Wang Kai comes from a product manager background and had no programming knowledge at all. He started with Cursor in 2025-05 and fully migrated to the Claude Code terminal workflow by 2026. His summary has reference value for all practitioners from non-technical backgrounds:AI Coding is not for writing code; it is a foundational collaboration capability in the AI era.
1. Core Mindsets
| # | Mindset | Explanation |
|---|---|---|
| 1 | AI Coding is a foundational skill, not for writing code | It is used for data analysis, automation, bots, MCP, and product demos; the core is "extending capabilities, not reducing costs and increasing efficiency" |
| 2 | Pursue consuming more tokens | Do not save money; a monthly fee of 200/250 USD is the investment with the highest ROI |
| 3 | Data analysis is the most disruptive use case | Cursor + API pulls data in one hour; in the past, large companies took 3 days |
| 4 | Hands are faster than the brain | Do it as soon as you think of it, judge after the Demo is out; the cost of validation in the AI era is extremely low |
| 5 | Do not learn, only use | Do not spend time learning technical principles; start building directly, and ask AI when you get stuck |
| 6 | Three communication styles | PUA / honest partnership / detailed reporting with materials; you also need to keep asking AI during coding |
2. Evolution Path (Wang Kai's personal experience)
graph LR
A[2024 前: 不懂编程<br/>需要技术配合] --> B[2025-05 Cursor 入门<br/>5 万美金月费起步]
B --> C[2025-06 用 Cursor 做 RSI 策略<br/>金融数据分析]
C --> D[2025-09 Claude Code 客户端<br/>开始接 MCP]
D --> E[2025-12 Claude Code 终端<br/>主力使用]
E --> F[2026-01 250 美金/月 Claude<br/>多窗口并行]
F --> G[2026-03 Chrome MCP + Loop<br/>AI 自动化任务体系]
G --> H[2026-04 多浏览器多 Agent<br/>token 消耗百倍]
3. Evolution of Cursor Usage
Stage 1: Building small tools
- Simple bots (Feishu bot sending stock data)
- API call wrappers
- MCP integration
Stage 2: Data analysis (disruptive)
- Running ten years of historical data to validate strategy hypotheses
- "Largest market cap company" strategy vs Nasdaq comparison → data proves the strategy underperforms the index
- 30-year annualized comparison of Moutai / Apple / S&P 500 / Berkshire
- Signal statistics (average number of signals per month, average holding days)
Key turning point: redefining Cursor from a "code tool" to a "sharp tool for data analysis"
- Work that previously required 3–5 days from technical teams
- Now takes 1 hour to pull data + generate charts + draw conclusions
Stage 3: Product Demos
- Self-built KeepRule (investment principle library)
- Buffett-wiki Demo
- owner-earnings intrinsic value calculator
- Within 24h → high-quality Demo + backend API + frontend interaction
4. Claude Code Terminal Workflow (main tool in 2026)
graph TB
A[一台电脑] --> B[终端窗口 1: Claude Code 主对话<br/>策略复盘 / 文档迭代]
A --> C[终端窗口 2: Claude Code 数据分析<br/>跑历史数据 / 出报表]
A --> D[终端窗口 3: Claude Code 控制浏览器<br/>自动化推广 / 注册账号]
A --> E[终端窗口 4: Claude Code Loop<br/>定时任务 / 视频生成]
A --> F[终端窗口 5: 与 RSI AI 策略攻防演练<br/>找漏洞 / 优化提示词]
A --> G[Xcode: 看 APP 效果<br/>对着 Demo 改细节]
Direction to maximize Claude Code's value(inspired by Anthropic's one-person marketing blog):
- It is not about clever tricks or gimmicks
- It is about "making all data/information generated in the workflow accessible to AI"
- The more complex the work and the more data there is, the greater the help from AI
5. What AI Coding Has Changed
5.1 Changing collaboration processes (the most profound change)
Internet era:
- Product writes requirements → design draws mockups → tech develops → joint debugging and testing → launch
- Cycle of several weeks
AI Coding era:
- Product builds Demo themselves (including backend API)
- Tech reviews + optimizes (runs it locally)
- Product details are modified directly on the Demo
- Cycle of 24–48 hours
Wang Kai's summary: previously thinking "AI lets me do things faster" was about cost reduction and efficiency increase; AI Coding is capability extension— things you could not do before, you can do now.
5.2 Changing the way information is discerned
Example(2026-02 post):
- Saw the investment strategy of "only buying the highest market cap US stocks"
- Before: first reaction was "very smart"
- Now: use TRAE + FMP API to pull 10 years of data for validation in 1 hour → the strategy underperforms the Nasdaq
- Conclusion:AI Coding gives ordinary people the ability to refute with data
5.3 Changing work habits
- Previously had the browser open for 7/8 hours
- Now mainly uses the Claude Code terminal, browser is only for viewing information
- Browser operations are all done through Claude Code (Chrome MCP)
6. Key Toolchain
| Tool | Use case | Notes |
|---|---|---|
| Claude Code: 250 USD/month | Main tool (2026) | Must-buy, token quota greatly expanded |
| Cursor | Cursor: early main tool → backup | Writing code, product demos |
| TRAE (ByteDance's overseas version) | Data analysis | Cheap, can use Claude API |
| Qoder (Alibaba) | Backup | Chinese-language scenarios |
| Codex | ChatGPT: backup | OpenAI's |
| Chrome MCP (official) | Controlling the browser | Must switch to "real browser + session reuse" |
| AgentMail | Agent-dedicated email (YC S25) | Essential for registering accounts |
| FMP API | Financial data MCP | Monthly fee ~100 USD, self-built MCP |
7. "Stuck Points and Breakthroughs" Wang Kai Experienced Personally
Stuck point 1: GPT-guided Python installation went in circles (2025-06-04)
- At first Wang Kai asked GPT to guide him through installing the Python environment
- "AI was very responsible and tried various solutions, but it didn't get installed after 1 hour"
- Lesson: in fields where you have zero judgment, AI will lead you in circles
- Solution: go directly to the Python official website to download the installer, done in 1 minute
Stuck point 2: Claude API calling Chrome MCP hit IP restrictions
- A global VPN/proxy didn't work either; the Claude client was usable but the API was restricted
- Solution: control the browser directly through Claude Code to bypass the API restrictions
Stuck point 3: New HN account was strictly restricted (2026-03-11)
- Wang Kai had Claude Code autonomously register an HN account + post comments
- The pipeline worked: email registration → verification code (Chrome MCP clicking iframe) → account registration → comment
- But posting links was rejected (HN anti-spam)
- Lesson: platforms treat humans and Agents the same; you need to warm up accounts
Stuck point 4: When Coding, didn't know how to PUA AI
- Using Cursor to build a trading strategy, didn't know how to make it output better results
- Solution: ask the Claude client "how to make Cursor give better results?"
- "AI PUA AI closed loop"
8. "Automated Tasks" Mindset for AI Coding (2026-04+)
Wang Kai's epiphany:Use Claude Code's Loop / terminal resident capability to achieve automated operation
graph LR
A[设定自动化任务<br/>e.g. 监控股价、爬资料、发推广] --> B[Agent 运行]
B --> C[10 分钟检查一次<br/>如果停止就让其继续]
C --> B
B --> D[每日生成报告]
D --> E[飞书 bot 通知]
Bandwidth expansion principle:
- Traditional management: 1 person leads 10 people
- AI Coding era: 1 person covers hundreds of things (through automated Agents)
- "Build a stably running system, rather than doing the work directly"
9. Must-read Materials (Recommended by Wang Kai)
- Anthropic 1-person marketing blog → see the extreme use of Claude Code in marketing
- Claude official Skills documentation
- Context engineering papers (2025-07 / 2025-10 LangChain x Manus)
- Karpathy LLM Wiki concepts
- Microsoft Agent ten-lesson course (recommended by Teacher Baoyu)
10. Three-Stage Mindsets for AI Coding
| Stage | Mindset | Tools |
|---|---|---|
| Beginner | Learn the three communication styles (PUA, partnership, reporting) | Cursor / Claude Code client |
| Proficient | Use AI Coding for data analysis, bots, product demos | + MCP / Claude Code terminal |
| Master | Automated task system / Agent clusters / multi-window parallel execution | + Chrome MCP / Loop / multiple browsers |
11. Advice for People from Non-Technical Backgrounds
Wang Kai emphasized (in multiple posts from 2025-05 onward):
- Don't learn, start building
- Directly spend 250 USD/month, use the most expensive option
- The goal is more token consumption, and discover use cases from it
- When stuck, immediately ask AI (Claude client) "how to make Cursor / Claude Code work better"
- Don't grind on small projects, quickly move on to the next scenario
- Hands are faster than the brain
12. Wang Kai Says
"AI Coding is definitely one of the foundational, most disruptive capabilities in the AI era; everyone must, must, must start using it."
"Spending another 250 USD per month is cheaper than missing one decision."
"Most tokens in AI Coding are definitely not spent on writing product code."
"A sharp tool for data analysis — that is simply Cursor's greatest value."
"The most valuable capability in the AI era is 'the efficiency and quality of exponentially consuming tokens.'"
References
- Wang Kai RSI AI Investment Strategy and the "Good Asset, Good Price, Good Volatility" Framework - Investment strategy (Claude Code is the main tool for data analysis)
- Wang Kai Agent Engineering Practice - Agent engineering (automated tasks, multi-Agent collaboration)
- Wang Kai's AI-Era Entrepreneurship Methodology and Business Judgment Framework - Entrepreneurship methodology (engineering capability for independent development)
- Wang Kai's AI-Era Decision-Making System - Decision-making system (AI Coding outputs assist decision-making)
- newtype AI Business, Industry and Investment Observations - Comparison from Lao Huang's newtype perspective