Accumulated from 736 posts tagged #AIProductSharing / #BusinessIdea. Wang Kai has browsed thousands of AI projects; this wiki does not copy lists, but extracts project mining methodology + systemic opportunity patterns + cross-year case comparisons.
1. Project Mining Methodology
graph TD
A[发现机会的 4 个视角] --> B[1. 大模型新特性<br/>能力变化产生新场景]
A --> C[2. 大平台新动作<br/>Apple/Google/Microsoft 生态变]
A --> D[3. 流量红利窗口<br/>SEO/Chrome/Reddit/AgentMail]
A --> E[4. 已验证模式复用<br/>互联网时代失败案例 + AI 重做]
4 information sources Wang Kai uses long-term
- Overseas AI project navigation sites: topai.tools "Latest" channel
- YC project database: ycombinator.com/companies
- AI project traffic analysis: AITDK plugin
- Industry vertical media: The Information / Anthropic Economic Report
2. Systemic Opportunity Patterns (sorted by time)
2.1 2023 Dividend Patterns
| Pattern | Case | Status |
|---|---|---|
| Vertical RAG(finance/legal/technology) | Reportify / DevvAI / Humata | ✅ Ongoing |
| AI site network SEO | Cutout.pro / glarity.app | ⚠️ Declining (killed by Google AI Overview) |
| AI images(photorealistic/face swap/background removal) | Midjourney / Stable Diffusion | ⚠️ Eroded by multimodal models |
| AI writing | Writer / Jasper / Notion AI | ✅ Ongoing but ARPU dropping |
| Vertical GPT bots(shell wrappers built with Coze etc.) | Abundant | ⚠️ Eroded by large model upgrades |
2.2 2024 Dividend Patterns
| Pattern | Case | Status |
|---|---|---|
| AI video(text-to-video via prompts) | Runway / Pika | ✅ Ongoing |
| AI coding | Cursor / Devin / v0 | ✅ Explosive growth |
| AI legal / AI medical SCRIBE | Abridge / Light | ✅ Ongoing |
| AI Coding data calibration | DataSnipper tens of millions in revenue | ✅ Ongoing |
| Relaxed commercial value thresholds for Chrome extensions | glarity / ZenithTrade | ⚠️ Traffic is hard to get |
| AI overseas podcasts(evolved from NotebookLM) | Wonderr / Spotify Podcast AI | ✅ Ongoing |
| AI Bot services(subscribed personalized assistants) | Granola / Shelpful | ⚠️ Boundaries eroded by Apple/Google |
2.3 2025 Dividend Patterns
| Pattern | Case | Status |
|---|---|---|
| Agent platforms | Manus / Devin / Cursor | ⚠️ Cash-burning war |
| New paradigm AI search | Perplexity / Metaso | ⚠️ Eroded by ChatGPT/Claude |
| AI browsers | Fellou / Comet / Dia | ⚠️ Early stage |
| Vertical Skill / MCP platforms | TokRepo / mcp.so | ✅ Early-stage dividend |
| AI quantitative trading tools | TabTab / Trading-R1 | ⚠️ Most are misused |
| Direct AI video/voice generation | Veo / ElevenLabs / Jimeng | ✅ Tools ongoing |
2.4 2026 New Dividend Patterns
| Pattern | Case | Status |
|---|---|---|
| AI psychology / emotional counseling | Xinguang APP / Claude IFS | ✅ Early stage |
| AI smoking cessation / chronic disease management(built by Wang Kai) | Wang Kai AI Smoking Cessation | ✅ Launched |
| Agent automated promotion | Self-built Claude Code + Chrome MCP | ✅ Dividend |
| AI Eye glasses (replacing mobile interaction) | Apple Vision / Ray-Ban Meta | ⚠️ Early stage |
| AI investment tools(not stock trading, but decision support) | KeepRule (built by Wang Kai) | ⚠️ Early stage |
| AI personal principle library / decision framework | Built by Wang Kai | ✅ Early stage |
| Robotics / physical AI | Figure / Tesla Optimus | ⚠️ Still needs time |
3. 5 Criteria for Judging Systemic Opportunities
Wang Kai has repeatedly emphasized (this is not a simple list):
- Is a large model upgrade an enabler or a killer?？
- Enabler (e.g., Cursor + better Claude) → long-term dividend
- Killer (e.g., simple GPT wrappers) → short-term dividend
- Will big platforms proactively build this? (Apple/Google/Microsoft)
- They will proactively build it (e.g., photo albums, notes) → don't do it
- They are unwilling to build it (e.g., niche scenarios) → you can do it
- Can token costs be covered?
- Immediate payback (e.g., $30 per AI PPT session) → doable
- ARPU model (e.g., $5 monthly fee) → must have extremely high LTV
- How long is the traffic dividend window?
- Dividend window < 3 months → get in and out fast
- Dividend window > 1 year → you can build an ecosystem
- Is there a data / channel / relationship moat?
- Yes (e.g., SaaS holding customer data) → plus point
- No (e.g., pure wrapper) → no plus point
4. Recurring "Systemic Scenarios"
4.1 AI Replacing Shallow-Interaction Tools
- Key insight: All light-interaction scenarios of "open APP → input information → get result" will be replaced by Siri / system-level AI
- Recommendation: Interaction paths cannot be short; lock into deep scenarios
4.2 AI Scribe (Document Replacement Scenarios)
- Police duty reports (YC Abel Police)
- Doctor medical records (Abridge)
- Lawyer documents (Spellbook)
- Meeting notes (Granola / Feishu Minutes)
- Wang Kai's judgment: "Willingness to pay is high in Europe and America, labor costs are extremely high, and AI cannot fully replace humans within 3-5 years"
4.3 Data Calibration / Data Consistency
- Financial report data calibration (tens of millions in revenue)
- Legal document conflict detection
- Financial report data cross-validation (DataSnipper)
4.4 Monitoring + Instant Response (AI never sleeps)
- FastFacts (built by Wang Kai before): monitor SEC filings for instant trading
- Monitor market news and trending topic anomalies
- Monitor futures prices + AI interpretation
- Police body cameras + AI reports
4.5 AI + Cameras (Visual Input)
- Health/disease recognition
- Learning tutoring
- Product label recognition
- Game commentary (AI assistance while playing LoL/Texas hold'em/Mahjong)
4.6 RAG's "Vertical Data Goldmine" Logic
- Find data that is search-engine-unfriendly but has monetary value
- For example, financial report data (Reportify), earnings calls (Reportify), technical documentation (DevvAI)
- Upgraded prompts + reranked sources + continuously maintained Q&A sets = surpassing Metaso/Perplexity
4.7 Private-Domain AI Search / Personal Wiki
- Quanta Quest (built by Wang Kai before, lessons from failure)
- Granola
- Me.bot
- Core moat: Connecting private data from Notion / Gmail / Drive / iCloud, etc.
4.8 AI + Education
- Card-based learning (turn K12/English/professional content into test questions)
- AI replication of Andrew Ng's courses for SEO
- Student cheating vs. school anti-cheating
5. "Redoing with AI" for Internet-Era Projects
Wang Kai's epiphany (2025-03-17 post):
- Demands that failed / never took off in the mobile internet era
- Become valid again in the AI era in many cases
- Cases: Book of Stomach, Taobao affiliate-style cashback (heavily invested by YC), AI earthquake monitoring, AI psychological counseling
Method to capture such opportunities:
- Sort out demands that were popular/tried during the mobile internet cycle
- Use forgetting logic to judge how many people remember them now
- See if AI can solve the bottlenecks that "couldn't be built" before
- Rebuild with AI and test
6. "False Opportunities" Wang Kai Repeatedly Warns About
| False opportunity | Why |
|---|---|
| Simple GPT wrappers | Eroded by the next model version, no moat |
| AI quant / get-rich-quick stock trading | Quant giants can't afford the burn, retail investors can never outperform |
| AI digital immortality / AI soul resurrection | Moral boundaries + missing data |
| General-purpose Agent / Minus replication | Already a red ocean |
| MCP toolkits / Workflow platforms | Large model companies build them themselves |
| RAG middleware | Directly supported by large models |
| AI-era SaaS (per-seat charging) | Eroded by OS-ification by ChatGPT etc. |
7. Directions Wang Kai Repeatedly Emphasizes as "Must-Watch Long-Term"
graph TD
A[长期价值方向] --> B[1. AI + 老需求<br/>戒烟 / 减肥 / 冥想 / 心理]
A --> C[2. 垂直行业服务<br/>财务代理 / 海关 / 招投标]
A --> D[3. 软硬结合<br/>智能玩具 / AI 玩偶 / AR 眼镜]
A --> E[4. 超级个体赋能<br/>独立开发者工具]
A --> F[5. 精神消费<br/>个人成长 / 游戏 / 体验]
8. Key Classic Cases (mentioned multiple times)
8.1 Monica (Xiaohong)
- Started by acquiring ChatGPT for Google
- Evolved into an AI all-in-one suite + ad buying
- Core capability: Traffic acquisition + multi-product reuse
8.2 Reportify (Brother Dong, former Xueqiu CTO)
- Financial vertical RAG
- Financial reports / research reports / earnings calls
- Core moat: Tens of thousands of high-quality financial documents + reranking
8.3 v0 / Stitch / Lovart
- AI design / UI generation
- Key learning point: Ten-thousand-token prompts + daring to pressure the model + providing examples
8.4 Salesforce Agentforce 3
- Outcome-based payment (no longer per seat)
- Epoch-making significance: Fundamental change in business model
8.5 Glarity.app (3 million monthly traffic)
- Users use the plugin to summarize videos → automatically generate SEO landing pages
- But deep conversion rate is low,Large SEO traffic does not necessarily mean high commercial value
8.6 ISRG (Intuitive Surgical)
- Wang Kai actually bought and sold
- AI + surgical robots, data + channel moat
8.7 Granola (AI notes, YC + Silicon Valley hit)
- Takes over all "context + information-intensive" work
- Founder philosophy: AI provides key information + assists decision-making, rather than making automatic decisions
9. Resource Toolkit (recommended multiple times by Wang Kai)
| Tool | Use case |
|---|---|
| Coze.com | Prompt-wrapped bots |
| Dify | Workflows |
| AirOps | Bulk SEO content |
| Writer | Long-form AI writing |
| Undetectable.ai | Humanizing AI content (avoid detection) |
| GPTzero | AI detection |
| NotebookLM | Paper / podcast AI learning |
| Stitch (Google) | UI / APP design |
| Claude Artifacts | Simple product demos |
| Manus | General-purpose Agents |
| Fellou | AI browsers (for testing) |
| Jimeng / Pika / Runway | AI video |
| Ideogram | AI Logo |
| Napkin AI | Article illustrations |
| Notebooklm | Video/PDF summarization |
| Eightify | YT summarization (nearly 10 million monthly traffic) |
10. Wang Kai Says
"AI application-layer opportunities come in waves; every time large model capabilities improve, there will be a wave of opportunities."
"To seize opportunities, you can prepare your team and response logic in advance; the core of every opportunity window is to compete on speed."
"Agents such as Qwen and Manus are still in fierce competition, and opportunities around Agents are actually abundant."
"After browsing projects beyond a certain volume, you will know that 'ideas are worthless; execution + business model + traffic' are what matter."
"Don't build shallow AI services that die as soon as large models upgrade."
References
- Wang Kai's AI-Era Entrepreneurship Methodology and Business Judgment Framework - Entrepreneurship methodology (execution framework after the idea)
- Wang Kai's AI Industry Judgments and Big-Company Competition Landscape Observations - Industry judgments (how big-company moves affect idea feasibility)
- Wang Kai's Agent Engineering Practice - Agent engineering (opportunities around Agents)
- Wang Kai's Overseas Promotion SOP and Overseas Multi-Account Matrix Practice - Overseas promotion (traffic acquisition after idea launch)
- newtype AI Business, Industry and Investment Observations - Lao Huang newtype idea comparison