This week, I attended an offline Microsoft AI sharing event hosted by Hongli Society. The first three sessions—focused on AI—captured key insights from three speakers.

How to Leverage AI

  1. Those who understand how to leverage AI—not just use it—will capture disproportionate value in the AI-internet era.
  2. What is a “bonus” (bonus/opportunity)? One bonus keeps you alive; two make a viable business; three can spawn a unicorn; four may take you public.
  3. Today’s most common AI applications: cutting costs, boosting efficiency, driving traffic, and building IP.
  4. Consider a typical scriptwriting workflow: editor finds articles online → writes drafts → planning director screens them → IP records final versions. Each editor produces 4–5 scripts per day. Automating just 1–2 steps with AI lifts output 5–10×—without sacrificing quality.
  5. AI implementation follows three steps: real problem → right tool → measurable result.
  6. Key to AI-driven efficiency: solve one concrete bottleneck in a specific workflow—and lift speed 5–8×.
  7. Low private-domain conversion? Even basic tasks like consistent WeChat Moments updates can be reliably automated with AI.
  8. Prompt engineering isn’t magic—it’s human experience layered onto AI rules.
  9. Three prerequisites for AI real-world adoption: start from genuine needs; train AI like an intern (iteratively, with feedback); never adopt AI just to say you did.

What Can AI Actually Do?

It’s misleading to ask “What can AI do?”—better to ask: What does it solve for us, right now?

AI-powered writing workflows include: extracting core themes, pulling standout quotes, and rewriting in new voices.

Three tiers of AI application:

  1. “Anyone can do it”: e.g., Feishu’s AI meeting notes—simple, plug-and-play tools that will multiply rapidly.
  2. Prompt engineering: Teaching AI how to work—not just phrasing prompts, but deeply understanding your product, user needs, and context. This skill can’t be trained; it’s earned through practice. Structured prompt templates rarely move the needle.
  3. Private-data fine-tuning: Only attempt this once you truly grasp both your data and your goals.

Three paths for enterprise AI adoption:

  1. Aggregating individual AI tools into team-wide practices.
  2. Adopting purpose-built SaaS solutions.
  3. Deconstructing your own workflows—and building custom, proprietary AI tools. This is where AI delivers lasting competitive advantage.

Future trends in AI business:

  1. Entering an era of performance surplus: Chinese large models matching GPT-4 isn’t a question of if, but when.
  2. Commercial AI usage will get pricier—token costs will rise.
  3. AI will crystallize into a distinct intellectual industry, with its own talent, IP, and economics.

Three recommendations for founders & leaders:

  1. Identify real pain points in your product or operations—and deploy AI for measurable cost reduction or efficiency gain.
  2. Avoid “AI campaigns.” Go deep, not wide. Pick one high-impact tool—and master it.
  3. Use it yourself. Firsthand experience beats all theory.

Bilibili & AI

  1. CapCut (JianYing) has massive untapped potential. DouBao now holds China’s highest monthly active users among domestic AI apps.
  2. On Bilibili, watch time increasingly drives organic reach—the longer viewers stay, the more the algorithm recommends. Prioritize time contribution, not just views.
  3. What makes a great short video? It explores knowledge gaps, delivers joy, or expresses emotion and values authentically.
  4. The “impossible triangle” of content: originality + high quality + high frequency—you can only optimize two at once.
  5. The “Golden Half-Second” workflow:
     1. Topic + internet intuition (what’s resonating now),
     2. Study top-performing covers on external platforms for visual cues,
     3. Refine the topic,
     4. Use AI to rework visuals, pacing, and hooks,
     5. Add polished motion graphics—win the first frame.
  6. AI dramatically accelerates editing—especially rough-cut assembly, subtitle generation, and asset tagging.
  7. Track impression-to-play ratio: For platforms, every impression is monetizable. Optimize for play-through, not just clicks.
  8. Study YouTube closely—its cover logic, thumbnail psychology, and recommendation signals align surprisingly well with Bilibili’s.
  9. Team scaling rule: You can add unlimited editors to boost output—but keep scriptwriters per account under three. More writers dilute voice and consistency.
  10. Build your own libraries: cover templates, topic banks, and visual asset repositories.
  11. Knowledge-video workflow:
     1. Gather sources using Consensus (a powerful research assistant),
     2. Ask targeted questions of documents to extract insights,
     3. Human editors refine and write the final script.
  12. Global verticals with huge local potential: garden design, cycling culture, home brewing—underserved but growing.

The Traffic Trap

An online K–12 enrichment institution has been stuck at ~¥30M annual revenue for two years—unable to break past its traffic ceiling.

Growth stalled. Profitability remained elusive.

The founder reached out for advice. We uncovered several issues—but the core problem wasn’t traffic volume. It was low operational efficiency and no defensible differentiation in a crowded market.

Their acquisition model relied almost entirely on fragmented private-domain partnerships—no sustained investment in public-domain channels (e.g., search, social discovery, platform algorithms). That’s a major strategic omission.

Without advantages in business model, team execution, or brand strength, the traffic they do acquire is inevitably marginal—“leftover” inventory from low-tier partners. Their current channels confirm this: tiny volumes, “one-fish-many-meals” dynamics, even some outright “coupon-hunting” platforms. This growth engine is structurally unsustainable.

Key conclusions:

  1. Traffic isn’t the root problem—it’s a symptom. What’s missing is traffic intelligence: the ability to attract, convert, and retain based on real insight and capability.
  2. When competitiveness is weak, traffic tactics must innovate—or borrow leverage. Copy-pasting old playbooks won’t work.
  3. Public-domain traffic isn’t just “another channel.” It’s scale infrastructure. Even a 0.1% share of Bilibili’s or Xiaohongshu’s daily active users dwarfs what any private partner can deliver. In practice, public traffic is functionally infinite—if you know how to earn it.

The Craft of Writing

  1. All outward communication requires writing skill: books, articles, ads, product copy, proposals—even internal reports.
  2. Writer’s block often stems from not having thought clearly enough yet.
  3. Resist the urge to write prematurely. Wait until your research is complete and your structure is clear—then write in one focused burst.
  4. For important work—books, whitepapers, keynote scripts—this discipline pays off. A solid mental framework + full source material = exhilarating flow.
  5. Inspiration doesn’t strike at your desk. Walk outside. Observe people. Talk to strangers. Read widely. Capture sparks immediately—build your personal “idea bank.”
  6. Purpose first: writing exists to convey ideas or drive action. Avoid clever wordplay that obscures meaning.
  7. Double meanings? Use sparingly. Most “witty” puns confuse more than delight.
  8. Write like you speak—with plain, familiar language. The closer your tone matches your audience’s daily thinking, the better it lands.
  9. Always write to someone. Picture one real person—give them a name, a job, a frustration—and draft as if speaking directly to them.
  10. Logic is non-negotiable—even when hidden. Your argument must hold together, even if readers never see the scaffolding.
  11. Prefer linear logic over parallel lists: guide readers step-by-step—through questions, stories, or cause-effect chains.

How to Ease Procrastination

One root cause: fear of the unknown.

A frequent consequence: chronic anxiety.

Yes—distracting yourself (scrolling, tidying, “researching”) temporarily dulls the anxiety. But it leaves the fear intact—and the task untouched.

A more effective approach: shrink the unknown.

Turn ambiguity into familiarity—step by step.

Some things are hard to demystify (e.g., publishing in top academic journals when standards are opaque and your expertise feels insufficient). But even then: read acceptance letters, study rejected papers, interview editors, collect rubrics. Let your brain build neural pathways around the task. The more familiar it becomes, the less your amygdala rebels.

Another culprit: perfectionism. It doesn’t raise standards—it expands the unknown. “Perfect” is undefined, unbounded, paralyzing.

So: kill perfectionism. Embrace “60-point delivery”—get it done, then iterate. Done is the foundation of good. Good is the gateway to great.

Can we eliminate procrastination entirely? Unlikely. As long as we face tasks beyond our current capacity—and as long as uncertainty and perfectionist reflexes remain human—we’ll meet procrastination again and again. The goal isn’t eradication. It’s reduction through clarity, courage, and compassion for our own limits.