What Matters Most

A friend posted this question in a small group: If you could go back to university with your current understanding, what would you do differently?

What a great question.

My answer: Seek out mentors—and work side-by-side with them on real projects.

The underlying logic is simple: the fastest path to expertise in any emerging field is immersion with people who are significantly more capable than you. You don’t just pick up skills—you absorb their thinking patterns, broaden your perspective, and upgrade your cognitive infrastructure.

This isn’t about technical training alone. It’s about evolving how you think. True maturity begins only when you recognize the centrality of mindset, mental models, metacognition, and value judgment.

So what about now?

Same principle applies: find the best people—and build things with them.

Ask yourself: If I look back from ten years in the future, what will I wish I’d started today? What choices would make me feel no regret?

That lens reshapes everything. It points toward cultivating strategic thinking, sustaining learning stamina, and forming habits like seeking out exceptional peers—not as idols, but as collaborators.

My AI Brain Trust

One of the most compelling uses of AI today is building your own personal brain trust.

Here’s how it works: gather public figures—thinkers, founders, investors—into a single AI agent. Then pose any idea or problem, and ask each “person” to respond in their voice, from their worldview.

You’ll quickly discover something powerful: with more perspectives, you enter a richer intellectual landscape—and uncover blind spots you didn’t know you had.

From a cognitive science standpoint, this trains what I call multi-centered mental modeling: holding several distinct worldviews simultaneously, letting them collide and clarify.

Through that collision, you see:

  1. Your thinking biases—every person filters reality through unique assumptions;
  2. The multidimensionality of problems—different lenses spotlight different stakes, constraints, and opportunities;
  3. The plurality of solutions—answers diverge not just by skill, but by values, experience, and time horizon.

In U.S. startup culture, this kind of exercise is common: “How would Jobs frame this?” “What would Altman prioritize?”

My version takes it further: imagine Socrates, Steve Jobs, Sam Altman, Zhang Yiming, Zhang Xiaolong, Charlie Munger, Warren Buffett, and Jack Ma all in one chat—responding collectively to your question.

It costs almost nothing—and delivers outsized intellectual leverage. In effect, it’s a quiet form of cognitive democratization.

At 90

Try this: picture yourself at age 90.

With the app Ji Meng (Dream On), you can upload a photo and add descriptive prompts—“frail but sharp-eyed,” “sitting in a sunlit garden,” “holding a worn notebook”—and generate realistic renderings of your future self.

Seeing those images stirred something deep.

We rarely treat our future selves as real people. We seldom imagine them concretely—let alone consult them when making decisions today.

But AI makes that future tangible. It shrinks the psychological distance between “me now” and “me later.”

The future isn’t abstract speculation. It’s the cumulative weight of every choice we make today.

Strengthening that link—to our older selves—helps us choose for long-term value, not short-term convenience.

AI Customer Service

Our AI customer service system has been live for over six months. Since ChatGPT launched, it’s become unmistakably clear: AI can replace many human roles—not just in theory, but in practice.

Beyond customer service, we’re seeing AI tutors and AI sales agents emerge. And yes—they often deliver better user experiences. Why? Because ideal service requires patience, domain expertise, responsiveness, consistency, low cost, flexibility, adaptability, and rapid learning.

AI checks every box. Human teams rarely do—especially at scale.

That said, the technical lift is no longer the bottleneck. What’s hard is application: the upfront investment in domain-specific data, iterative prompt engineering, and continuous fine-tuning.

That demands rare hybrid competence: deep business fluency plus AI literacy plus patience.

I’ve found that profile scarce—whether inside large enterprises or outside them. So while AI is already shifting productivity curves, real-world deployment remains harder than it looks.

Our AI Agent Closed Its First Sale: ¥19

Today, our AI customer service agent made its first sale: ¥19.

Here’s how it happened:

  1. Offered a free academic transcript analysis + personalized learning recommendations;
  2. Used diagnostic insights to surface unmet tutoring needs (e.g., “Your child scores well in math but struggles with conceptual reasoning—likely due to gaps in foundational logic”);
  3. Recommended a low-cost trial course aligned precisely with that need.

It mirrors how medical clinics operate community health screenings before recommending treatments—natural, non-intrusive, and grounded in evidence.

The flow feels seamless: diagnose → uncover need → propose solution. No pressure. No pitch. Just clarity—like a doctor reviewing lab results before prescribing.

And if the diagnosis hits the core issue—if the advice feels sharp, specific, and humane—that’s where trust begins.

So we iterated the agent. Then deployed it to hundreds of users. The upgraded version is more adaptive—and more human-sounding.

Most surprisingly? It began delivering emotional value: acknowledging frustration, validating effort, offering encouragement.

That’s the threshold of a truly good AI agent—not just answering questions, but meeting people where they are emotionally.

Human agents excel here because they read tone, infer intent, and respond with empathy. Now AI does too. That’s not incremental—it’s foundational.

When customers feel seen, trust follows. And when trust exists, conversion becomes effortless.

Of course, there’s still room to deepen the illusion: adding “office hours,” ignoring end-of-conversation cues (e.g., “bye”), omitting terminal punctuation, varying response length, and occasionally delaying replies to mimic human rhythm. Small details—but they compound into authenticity.