arrives, there won’t be just one.” Founders cannot pull the timetable forward by force of will.
Traditional ChatBot metrics such as time spent and conversation turns may be the wrong anchors; retention and monetization matter more, while the larger product opportunity lies in shared context. Binson sees conversation as “against human nature, with too high a barrier.” Search-like tasks should ideally be solved in one turn, with users searching and leaving. Free distribution at scale is not much of a moat either, because a large platform can catch up with tens of millions of users within 1–2 months. He is more bullish on AI seeing the screen, sharing project libraries, and sensing the physical environment: “Sharing context will become increasingly important.”
Binson’s path from Baidu and 360 to Bilibili and then entrepreneurship shows that a big-tech executive résumé can be both an asset and a liability. He rose continuously at Baidu and collected nearly every major award, gained an operating perspective at 360, and at Bilibili put ecosystem ahead of commercial efficiency. But he also admits he started too late: executives win inside platforms through resources, rules, and upward management, whereas entrepreneurship has “no referee” and must be judged by the market. If the goal is simply to make money, his conclusion is blunt: “Don’t start a company.” For someone with a valuable skill, staying at a large platform may offer a higher expected financial return.
Deep dive
1. A Highway Crash Turned Death from an Abstraction into a Scene
The accident happened during a business trip to Zhuhai in October. The taxi suddenly stalled in the highway’s fast lane. The driver neither moved toward the slow lane or emergency shoulder nor switched on the hazard lights; seconds later, a following vehicle hit it at roughly 100 km/h.
Before impact, Binson was leaning forward from the back seat between the two front seats to ask what was happening, so he never heard the driver’s reply. When he woke up, all he felt was “excruciating pain.” Only later did rescuers tell him that the vehicle had spun roughly 720 degrees.
More than a dozen ribs on the right side fractured, as did ribs on the left; displaced bones punctured his lung, and his arm was also broken. His head did not take a direct impact, and his spine was checked first, so he remained conscious from the scene through the hospital and experienced the pain and fear in full.
The vehicle had caught fire. A veteran who had traveled from Beijing for work and a China Southern ground-crew member used fire extinguishers and smashed open the left-side window. Binson abandoned his backpack and phone and told them to pull him out before the glass had been cleared: “For my wife and my child, I can’t die here.”
2. The ICU’s Torture Also Included Losing Control
Emergency checks were ordered by lethality: first ruling out a brain hemorrhage, then examining the spine and internal organs; the ribs and arm came later. It took a doctor asking, “Where is your family?” for Binson to realize this was not an injury he would simply leave after surgery.
While he waited after surgery, an ambulance from Guangzhou arrived carrying a box containing a kidney. He mistook it for the ambulance a friend had said might transfer him from Guangzhou, and assumed he needed a kidney transplant: “We’re done for—I need a new kidney.” When he woke up 2 days later, he did not even know which operations he had undergone, which intensified the panic.
His 6 days in the ICU moved through 4 psychological stages: first, not knowing how badly he was injured; then relief at surviving; then anxiety after beds around him suddenly emptied, making him worry that he too might disappear; and finally a clear trauma response. “You feel completely helpless. I couldn’t sleep at all for those few days.”
He repeatedly dreamed that he was trapped beside a reservoir and a cliff near the highway, unable to move, while the only child who appeared was too afraid to rescue him. A psychologist confirmed a typical traumatic stress response and compared it to still feeling the ground move after getting off a roller coaster. Once he knew his mind was not broken, he began to relax, sleep, and exercise actively in the ICU.
3. The First Bowl of Congee Compressed Happiness Back to Life’s Basics
As soon as Binson woke and confirmed his physical condition, his mind immediately returned to the company; he even called his partners to discuss work. Only afterward did he realize, “First, I shouldn’t have done that, and second, there was no need.” Once he left the ward, he simply turned off his phone and let his family decide what was important enough to pass on.
After moving from the ICU to a regular ward, his first bowl of congee made him feel that “life can be very simple, and happiness can be very simple.” A few days later, secretly eating a bite of meat produced the same realization: being able to eat, drink, and use the bathroom normally was already enough to constitute happiness.
He read a book by a professor of psychiatry who had spent years researching end-of-life care and later became bedridden after suffering a stroke in old age. People nearing death rarely regret not earning enough. They more often regret missing time with family, failing to commit to what they truly wanted to do, or waiting too long to reconcile with people they had been unable to make peace with.
4. Stepping Away from Work Was Both Recovery and an Organizational Stress Test
曲凯 shared that he used to misunderstand the idea that “the more things there are and the more decisions need to be made, the more you should sleep first.” By the next day, many issues no longer seemed so important, and decisions made under pressure and haste were not necessarily correct. Repeatedly falling into firefighting often means the mechanisms upstream were never properly built.
曲凯 said founders often treat “I have to personally watch everything” as a form of responsibility. Binson’s correction was that letting go is not enough; mechanisms must be built in normal times, or the company cannot operate normally in the founder’s absence.
The accident came as financing agreements were being advanced, yet the team handled the related work smoothly and the business and user growth did not stop. Binson even observed that because no one was waiting for him to make the final call, team members realized that “we now have to make every decision ourselves,” which instead unlocked more energy.
He did not begin returning to the office for a few hours every 2–3 days until 3 months later. The business direction had not changed, but viewing the team and organization from the outside gave him a pair of seemingly contradictory shifts: more patience with objective timing, and greater resolve on matters he had already thought through.
5. Calm Did Not Soften Management; It Accelerated Delegation and淘汰
His first move after returning was to expand decision rights. Regardless of seniority or prior résumé, anyone with an idea could run an experiment and receive resources. 曲凯 added the prerequisite: delegation must be matched by talent quality, or “authorization” simply becomes a situation in which no one is accountable.
His second move was to conduct a strict year-end performance review and actively let some employees go. Binson had previously feared conflict and tended to be a nice manager. His partner reminded him that “kindness does not command an army”: the company needed to be lean rather than large, and everyone had to keep pace with its development.
His boundary is that delegation is not “autopilot.” People unwilling to learn, standing still, or merely riding along on the train should leave. Experiencing a brush with death did not make him more permissive toward everyone; it reinforced his responsibility to “help everyone get the job done.”
6. If Entrepreneurship Is Only about Money, Its Expected Wealth May Trail Big Tech
Binson breaks the meaning of life into 3 layers: security; social relationships and recognition; and self-exploration, ultimately centered on “change and growth.” The accident confirmed for him that people are first creatures of experience and emotion, moving outward from the self to family, friends, and society before returning to self-exploration.
He believes this generation in China, shaped by memories of scarcity, is unusually sensitive to material goods and money. During an exchange at Stanford, by contrast, he saw many people whose first motivation was to do something “really remarkable,” even to change humanity or society; money was not necessarily the primary driver. If the sole objective is to get rich quickly, his advice is: “Don’t start a company.”
7. Baidu’s Award Grand Slam May Also Have Kept Him from Leaving Too Soon
After joining Baidu through campus recruiting, Binson spent years in the Hive and search organizations, moving from coding into product and management. He was promoted at least once almost every year and maintained the highest performance rating. He collected nearly every award available, including Best Individual, Best Manager, Best Team, the President’s Special Award, and the $1M Award.
At the time, Baidu had only roughly 100 directors. An ECF once compared a director to “the governor of a province.” But after winning the awards, he gradually realized that “it’s actually not that big a deal.” The real value was having caught the internet’s high-growth cycle: “When the water is high, the fish grow big.”
He once rejected outside offers paying 2–3x more because he believed the early priority should be learning. If growth was fast enough, he should ask whether compensation could increase 10x over 3–5 years, rather than cashing in immediately for 2–3x and then spending years selling old experience.
In hindsight, he admits that leaving Baidu in 2017 was already late. Toutiao and Kuaishou had extended offers in 2014–2015, but when you are on the best ship of the moment, it is hard to see that faster ships are already alongside. The “first miss” was not a failure to identify the company; it was underestimating the speed of the industry’s transition.
8. 360 Made a Function Leader Accountable for the Operating Result for the First Time
From 2017 to 2020, Binson started in 360 Search monetization before taking over information-feed monetization and other consumer businesses. 周老板’s key attraction was not compensation but pulling him out of performance optimization and technical optimization and making him accountable for revenue and the overall operating result.
His working style was not to repeatedly complain upward but to deliver results with solutions: “I know about all these difficulties. If I could solve them, why would I need you?” He committed at the beginning of the year to revenue and monetization efficiency and tried to exceed both by year-end. This was also why he considers himself one of the few executives who was rarely scolded by 周老板.
Binson valued the simplicity and directness: say what you want, say what you dislike, but do not deceive. Raising a problem is not scarce. What is scarce is answering, under existing constraints, “What can you solve, and what can you do?”
9. Bilibili’s Choice Put Ecosystem Ahead of Commercial Efficiency
When choosing a platform again in 2020, Binson studied Bilibili, Xiaohongshu, and Pinduoduo. His judgment was that “Pinduoduo is a company I could own shares in, but not work for,” while Beijing’s talent market was already relatively saturated and several Shanghai companies were short of senior talent. Leaving Beijing after more than a decade was not the core obstacle.
What attracted him to Bilibili was not narrow anime culture or danmaku, but its role as a content and information community with cross-generational reach. China lacked a YouTube-style outlet for high-quality content; creators were proud to call themselves “Bilibili creators,” and professors were willing to teach courses there. High penetration among young users pointed to long-term incremental growth.
While responsible for the UP主 ecosystem and monetization, he refused to copy big-tech playbooks directly. Many candidates before him had immediately suggested switching to vertical swiping and adopting Douyin’s traffic model. His starting point was to understand why mid- and long-form video, UGC, and the community had worked in the first place.
A dual-column feed is better for exploration, smaller creators, and community; a single-column swipe feed is better for head content and monetization. Binson used Kuaishou as an example: copying the model can quickly lift revenue, but may erase the original character. “Two very simple changes can bring very different things underneath.”
10. A Big-Tech Executive’s Capability Model Does Not Convert Directly into Startup Odds
Binson is blunt that executives have relatively low startup success rates. The platform, resources, and rules are supplied by the company; executives level up, manage upward, and can win by producing 3 standout results out of 10. In entrepreneurship, no one defines whether the long board or short board matters more: “There is no referee in entrepreneurship. The market is the one doing the validation.”
He ultimately decided to leave in 2023 because he had already seen the Transformer architecture and had already worked out the direction. If he knew he had the ability to do it but failed to act, he would regret it 10 or 20 years later even if he had made money. That regret was less acceptable than the pressure and opportunity cost of entrepreneurship.
After starting the company, he began by removing “big-company thinking” from himself: no private office, sitting with the team instead; fewer reporting layers and less hierarchy; no “little patch of turf”; and no internal horse-racing, because an early-stage company should be working together on the same job.
Once big tech enters a stock-based competition phase, the higher the business level, the more likely employees are to spend 80% of their energy away from the business itself. As the pie shrinks while bosses still demand growth, organizations continually replace people and repeat old projects. Frontline employees eventually become numb, mechanically executing the next reporting cycle.
11. Understanding the Boss Requires Correcting Both Information Gaps and Memory Bias
曲凯 argued that a person’s decisions are only roughly 60%–70% correct; what matters is getting the critical moments right. Because subordinates occupy different positions and have different information, they may understand only half of a decision. Subjectively, they see the boss as being right only 30%–40% of the time, while remembering failures more vividly, so the conclusion naturally becomes: “The boss is stupid.”
曲凯 added the survivor bias in project selection. Employees repeatedly cite 3 projects the boss rejected that later surged, while forgetting the dozens of other rejected projects that genuinely failed. Binson agreed and suggested first calming down and looking for missing information, then asking whether the decision was simply within the other person’s normal 30% error rate.
Binson trains commercial judgment through a method resembling shadow mode in autonomous driving. Before a company or historical figure makes a choice, he first writes down, “What would I do?” and then checks it against the real-world result. Studying history should not mean memorizing answers; when Cao Cao faces a specific context, one should pause and run one’s own decision experiment first.
12. Duoduo Game Partner Started from Not Competing for User Time, Not from Building a ChatBot
When choosing a direction in 2023, Binson judged that as everyone crowded into foundation models, he should build applications. The team briefly made a product resembling “cute AI,” but shut it down after only 1 week and returned to exploring the intersection of virtual characters and games.
His first-principles judgment was: “AI-generated content cannot be better than human-made content, cannot be better than games, and cannot be better than short video.” AI can improve efficiency and reduce the cost of creation, “but it cannot be the director.” The product therefore should not compete for another block of exclusive user time.
“To go big, start small” was implemented through successive narrowing: from young people to anime and gaming users, then to those who commonly play games. The team was not trying to compete with game companies on game content. It instead asked: “Videos have danmaku. Why can’t playing games offer something similar?”
The resulting product is an anime-style character that watches the game beside the user, understands the screen, chats, and offers攻略. It turns danmaku into interaction while providing emotional value. The earliest inspiration was closer to a virtual streamer avatar, with AI replacing the person behind the avatar; the desktop pet emerged as the form users discovered through iteration.
13. A Narrow Wedge Must Still Answer Why a Giant Cannot Kill You Once It Enters
Binson rejects the self-comforting belief that “Tencent won’t do it” or “ByteDance won’t do it.” A founder cannot hold Zhang Yiming’s head down and stop him from building something, just as Wang Xing cannot prevent Zhang Yiming from entering local services. The only controllable question is whether the startup still has an independent ecosystem position after the giant enters.
His test is direct: “If you are doing the same thing as him, within his range of capabilities, you definitely cannot beat him.” The moat therefore cannot simply be being first to launch a similar feature. It must come from a game-data and product ecosystem that is difficult for a giant to flatten directly with existing distribution and organizational capabilities and that requires continuous construction.
For Duoduo Game Partner, RAG is only an implementation method. The real accumulation is the game-by-game construction of knowledge,攻略, and visual recognition. The team already has a standardized process: it can support new games such as Black Myth: Wukong and Infinity Nikki within 1–2 days of launch, and has begun letting users contribute to the game knowledge base and visual data.
14. Hiring and Removing Big-Tech Inertia Are the Hardest Organizational Problems after Starting Up
Even after years of business and management training at multiple big-tech companies, Binson says the biggest problems he encountered after starting up were still hiring and getting everyone to adopt a consistent method. He initially preferred big-tech candidates, but later found that they first had to be trained as “individual developers” capable of rapid cross-functional experimentation.
The fastest growers are often people with 2–3 years of work experience: they have basic skills and business understanding but are not yet locked into old processes. More experience is not inherently bad, but people must actively acknowledge that it can become a burden—“discard all your past experience and don’t treat it as seniority.”
Being emotionally invested in the product is largely decided on day one and cannot be manufactured through onboarding. Someone who has never played anime-style games can learn, but genuinely identifying with the direction and being willing to treat it as their own is more a hiring criterion than a training outcome.
The company is primarily based in Beijing and recently established an office in Shanghai. It is hiring across engineering, algorithms, product, operations, content growth, and game art; Japanese-language backgrounds are especially welcome for the Japan business. The common bar remains self-direction and genuine alignment.
15. Six Million Users Does Not Mean the Company Is Already on Track
Binson disclosed that Duoduo Game Partner has roughly 6 million registered users, mainly in China but also across overseas markets including Japan. Retention and monetization are better than those of typical AI products. 曲凯’s challenge was that desktop pets naturally retain users. Binson accepted the point: “If it naturally retains users so well, why wouldn’t I use it?”
The product did not begin by designing a desktop pet and then searching for demand. It discovered through iterating on an ambient character that the desktop pet fit the time users spend playing PC games. Binson’s product philosophy is: “A product is not designed by a genius on day one. It grows and iterates with users along the way.”
He believes the team found PMF relatively quickly, within 2024, and that entering from the companionship angle provided some first-mover advantage. But the process still depended on the objective laws of industry development.
Monetization does not rely on gacha. It comes mainly from skins, gifts, and memberships, with memberships accounting for roughly one-third. The team has more than 40 people; PC has reached break-even, while mobile has not fully broken even. So “doing okay” on revenue is only relative to the sector and does not mean the company is profitable overall.
His current state is “optimistic, but cautiously optimistic.” The AI market is large enough and long enough, but Duoduo Game Partner is not yet fully on track. The next goal is to expand overseas and Japan, ultimately bringing domestic and overseas revenue to roughly a 50/50 split.
16. The AI淘汰赛 Has Begun, but the Killer App Remains Bound by Objective Timing
Binson believes products such as DeepSeek will accelerate user education and help more people feel AI’s capabilities. On the industry side, however, 2025 will mark the start of the淘汰赛: more companies may go bankrupt, shut down, or pivot, while competition intensifies materially.
A second positive force is that people from foundation-model companies are starting to build new applications. Most first-wave application founders learned AI on the fly from traditional internet businesses. The second wave has hands-on experience and a clearer understanding of what models can and cannot do, which should improve the efficiency of industry exploration.
Progress in models and voice anthropomorphism has partly exceeded his early expectations, while multimodal perception has fallen short. GPT-3.5 was already a massive leap; the move from 4 to 5 should never have been assumed to happen instantly. For Duoduo Game Partner’s current business, the models are basically sufficient; he simply hopes multimodality advances faster.
He opposes founders trying to force the market to mature early. PMF and application breakouts follow objective patterns; before the time is right, one team cannot push an entire industry into breakout mode. “When the killer app arrives, there won’t be just one.” Calm therefore does not mean giving up; it means refusing to substitute anxiety for construction.
17. The Next AI Interface Is Shared-Screen Context, Not More Conversation
Binson believes the ChatBot may not be the right end state: “It goes against human nature, and the barrier is too high.” Search should be search-and-leave; if the answer is accurate, the interaction should ideally end in one turn. Conversation turns and time spent may therefore be poor metrics. Retention and monetization remain more reliable: whether users vote with their feet.
The free-for-scale, burn-to-finance path is also difficult to sustain. Even if a startup accumulates tens of millions of users, a large platform with an existing entry point may catch up within 1–2 months. Being small and beautiful is not necessarily enough either; user scale without data or product moats is equally meaningless.
He further argues that the AI era may have no MVP. When mobile-internet technology matured, an MVP could reach 60 points and then validate demand. An AI product may only reach 20 points because of model limitations. Failure at that level does not mean it would still fail at 60 points; both the product form and business model are waiting on the technology variable.
The larger opportunity is for AI to see the screen, project libraries, and ultimately the physical world while sharing the user’s context. Better on-device capabilities can ease privacy concerns. As PC, iOS, and Android gradually open permissions, screen sharing may become as common as sharing a PPT in a meeting. Duoduo Game Partner will continue changing form because “we haven’t reached the final round yet. Change is what’s right; not changing is what’s wrong.”