New Year 2026 Conversation with Koji: The Year of R
December 14, 2025
This episode marks another year of cross-show collaboration between "This Is For Real" (Cihua Dangzhen) and "Crossroads" (Shizi Lukou). Yusen Dai, Managing Partner at ZhenFund, and Koji from Crossroads pick up where they left off last year, reviewing 2025 and looking ahead to 2026.
In last year's opening conversation, we called 2025 the "Critical Year for AI" and the "Year of Agent," predicting more applications would move from concept to real deployment. Looking back now, that call is being validated — AI applications exploded this year, and the first truly meaningful Agent products emerged, including Manus, Genspark, and Claude Code. Meanwhile, commercialization of AI applications is accelerating.
This year's conversation also takes us back 18 years, recalling what Xing Wang and Yiming Zhang were like when they were still in their twenties — and from there, expanding into longer-term reflections on entrepreneurship, innovation, and investing. Of course, we also share our outlook for 2026. To sum it up in one phrase: the "Year of R":
- Return (commercial payoff): As investment scales up, people will focus more on AI's real returns. Quality of growth matters more than speed of growth.
- Research (frontier research): Current AI research paradigms are hitting bottlenecks; new breakthroughs are needed to unlock the next phase.
- Remember (user memory): Memory will become a key differentiator for AI applications. Proactive Agent could be a 10x opportunity.
Looking ahead to 2026, AI will continue evolving rapidly, with both opportunities and challenges. But no matter how times change, ZhenFund will always stand with founders. We believe that truly excellent entrepreneurs are the constant in an ever-changing world.
02:56 Reviewing 2025
- Last year, we said 2025 would be the "Year of Agent." Time to check the answer.
- Last year, we argued ChatGPT wasn't yet AI's iPhone moment — we were still in the BlackBerry era. What about now?
- The Ghibli moment happened this year? Felt like years ago already.
- The "Year of Agent" call was right, but "year one" means this won't be solved in a single year — Karpathy called it "Decades of Agent."
- How's the Doubao mobile assistant working out?
12:58 What's the next opportunity unlocked by models?
- Agentic capabilities have unlocked maybe 10-20% so far. At 80%, AI will be able to do vastly more.
- Multimodal fusion is the next big shift.
- There's a benchmark called Zero Bench — the strongest models now score 5 points, but in a year might hit 60-80.
- Elon Musk said: humans are basically machines that constantly receive visual input.
17:10 Has the "phase transition" happened?
- AI capability is like boiling water: before 100°C, you can only make coffee; at 100°C, you instantly unlock the steam engine.
- AutoGPT in 2023 was a concept; Manus in 2025 is a product with genuinely great UX — the difference between them is the water temperature.
- Ilya said AI has a scaling period and a research period; paradigm breakthroughs are needed to keep heating the water.
20:00 Products that have excited us lately
Yusen:
- Typeless: Not just voice input — it learns how you speak differently in WeChat versus Lark.
- Doubao mobile assistant: Still in preview, but the first time I felt AI could actually help you complete a task with real degrees of freedom.
- Honestly, the excitement level is lower than late last year — last year felt like a massive tide surging in; this year feels more like evolution from 1 to 10.
Koji:
- Product of the year: Sunday robot — the first robot that made me want to buy one and bring it home.
28:54 Koji's "Wudaokou Stories"
- "The first email from Xing Wang" — Koji reads the letter
- What did Xing Wang say to Koji, pointing at the Wudaokou crossroads?
- Yiming Zhang back when he was still coding: compiling took ten-plus seconds. While others spaced out browsing the web, Yiming muttered: "What can I do with these ten seconds?"
- Picasso said: "When art critics get together, they talk about form, structure, and meaning. When artists get together, they talk about where to buy cheap turpentine."
41:39 How would Xing Wang and Yiming Zhang be different in the AI era?
- The "Tao" of entrepreneurship doesn't change: learning ability, leadership, innovation, willpower.
- But the "Shu" is upgrading: 20 years ago it was Copy to China, 10 years ago it was China-specific models, now it's global from day one, world-first.
46:36 In an era without maps, how do you build? How do you invest?
- What's the same between 2009 and today? Everyone senses opportunity, but nobody knows the answer.
- Two things that definitely won't be wrong: take active action + go where excellent people cluster.
- The founder of Walmart's kids dreaded road trips with Dad — because in every small town, he'd stop to examine supermarket shelves.
- Project direction is hard to judge, but the quality of excellent founders is relatively easier to assess.
54:01 How do you find Yao Ming in the AI era?
- If Yao Ming is in the room, you won't miss him. But the NBA also has players under 5'6".
- Resilience matters more than ever: faster iteration in the AI era means more frequent failure.
- You need your own judgment on technology trajectories: build what models can support in 6-12 months, so you become a pioneer rather than a martyr.
- Zhilin Yang in 2023 judged long context to be crucial; Xiaohong saw the Agent opportunity because he built an AI browser first.
- The AI table deals 10-20 cards a year. Missing on one hand is normal — fold and play the next.
01:03:44 Advice for founders who want to come on a podcast
- Don't need to package yourself as super successful, or try to seem super smart.
- The real tragedy isn't saying something wrong — it's working your ass off on something nobody knows about, nobody discusses, nobody cares about.
- Media needs good content; good content = interesting + useful + resonant.
01:08:58 Entrepreneurship vs. investing: two completely different lifestyles
- What's the investor job like?
- Entrepreneurship: going extremely deep on one thing. Investing: knowing a little about many things.
- Entrepreneurship has a progress bar but is exhausting; investing has no progress bar but is anxiety-inducing.
- Don't take what investors say too seriously — if they really knew, they might not be investors.
01:13:23 [Key Outlook] 2026 Prediction: Year of R
Yusen's annual framework: three R's will define AI trends in 2026, making it The Year of R.
Return
- For the past three years, the market traded on the "I" in ROI (Investment). Next year, more attention will go to "R" (Return).
- ChatGPT subscription price hikes are hard: intelligence that sold for $200 last year now sells for $20.
- What used to be valuable becomes worthless — AI won't earn programmers' salaries; it will devalue programming itself.
- Silicon Valley is already shifting: investors increasingly care about gross margin, retention, and cash flow, not just growth.
Research
- Ilya says we've re-entered a research phase; Demis says we need 1-2 more breakthroughs to reach AGI.
- New Silicon Valley trend: invest in Neo Labs, hoping they explore research paths different from the leading model companies.
- Thinking Machines Labs valued at $50 billion — more than the combined valuation of all Chinese generative AI startups.
Remember
- Memory is the key differentiator for AI applications; ChatGPT retention shows a smile curve.
- I asked ChatGPT where to go for Spring Festival — its recommendations were far better than Gemini's, because we'd been talking for three years.
- Next step: Proactive Agent — a good assistant doesn't wait for the boss to call, but proactively prepares the materials.
01:42:51 Our annual recommendations
- AI product of the year?
- Podcast of the year?
- Book of the year?
- What's your happiness index this year?
2025 opening conversation with Koji: AI's Critical Year, Agent's Year One
Good AI should be like Doraemon | Koji x Yusen
"This Is For Real" (Cihua Dangzhen) is a general business podcast produced by ZhenFund, where the ZhenFund investment team will share the latest hot topics and industry insights with leaders across various fields.
Founded in 2011, ZhenFund is one of China's earliest angel investment institutions. Since its inception, ZhenFund has been actively seeking out the best entrepreneurial teams and era-defining investment opportunities in artificial intelligence, chips and semiconductors, robotics and hardware, healthcare, enterprise services, new energy, cross-border expansion, consumer lifestyle, and beyond.
ZhenFund — your first stop in entrepreneurship!