How Do Pros Use AI? How Do Regular People Learn AI? How Do Investors Bet on AI? | A Conversation with Ke Dai Biao Li Zheng

🚥 This episode of Crossing lands at an interesting inflection point: halfway through 2026, AI has become powerful enough to change how we work, but still falls short of "frictionlessly getting things done for you."

➤ In the first half, I invited a power user in my circle — Kelizheng Lizheng — to share his insights. He holds a PhD in economics from Cornell, founded Superlinear.Academy, and has worked at Amazon, Meta, and Tencent. He joined Statsig at its earliest stage, which was later acquired by OpenAI.

Kelizheng Lizheng explained why power users keep getting stronger with AI — they're constantly asking "what can I do for AI?" and delegating everything in life and work to it.

He also shared advice for ordinary people on using AI well — starting with this: stop using ChatGPT-style chat and switch to agentic tools like Codex, Claude Code, Cursor, and Manus.

➤ In the second half, we swapped roles. Kelizheng Lizheng interviewed me: my mid-2026 observations and reflections on AI venture capital.

I shared the hope and frustration I've felt on the front lines — the hope being unprecedented entrepreneurial and capital enthusiasm, plus the fact that "the optimal way for humans and AI to collaborate likely hasn't been invented yet," meaning friction itself is opportunity; the frustration being that commercial value has concentrated massively in foundation models, which have swallowed the vast majority of AI-generated returns, while the application layer often feels like "picking up pennies in front of a bulldozer," and the old software playbook (identify need → build product → sell to more people) has become harder.

Right now, startup and investment opportunities are polarizing: either leap for the sky (AI for Science / Physical AI) or dig deep into the ground (industry-specific workflow transformation and delivery / FDE).

Finally, I shared five AI startup directions I've been tracking lately. AI founders, feel free to reach out.

——————

P.S. — I highly recommend following Kelizheng Lizheng and Yage's Superlinear Academy. When we got into what "learning AI" should actually mean and how to approach it, they laid out a particularly strong framework.

🎬 Our video podcast is now live on @Koji Yang Yuancheng's channels on WeChat Video, Douyin, Xiaohongshu, Bilibili, YouTube, and other platforms.

📒 The transcript is published on the @CrossingCrossing WeChat official account.

🟢 🔴 🟡 Part 1 — Kelizheng Lizheng: How Do Power Users Use AI? How Should Ordinary People Learn AI? 🟢 01:40 What Separates Power Users from Everyone Else

What are the two key differences between power users and ordinary people?

Wanting to learn about world models, he didn't read papers — he had AI write him a 30,000-word office romance melodrama. Why did this work?

🟢 04:33 The Single Most Important Thing When Learning AI This Year

Why does he insist you "stop using ChatGPT immediately" and switch to Claude Code / Codex / Cursor?

When factories swapped steam engines for electric motors, efficiency only rose slightly — the real leap required one specific thing. What was it?

Using a chat box versus using agent tools: "a three-point gap."

🟢 10:56 Good Skills and Context Are the Real Moat

Same tools, same models — in the end, only one thing creates distance.

What makes a good skill, fundamentally?

Can skills be productized and sold?

The third step to using AI well — accumulating context and distilling context. Why are these two separate actions?

🟢 15:03 Crossing the River by Feeling the Stones — But the Stones Have Disappeared

— What exactly is the missing stone?

Looking from both demand and supply sides, why has the old startup logic suddenly stopped working?

His previous company could list a hundred reasons "you should buy us." Now customers chat with Codex for two minutes and leave.

🟢 22:01 AI Will Bring a Second Renaissance

Why is he so optimistic about his own children's future?

When AI thoroughly replaces the "cog in the machine," people will be forced to do only one thing.

Future education may no longer be about mastering one discipline, but "knowing a little" about a hundred disciplines.

"You thought learning AI meant learning nouns. Actually, you're learning verbs."

.

🟢 🔴 🟡 Part 2 — Koji's Share: Mid-Year Observations and Reflections on AI Startup Investing 🟢 25:49 The 2026 Startup Experience: The Other Side of the Bubble

The Paperboy AI founder's line that stuck with Koji: the optimal way for humans and AI to collaborate likely hasn't been invented yet.

"Software is dead," "pessimists are often right" — but why can only optimists actually succeed?

Unprecedented capital frenzy and obvious froth are two sides of the same coin.

🟢 27:54 Picking Up Pennies in Front of a Bulldozer

One chart shows: 90% of the business value created by this AI wave has been captured by model companies.

If picking up pennies in front of a bulldozer only yields pennies, what are the two remaining paths to survival? Why must one "leap for the sky" and the other "dig underground"?

AI for science — for what exactly?

Behind the billion-dollar valuations in embodied intelligence.

🟢 32:51 FDE: Bringing Digital Employees Into Companies

OpenAI and Anthropic both issued announcements around the same time, introducing the same new term.

What exactly is an FDE (Forward Deployed Engineer)?

This role has long existed (pre-sales, customer success, embedded engineer) — but what happened when it was renamed?

With such massive talent demand, why hasn't supply exploded yet?

🟢 43:30 What Does the New-Era VC Look Like?

Narrative connector, connector as a service, new-type financial partner — what specific founder problems do these three roles solve?

"Self-strengthening leads to universal strength" — why does a founder who needs too much VC guidance raise serious concerns?

A good investor should be like a co-pilot, not someone shouting about every traffic light.

🟢 50:00 Five Directions

Everything agent: sandbox, memory, communication, payments...

Designing products for agents: Lark's reputation surge and its CLI — what's the connection? When an agent can't find your tool, you "effectively don't exist."

Video models: Keling AI at $500 million, Seedance at $1.5 billion ARR, a wedding photographer in a small Yunnan town making a globally viral short film.

Software won't die, and GUIs still have a future.

Everything voice-related.

Subscribe to Crossing: 🚦 We track the industry shifts and new entrepreneurial opportunities brought by the new wave of AI technology.

🚦 Crossing is Steve Jobs' metaphor for Apple — standing at the intersection of technology and liberal arts, where great products are born. AI is transforming every industry. We seek out, interview, and bring together a new generation of AI founders and active participants in the AI era. Together with them, we explore and embrace the new changes and new possibilities.

👦🏻 Host Koji: I founded Crossing, launched AI Hacker House — a community space for the new generation of AI founders — and serve as Venture Partner at ZhenFund. I believe technology, especially AI, represents the greatest value-creation opportunity of our generation. Koji on Jike, Koji's website