The Right Path

I highly recommend the book The Right Path: 26 Years of Reflections at Yum China.

The author is Su Jingxu—a legendary business leader who led Yum China for 26 years. This management classic captures how he and his team navigated the immense opportunities brought by China’s reform and opening-up—and how they continuously adapted to seize evolving advantages.

Its core idea is deceptively simple yet profound: Greatness lies in simplicity, sustained consistently.

Here’s what that means in practice:

  1. The integration of “Dao” (principle) and “Shu” (method): Values drive direction; methodology enables execution.
  2. Success isn’t about tricks or shortcuts—it’s the result of long-term commitment to the right path.
  3. Decision-making is paramount: High-impact success comes from accumulating both high-quality and high-volume decisions. “Dao” is your compass—integrity, excellence, people-first values. “Shu” is your toolkit—decision models, know-how, org design, time management.
  4. Know-how is the core asset of any team or company. It’s not just experience—it’s reusable, transferable, decision-guiding knowledge. Continuously build it, then use it to make better calls.
  5. To improve decision quality, cultivate disciplined habits: Clarify the situation → define the objective → generate multiple options → anticipate consequences → choose the best path.
  6. Management, at its heart, is “managing yourself, aligning others.” Forceful control rarely works. True effectiveness lies in enabling others to succeed—and thereby succeeding yourself. As the ancient Su Shu says: “Virtue is what people gain—allowing all things to fulfill their true desires.” Only now am I beginning to grasp the depth of such wisdom.
  7. The core of organizational culture is simple but hard: “First, identify what’s right. Then, do it right.”
  8. Long-term thinking means mastering the small things first—only then do strategy and innovation become meaningful.
  9. Time management: Keep your to-do list ruthlessly short—and clear it fast.
  10. Advice for young professionals (which I fully endorse): When choosing your first job, prioritize skill-building over salary. What you earn in your first few years pales beside what you’ll earn across your lifetime—if you invest wisely in capability early on.

Blind Spots in Building AI Products

This week, I spoke with several top-tier experts from traditional industries—founders of elite consulting firms, CEOs of leading PR agencies, nationally recognized lawyers. In their domains, they’re undisputed authorities.

Their insights often challenged my assumptions—sometimes overturning what I’d taken for granted.

But what struck me even more was a shared trait: nearly all of them harbor an “AI product dream.”

Underlying this is a reasonable belief—that AI can fundamentally disrupt and reshape workflows in their fields.

That mix of anxiety and excitement? It’s widespread among domain experts riding the AI wave. Logically sound—but translating that into actual products? Not so straightforward.

When I asked them to describe their AI product idea, most couldn’t articulate it in one clear sentence. They’d share scattered thoughts, promising angles, clever features—but no crisp definition of what the product actually is, or how it solves a specific problem in one sentence.

Why is it so hard—even for world-class experts—to define a product in their own field?

I think it cuts across three layers:

  1. Mindset mismatch: They’re trained in service logic—focused on what to do, not how to productize it. Product thinking starts with how: concrete user scenarios, defined workflows, measurable outcomes.
  2. Cognitive blind spot: Deep expertise breeds familiarity—and familiarity blinds. It’s harder to step outside your own lens and see needs through the user’s eyes.
  3. Capability gap: Moving from expert practitioner to product builder demands new muscles—user journey mapping, interaction design, business model validation. These aren’t part of their native toolkit.

Prompt Services

A friend asked: Are there professional prompt-engineering companies in China?

I replied: Not really. Most work is done by individuals—and no sustainable business model has emerged yet.

Why?

  1. Writing good prompts ≠ understanding the business. To craft truly valuable prompts, you must first deeply grasp the domain’s workflows, standards, and mental models. That upfront discovery work consumes most of the time—not the prompt writing itself.
  2. Generic prompts have little commercial value. Valuable ones? Often sit beyond the budget of typical buyers.
  3. Truly enterprise-grade prompts require continuous iteration. There’s no “final version”—only increasingly demanding versions, shaped by real usage and shifting goals.
  4. As models advance, prompt requirements evolve too—especially for abstract, reasoning-heavy tasks. That’s a different kind of skill entirely.
  5. In high-value AI use cases—like AI agents—prompts alone are insufficient. You need tool integrations, data pipelines, and orchestration logic to deliver real solutions.
  6. So, to build commercially viable, paid-for prompts, you need rare cross-disciplinary fluency: deep domain knowledge + mastery of standards + abstract thinking + prompt craft + AI tooling + product architecture. And even then, weeks of immersion may be needed just to map the logic.

As a standalone service, prompt engineering doesn’t scale well.

But productized around prompts? Yes—much stronger:

  • Vertical-specific prompt libraries + embedded consulting
  • Prompt tooling platforms + expert support tiers
  • Prompt training programs + certification pathways

How to Price AI Products

Over the past two weeks—driven by a new AI project—I’ve been diving deep into AI pricing logic.

One standout resource came from Madhavan, who’s helped over 250 companies (including 30 unicorns) refine pricing and monetization. Author of Monetizing Innovation, he recently released Scaling Innovation. His insights were sharp and actionable.

Key takeaways:

  1. Don’t fixate on just market share—or just wallet share. Many companies fall into the “single-engine trap”: either chasing growth at all costs, or squeezing revenue before proving value. Both backfire.
  2. AI pricing ≠ SaaS pricing. With SaaS, you could iterate slowly on pricing. With AI, you must get it right from day one. Example: An AI IDE priced at $20/month trains users to believe that’s what AI is worth. Later price hikes face near-total resistance. Madhavan warns some tools may already be “priced into oblivion.”
  3. His 2×2 pricing matrix is powerful:
    • X-axis: Attribution (Can you quantify the value AI delivers?)
    • Y-axis: Autonomy (Can AI complete the task independently?)
      Highest value? High attribution + high autonomy—e.g., Intercom’s Fin, which resolves customer queries autonomously at $0.99 per resolution. Such products can capture 25–50% of the value they create—versus 10–20% for traditional SaaS. For the first time, AI lets you measure ROI—not just guess.
  4. Simplicity sells. Madhavan champions “beautiful simplicity.” Can customers instantly grasp your pricing logic?
    • Superhuman: $1/day = 4 hours saved
    • Subway: $5 footlong
      Pricing is your story. Tell it clearly—and customers feel the value. Overcomplicate it, and they hesitate, compare, delay. Simplify it—and they buy faster, renew more readily.
  5. In negotiations, never give without getting. Every discount should come with reciprocity—e.g., a commitment to quarterly value reviews or internal advocacy. That builds stickiness and strengthens your leverage.
  6. POCs shouldn’t prove “it works”—they should co-build ROI models. Charge for POCs. It filters serious buyers and prevents free-riding.
  7. Apply the 20/80 rule rigorously: 20% of your features drive 80% of willingness-to-pay—and those are usually the easiest to build. Yet many teams give them away for free in MVPs. Madhavan suggests renaming MVP to MVP = Most Valuable Product.
  8. Review pricing frequently—but don’t flip models lightly. SaaS adjusted pricing every 2 years. AI companies may need to reassess every 6 months. But changing your pricing model (e.g., from per-seat to per-outcome) only makes sense if your product’s attribution or autonomy has genuinely shifted. Otherwise, confusion and churn follow.

This week, I used Trickle.so extensively to prototype lightweight tools—and the experience was outstanding.

With natural language alone, you can turn ideas into functional, deployable apps in minutes. Its built-in database and auto-deployment remove major friction points common in similar tools.

It exemplifies what I believe is the right path for AI-native applications: focus relentlessly on one core scenario—and execute it at world-class level.

That singular strength shapes user perception, drives word-of-mouth, and anchors your brand.

A strong product leaves users saying, “This one thing? It’s unbeatable—and that’s why I tell others about it.”

Take our own education tool: Teachers often use it alongside competitors—even though feature overlap is high. When I asked why, several said: “Your grade management is simply the best.”

Same principle applies broadly to AI tools today: They all do many things—but each has one thing it does notably better.