When to open: You want to decide which tool to use for AI Coding (choosing among Cursor / Claude Code / Codex / Cline / OpenCode), understand the evolution of the entire track from 2024 to 2026, or choose between "domestic vs overseas" tool stacks. This wiki does not repeat Claude Code Workflow (Cross-Author Thematic Synthesis) details of Claude Code, but instead focuses on "the evolution path of the entire AI Coding tool stack + division of labor philosophy + 4 code retrieval methods + the 0-80 vs 80-100 life-or-death line".
One-sentence core:AI Coding is not "knowing how to write code", it is "a foundational collaboration capability in the AI era" —— consensus of the three schools: Wang Kai / Lao Huang / Xiaohongshu.The main battlefield shifted 3 times in 18 months(Cursor 2024 → Claude Code 2025-H2 → multi-CLI Agent 2026).
0. What is "AI Coding" from 4 perspectives
| Perspective | Source | Core positioning | One sentence |
|---|---|---|---|
| Tool stack observation school | newtype · AI Coding Practice and Tool Stack Evolution / newtype tool stack: Oh My OpenCode customization + newtype CLI multi-Agent team(Huang Yihe) | Evolution timeline of the AI Coding tool stack | "Keep an eye on the best: Cursor → Claude Code → OpenCode" |
| Product manager disruption school | Wang Kai's AI Coding Practice: From Cursor to Claude Code Terminal Workflow / Wang Kai's Claude Code Terminal Workflow Methodology (Full Version)(Wang Kai) | AI Coding is a foundational collaboration capability, not for writing code | "Data analysis is the most disruptive / token consumption is a hundredfold / collaboration process is restructured" |
| Non-technical independent developer school | Xiaohongshu / Vibecoding Practice Collection (Games/Tools/Mini Programs) / Xiaohongshu / AI Tools + Skill Tutorial Collection / newtype · Vibe Coding Independent Product Practice | Vibe Coding = You don't write code but your depth of thinking cannot drop | Three core mindsets (state requirements clearly / progress step by step / be a picky client) + 0-80 vs 80-100 life-or-death line |
| Domestic no-scientific-internet school | dontbesilent AI Self-Media Course / 13 Lessons Systematic Methodology(Lao D) | Openrouter Key + AI command line + Don't ask again | "Claude Code is the real Agent; most domestic shell wrappers are not" |
Key observation:These 4 schools look at the same set of tools (Cursor / Claude Code / Codex / Cline) but use them completely differently: Lao Huang looks at "tool stack evolution", Wang Kai looks at "collaboration process restructuring", Xiaohongshu looks at "non-technical people becoming chief directors", Lao D looks at "domestic adaptation".
1. Tool stack evolution timeline (cross-author synthesis)
graph LR
A["2024-08~09<br/>Cursor 震撼<br/>(老黄 / 零代码 Chrome 插件)"] --> B["2024-09<br/>v0+Claude+Cursor 三件套<br/>(老黄 newtype.pro 上线)"]
B --> C["2025-02<br/>AI Coding 赛道全景<br/>DeepSeek/Claude 3.7"]
C --> D["2025-03<br/>Cline 自建 MCP<br/>(老黄 Obsidian MCP)"]
D --> E["2025-04<br/>Gemini 2.5 Pro<br/>大上下文扫码"]
E --> F["2025-05<br/>王凯 Cursor 入门<br/>5 万美金月费"]
F --> G["2025-05~07<br/>Claude Code 走入主流<br/>Plan Mode / Ultra think 32K"]
G --> H["2025-08~12<br/>Claude Code 主战场<br/>Sub-agents/Slash/Skills"]
H --> I["2026-01+<br/>OpenCode/Codex/newtype-os<br/>多 Agent 编排"]
I --> J["2026-04~<br/>多浏览器多 Agent<br/>Chrome MCP + 飞书 CLI"]
3 main battlefield shifts:
- 2024-08: Cursor shock (Lao Huang "stronger than ChatGPT")
- 2025-07: Claude Code enters the mainstream ("showed me the possibility of a relatively general-purpose Agent")
- 2026-01: Multi-CLI Agent orchestration (Lao Huang newtype-os + OpenCode / Wang Kai multi-browser matrix)
See newtype · AI Coding Practice and Tool Stack Evolution §0 for the full timeline.
2. Cross-author consensus (6 points)
2.1 "AI Coding is a foundational collaboration capability, not for writing code"
Wang Kai(Wang Kai's AI Coding Practice: From Cursor to Claude Code Terminal Workflow §1):
"AI Coding is a foundational skill, not for writing code —— it is used for data analysis, automation, bots, MCP, product demos; the core is'extending capability, not reducing costs and increasing efficiency'. "
Wang Kai's personal experience: Came from a product manager background with zero programming knowledge → used Cursor to build an RSI investment strategy in 2025-05 → pulled 10 years of data in one hour using Cursor + FMP API,a workload that used to take big companies 3 days.
Lao Huang corroborates(newtype · AI Coding Practice and Tool Stack Evolution §10):
"You must have programming thinking(you need to understand concepts like variables / functions / classes / asynchrony / API calls / databases),you don't need programming experience(you don't need to memorize specific syntax)."
Xiaohongshu corroborates(Xiaohongshu / Vibecoding Practice Collection (Games/Tools/Mini Programs) §1):
"Pure liberal arts student, never learned Python,made a logically complete Stranger Things-style game by chatting with code. "
2.2 Multi-tool division of labor is the norm; using a single tool to handle everything is an anti-pattern
Lao Huang's classic three-piece suite(newtype · AI Coding Practice and Tool Stack Evolution §2 / newtype · Vibe Coding Independent Product Practice §4):
| Tool | Role | Stage |
|---|---|---|
| v0.dev | UI / design drafts | Visual layer |
| Claude(Web/desktop) | Framework / architecture | Abstraction layer |
| Cursor / Claude Code | Details / debug | Implementation layer |
"Handing everything over to one tool will most likely not yield good results. These AI tools are not yet strong enough for that,each has its own strengths. So, accept this reality,stop yapping, JUST DO IT. "
Anti-pattern: Trying to use Cursor alone to handle everything from UI to deployment → will inevitably hit a wall.
2.3 The 4 code retrieval methods each have their own strengths (Lao Huang exclusive)
newtype · AI Coding Practice and Tool Stack Evolution §4 (2025-06 Post 1392 highlights):
| Tool | Retrieval method | Advantages | Disadvantages |
|---|---|---|---|
| Cline | Real-time search (imitates humans, high-speed file content + regex) | Accurate, real-time | High token consumption |
| Claude Code | Agentic Search (dynamically decides content/filename search + multi-round strategy) | Highest accuracy | Even higher token consumption |
| Cursor | Classic Embeddings + RAG | High cost-effectiveness | Limited by Embedding comprehension capability |
| WindSurf | Same as Cursor but with low latency + enterprise deployment | Enterprise-grade | Slow initial indexing |
Cline's core reason for rejecting RAG: Function definition at line 200 / call at line 300 / dependency at line 400 → RAG can only read fragments, breaking the logic.
See Claude Code Workflow (Cross-Author Thematic Synthesis) §1.2 Agentic Search vs RAG.
2.4 Three tiers of Ultra think (late 2025 consensus)
newtype · AI Coding Practice and Tool Stack Evolution §6.2 (2025-09 Post 1558):
| Keyword | Thinking tokens |
|---|---|
Ultra think / ultrathink |
32K(HIGHEST) |
think hard / think harder |
10K(MIDDLE) |
think |
4K(BASIC) |
| (none) | 0(NONE) |
Lao Huang's practical principle: When Claude Code gives a wrong answer,first add "Ultra think" and retry, 90% of the time it's insufficient budget, not the model being bad.
Cal Rueb's counterintuitive addendum(Claude Code Best Practices (Cal Rueb · Anthropic Internal Talk)):Don't use ultrathink every time; it wastes tokens and slows down responses. If you repeatedly use ultrathink, it means the task granularity is too large and needs to be split smaller.
2.5 Plan Mode is the first choice for complex tasks (consensus)
newtype · AI Coding Practice and Tool Stack Evolution §6.3:
- Turn on Plan Mode, tell Claude the requirements + thinking (especially difficulties and possible solutions)
- It plans first,does not directly touch the code
- Suitable for complex tasks (Lao Huang Obsidian MCP
notes_insightnew feature)
Wang Kai corroborates: "Claude Code produces product architecture: don't base it on the current product state; first clearly state the product goal then let it produce the architecture + reverse-compare what the current state is missing" (equivalent to Plan Mode).
Lao D corroborates: "Let the AI first understand the requirements + discuss with you + clearly answer all the questions it raises → higher success rate" (lightweight Plan Mode).
2.6 "JUST DO IT" / learning programming while doing (consensus but with rhythm differences)
Lao Huang(newtype · AI Coding Practice and Tool Stack Evolution §10):
"JUST DO IT. Whether doing content or development,absolutely do not wait until you have learned everything before starting. Everything is solved while moving forward. "
Wang Kai(Wang Kai's AI Coding Practice: From Cursor to Claude Code Terminal Workflow §1):
"Don't study, just use —— don't spend time learning technical principles, just start doing it, ask AI when stuck."
Xiaohongshu(Xiaohongshu / Vibecoding Practice Collection (Games/Tools/Mini Programs) §1):
"Nurture-style programming is very addictive. First build the skeleton, then add flesh and blood, finally add the finishing details."
Lao D(dontbesilent AI Self-Media Course / 13 Lessons Systematic Methodology §L08): "Practical case: transcribe two audio files → Claude Code automatically writes a Python script to call the API → automatically fixes errors → generates TXT → automatically writes code again → completed".
But Lao Huang's exclusive warning: "Don't believe those UP hosts who say zero foundation can also develop.You still have to understand programming and have programming thinking, otherwise you'll be infuriated by tools like Cursor."
3. Explicit disagreements (4 points —— not smoothed over)
3.1 Disagreement 1: Should domestic tools be used?
| Perspective | Position |
|---|---|
| Lao Huang(newtype · AI Coding Practice and Tool Stack Evolution §3) | "Have never touched domestic ones at all" — 2025-02 Post 556: "If something is good enough, or I'm sure it's the best at the current stage, then I won't look at others anymore. For AI Coding currently I only look at Cursor and GitHub Copilot" |
| Wang Kai(Wang Kai's Claude Code Terminal Workflow Methodology (Full Version) §3.3) | Multi-model mixed stack includes domestic Kimi — "Claude Code as main force + Codex CTO review + Gemini/Kimi as fallback". Kimi 2.6 has stunning design aesthetics (Xiaohongshu / Vibecoding Aesthetics + Creator Economy Collection §2) |
| Lao D(dontbesilent AI Self-Media Course / 13 Lessons Systematic Methodology §L01) | Domestic first choice DeepSeek + Monica + Yuanqi + Doubao — full domestic adaptation |
| Xiaohongshu vibecoding | Mixed domestic + foreign — Kimi 2.6 / Codex / Cursor / Gemini / Jimeng |
| Wang Kai's sarcasm(Wang Kai's Claude Code Terminal Workflow Methodology (Full Version) §13) | "TRAE (ByteDance's overseas version) is cheap and usable with the Claude API / Qoder (Alibaba) serves as a backup for Chinese-language scenarios" |
Real disagreement:Whether domestic tools can enter core workflows. Lao Huang firmly sticks to "only watch the best" (believing domestic tools lag too far behind); Wang Kai / Lao D use domestic tools in practice because of domestic adaptability + design aesthetics + pricing.
Judgment criteria:
- Overseas users + indifferent to cost → Lao Huang's route (Cursor + Claude Code + no domestic tools)
- Domestic users + no scientific internet access → Lao D's route (DeepSeek + OpenRouter + Monica)
- Matrix operation + multi-model hybrid deployment → Wang Kai's route (Claude Code + Kimi/Gemini as fallback)
3.2 Disagreement 2: Is it worth running local large models for AI Coding?
| Perspective | Position |
|---|---|
| Lao Huang(newtype · AI Coding Practice and Tool Stack Evolution §11.2 / 2025 Post 593) | "Local large models are not recommended" — "The coding capability of local large models cannot match Claude 3.5 Sonnet,it is not advisable to spend twenty to thirty thousand yuan on this. Use Cursor / Claude Code subscriptions;even if API costs are high, they are cheaper than hiring developers" |
| Xiaohongshu vibecoding(Xiaohongshu / AI Tools + Skill Tutorial Collection §9) | Want to run nanochat locally / Karpathy's 243-line reproduced LM — for learning purposes, not production |
| Wang Kai | Has not touched local large models — directly pays for a $200 Claude subscription |
Real disagreement:Different use cases:
- Learning / understanding LLM principles → run nanochat / Ollama / LM Studio locally (refer to newtype · AI Tool Stack Practice and Evolution)
- Production / real development → Lao Huang/Wang Kai's position, subscribe directly
3.3 Disagreement 3: OpenCode camp vs Claude Code camp?
| Perspective | Position |
|---|---|
| Lao Huang(newtype tool stack: Oh My OpenCode customization + newtype CLI multi-Agent team) | OpenCode camp — Oh My OpenCode 1729-line customization guide / 10 built-in Agents localized to Chinese / SUL-1.0 license / Fork repository / npm publish |
| Wang Kai / Lao D / Cal Rueb | Claude Code camp — Official Anthropic + 4 major stacks: Skill / MCP / Slash / Sub-agents |
Why Lao Huang takes the OpenCode camp:
"Oh My OpenCode is a powerful plugin for OpenCode,with multi-model orchestration + parallel background agents + LSP/AST tools — more flexible than native Claude Code,supports full Fork + Chinese localization. "
Typical OpenCode advantages:
- Multi-model (switch models directly with OpenRouter)
- Parallel background agents (10 built-in Agents + custom ones)
- Fully forkable (free for non-commercial use under SUL-1.0)
- Chinese localization (automatically replace Agent names via perl)
Claude Code camp advantages:
- Complete Skill system
- Mature MCP ecosystem
- Continuous official investment from Anthropic
Real disagreements are not smoothed over:This is the fundamental divergence between open source / customizable and commercial / fully integrated. Lao Huang uses both camps (Claude Code for writing code + OpenCode packaged into newtype CLI 8-Agent content team).
3.4 Disagreement 4: Is Vibe Coding the "ultimate form" or something that "will be cleaned out"?
| Perspective | Position |
|---|---|
| Lao Huang(newtype · AI Coding Practice and Tool Stack Evolution §9.3 / 2025-08 Post 1789) | Prophecy of Vibe Coding cleansing services — "When you pass the validation period of using Vibe Coding products and start scaling up for real,you will very likely need such a service. Because code written by AI may contain many hidden pitfalls" |
| Lao Huang's personal experience(same post) | Prompt House user login verification, subpage flow suddenly became unstable after adding more subpages(repeated refreshes) → spent a large amount of time troubleshooting together with Codex |
| Xiaohongshu vibecoding(Xiaohongshu / Vibecoding Practical Collection (Games/Tools/Mini Programs) §2) | The 0-80 vs 80-100 life-or-death line — "Web coding takes a product from 0 to 80, but 80 to 100 is the state where the product truly becomes usable" |
| Xiaohongshu + Lao Huang | Consensus: Vibe Coding cannot replace thinking about architecture / performance / SEO / security |
Key anti-pattern(Lao Huang §9.4):
"With Vibe Coding, you don't have to write code, but the requirements in other aspects are not lowered at all. "
Why:
- Architectural thinking still needs to be done
- Phased refactoring still needs to be done
- SEO / performance / security / user experience still need to be done
- Otherwise AI will accumulate more and more code that "barely runs" → eventually explode
Xiaohongshu's "80→100 difficulties"(exclusive):
- Non-code logic(dictionary integration + image-text matching + emoji word mapping = dirty work)
- Review time(2 months for mini-game review but mini-program code cannot pass)
- User experience(80→100 is the real life-or-death dividing line)
- Interest > technology(development completion relies on self-motivation, no monetary rewards in the early stage)
4. Wang Kai exclusive: How AI Coding changes collaboration workflows (the most profound perspective)
Wang Kai's AI Coding practice: From Cursor to Claude Code terminal workflow §5:
4.1 What AI Coding has changed
Internet era:
- Product writes requirements → design draws drafts → tech develops → joint debugging and testing → launch
- Cycle of several weeks
AI Coding era:
- Product writes Demo themselves(including backend APIs)
- Tech review + optimization (run locally after receiving)
- Directly modify product details on the Demo
- Cycle of 24-48 hours
Wang Kai's summary: Previously, the idea that "AI makes me do things faster" was about cost reduction and efficiency improvement;AI Coding is capability extension — things you could not possibly do before, you can do now.
4.2 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: Spent 1 hour using TRAE + FMP API to pull 10 years of data for verification → the strategy underperforms the Nasdaq
- Conclusion:AI Coding gives ordinary people the ability to refute with data
4.3 Changing work inertia
- Previously opened the browser for 7/8 hours a day
- Now mainly uses the Claude Code terminal, browser only for viewing information
- All browser operations are done through Claude Code (Chrome MCP)
4.4 Evolution path of token consumption (Wang Kai's personal experience)
| Stage | Tools | Monthly expenditure | Key changes |
|---|---|---|---|
| Cannot code | Claude + ChatGPT | < $100 | Ordinary conversation |
| Using AI Coding | Cursor, Qoder, Trae, v0 open simultaneously | ~ $200 | Token volume ×10 |
| VS Code + Claude Code | By project | rise | Still constrained by project-based thinking |
| Terminal + Claude Code | Breaking away from project-based thinking | $125 → $250 → $200 | Token volume ×2 again |
| Multi-terminal windows + multi-model | CC, Codex, Gemini, Kimi each running | Same as above | Token volume ×3 again |
| Multi-browser concurrency + Skill + cyclic tasks | Chrome MCP × N + Feishu CLI + 10-minute patrol checks | $200, billed by API, $50 used in 3 hours | "Split work into dozens of Skills" |
5. Three core mindsets of Xiaohongshu vibecoding (exclusive)
Xiaohongshu / Vibecoding Practical Collection (Games/Tools/Mini Programs) §1 —— Extracted from Stranger Things game video:
| Mindset | Content |
|---|---|
| 1. State requirements clearly | You don't need to know programming languages.Like directing a vendor company Tell the AI: "1980s retro UI style, visual effects should have the texture of gloom and electromagnetic interference" |
| 2. Progress step by step (nurturing style) | Never aim for perfection in one go. First build the skeleton, then add flesh and blood, finally add the finishing details.AI will calculate appropriate values based on your content; watching your ideas land bit by bit, nurturing-style programming is very addictive |
| 3. Be a picky client | Type directly wherever something looks off,tell it how to adjust in natural language. "Code will flow and automatically reorganize like water according to your aesthetics" |
Core insight (2026): The greatest significance of Vibe coding for people without a technical background is —— for the first time, aesthetics runs ahead of tools.
"You are no longer a novice judged by complex code,you can also be the chief director of this world. "
"Sometimes you are only one chat box away from being the creator."
6. Lao Huang's OpenCode camp exclusive (newtype CLI)
newtype tool stack: Oh My OpenCode customization + newtype CLI multi-Agent team:
6.1 Oh My OpenCode customization guide (1729 lines)
10 built-in Agents + Chinese localization:
| English name | Chinese localization | Role |
|---|---|---|
| Sisyphus | Sisyphus | Main Agent / default orchestration |
| oracle | Oracle | Architecture consultant |
| librarian | Librarian | Document/knowledge retrieval |
| explore | Explorer | Code exploration |
| frontend-ui-ux-engineer | Frontend Engineer | UI / UX implementation |
| document-writer | Document Writer | Document generation |
| multimodal-looker | Multimodal Observer | Image / multimodal |
| Metis | Metis | Planning consultant |
| Momus | Momus | Plan review |
| orchestrator-sisyphus | Orchestrate Sisyphus | Multi-Agent orchestration |
2 customization methods:
- Configuration file(
~/.config/opencode/oh-my-opencode.json)— Entry level - Fork + GitHub / npm — Advanced (can change Agent names + full Chinese localization)
6.2 newtype CLI 8-Agent content team
8 professional Agents:researcher / writer / editor / fact-checker / extractor / analyst / archivist / chief
8 core commands:nt research / write / edit / fact-check / analyze / extract / archive / pipeline
Multi-Agent communication conventions:JSON envelope + NDJSON stream + 4 exit codes
7. Lao D's domestic adaptation stack (exclusive)
dontbesilent AI self-media course / 13-section systematic methodology §L08 —— Using Claude Code in China without scientific internet access + without paid membership:
Installation steps:
- Download an AI-empowered command-line tool (to replace the native Terminal) — you can ask directly when encountering errors
- Go to the Claude Code official website, copy the one-line installation command → paste it
- When errors occur, the AI command line gives prompts → click install to proceed
- Use OpenRouter instead of Claude official membership(starting at ¥40)
- Use third-party tutorials to connect the OpenRouter Key to Claude Code
- Launch Claude Code → chat with it directly in Chinese
Practical case(transcribing two audio files):
我 → "这个文件夹有 2 个音频,用 ElevenLabs API 转成文本"
Claude Code → 自动写 Python 脚本调 API → 报错自动改 → 生成 TXT
我 → "再把 TXT 转 Word"
Claude Code → 又自动写代码 → 完成
8. Vibe Coding commercial realization: Astro Kid's 400 million financing (Xiaohongshu exclusive)
Xiaohongshu / Vibecoding Practical Collection (Games/Tools/Mini Programs) VII.Astro Kid's 400 million financing = platform-based realization of Vibe Gaming(2026-05-17):
- TikTok-style AI game feed
- $100 million creator fund
- 20 million users in August
Significance: Vibe Coding has upgraded from "personal hobby" to "platform-based 400 million financing"—— confirming Lao Huang's judgment that "Agents are new content" (AI Agent Architecture and Multi-Agent Orchestration (Cross-Author Thematic Synthesis) §2.4).
8b. Shensi Circle AI Coding company case perspective (exclusive)
Entry:Shensi Circle author complete works entry
Topic wiki(added 2026-05-18):
- Shensi Circle / AI Coding Company Case Studies (Cursor/Lovable/Bolt/Devin/Replit) —— Cursor $50 billion valuation + Lovable $50 million ARR in 6 months + Readdy.ai $5 million ARR in 4 months
8b.1 Cursor "Training as Product"
- Composer model token efficiency 4x(2025-11 Sasha Rush)
- microVM same-environment training + production —— Training-as-product philosophy
- Anti-framework route —— Do not use LangChain, write the underlying layer yourself (consensus among Cursor / Perplexity / Lovable)
8b.2 Lovable $50 million ARR in 6 months
- 1 million users in 10 months
- Growth lead goes public6 new growth tactics(Traditional growth strategies have completely failed)
- No paid traffic / No keyword SEO / Heavy UGC + showcase + community
8b.3 Readdy.ai paradigm (Chinese-founder exemplar)
- Approaching $5 million ARR in 4 months
- Lanhu + MasterGo team (deeply focused on design tools for many years)
- Multi-page website generation + Production-ready code(HTML/React/Vue)
- Paradigm judgment: "Turning design from an 'operational skill' into an 'expressive ability'"
8b.4 Devin + Claude Code tripartite rivalry
"After Cursor, Devin, and Claude Code, another dark horse in AI coding is rapidly rising"
3 major programming Agent paradigms:Cursor (IDE enhancement)/ Devin (autonomous Agent)/ Claude Code (CLI collaboration).
8b.5 The 80→100 life-or-death line (author of Cilegeci)
"Web Coding solves the product from 0 to 80, but 80 to 100 is where the product truly reaches a usable state"
See Aesthetic / creation democratization (cross-author topic synthesis) §5.
9. Current (2026-05) mainstream tool stack comparison
| Tool | Lao Huang current | Wang Kai current | Xiaohongshu vibecoding | Lao D domestic |
|---|---|---|---|---|
| Main battlefield | Claude Code (primary) + Cursor (secondary) + Codex (added 2026-04) | Claude Code terminal($250/month) | Cursor + Codex (after 2026-12) | Claude Code via Openrouter |
| UI design | v0.dev (early product)/ Keynote (late-stage Go Fundamental) | — | Figma + Codex | — |
| Code search | Gemini 2.5 Pro (large-context code scanning) | — | Gemini 2.5 solves core functions in early stage | — |
| Multi-Agent | newtype OS(Chief/Deputy/Specialist) | Multi-terminal multi-browser matrix | Single Agent as primary | LobeChat conversation branching |
| Local large models | Not recommended | Do not use | Learning nanochat | Do not use |
| Model combination | Claude (planning) + Gemini (large context) + Sonnet/Flash (execution) | Claude Code + Codex CTO + Gemini + Kimi | Kimi 2.6 design + Codex refactoring | DeepSeek + Monica + Yuanqi |
| Usable in China | ❌ | ❌ (Requires overseas infrastructure) | ✅ Partial | ✅ Full pipeline |
10. How to choose — AI Coding paths by identity
| Who you are | Primary tools | Starting point |
|---|---|---|
| Complete beginner / non-technical / based in China | Lao D route | Openrouter + AI command line + Don't ask again, chat directly in Chinese |
| Non-technical + want to build indie products | Lao Huang route | v0 + Claude + Cursor three-piece set → switch to Claude Code main battlefield |
| Product manager / want to do data analysis | Wang Kai route | Cursor + FMP API data analysis → Claude Code terminal multi-window → multi-browser matrix |
| Individual creator / want to make small games | Xiaohongshu route | Three core mindsets (state requirements clearly / progress step by step / be a picky client) + 0-80 vs 80-100 vigilance |
| Want full customization / Sinicization | Lao Huang OpenCode camp | Oh My OpenCode Fork + 10 Agent Sinicization + perl auto-replacement |
| Enterprise-level / teams | Claude Code + Cal Rueb engineering | Multi-layer CLAUDE.md + Hooks + three-tier Permission |
Universal starting order(For any identity):
- Install Claude Code or Cursor (depending on budget)
- Must have programming thinking(5 concepts: variables/functions/asynchronous/API/database → watch a few Python introductory videos)
- Learn the three core mindsets:state requirements clearly + progress step by step + be a demanding client
- Add
Ultra thinkkeyword (retry when answer is wrong) - After resolving each issue → update CLAUDE.md(Lao Huang)
- Know the 0-80 vs 80-100 life-or-death line (Xiaohongshu)
- See the prophecy of Vibe Coding cleansing services (Lao Huang)→ Architecture / performance / SEO thinking still needs to be done
11. Reference list
Main original author wikis:
- newtype · AI Coding Practice and Tool Stack Evolution —— Lao Huang AI Coding timeline / 4 retrieval methods / Ultra think 32K / Plan Mode / Vibe Coding cleansing prophecy / anti-pattern quick reference
- newtype tool stack: Oh My OpenCode customization + newtype CLI multi-Agent team —— Oh My OpenCode 1729 lines of customization / 10 Agent Sinicization / newtype CLI 8 commands / SUL-1.0 license
- newtype · Vibe Coding Indie Product Practice —— Lao Huang indie products / Web → Mac App / full-stack breakdown / Go Fundamental
- Wang Kai AI Coding Practice: From Cursor to Claude Code Terminal Workflow —— Wang Kai 385 posts / collaboration workflow restructuring / disruptive data analysis / token consumption evolution path
- Wang Kai Claude Code Terminal Workflow Methodology (Complete Edition) —— Wang Kai Claude Code overview / terminal window as Agent / 10min patrol / multi-model mixed orchestration
- Xiaohongshu / Vibecoding Practice Collection (Games/Tools/Mini Programs) —— Three core mindsets / 0-80 vs 80-100 / Cilegeci / Astro Kid 400 million financing / desktop pet / 30-day ARPG
- Xiaohongshu / AI Tools + Skill Tutorial Collection —— Pixel characters / Vibe Marketing Skill / IELTS Skill three-piece set / nanochat
- Xiaohongshu / Vibecoding Aesthetics + Creator Economy Collection —— Aesthetics first / tuning Kimi 2.6 / 3 tricks to achieve 100k-level design effects
- dontbesilent AI Self-Media Course / 13-lesson systematic methodology —— Lao D domestic 13 lessons / Openrouter / practical audio transcription
Related topic wikis (this series):
- Claude Code Workflow (Cross-Author Topic Synthesis) —— Claude Code Workflow (Batch 1)
- AI Agent Architecture and Multi-Agent Orchestration (Cross-Author Topic Synthesis) —— AI Agent Architecture and Multi-Agent Orchestration (this batch Batch 2)
- Vibe Coding and Indie Development (Cross-Author Topic Synthesis) —— Vibe Coding / Indie Development Deep Dive (Batch 5)
- Claude Skill Engineering (Cross-Author Topic Synthesis) —— Skill Engineering (Batch 3)
History
- 2026-05-18:Topic-dimension wiki first release. Synthesized and rewritten from 9 author wikis. Explicitly marked 4 divergences: domestic tools / local large models / OpenCode vs Claude Code camps / whether Vibe Coding will be cleansed
Sources and related materials
- newtype · AI Coding Practice and Tool Stack Evolution
- newtype tool stack: Oh My OpenCode customization + newtype CLI multi-Agent team
- newtype · Vibe Coding Indie Product Practice
- Wang Kai AI Coding Practice: From Cursor to Claude Code Terminal Workflow
- Wang Kai Claude Code Terminal Workflow Methodology (Complete Edition)
- Xiaohongshu / Vibecoding Practice Collection (Games/Tools/Mini Programs)
- Xiaohongshu / AI Tools + Skill Tutorial Collection
- Xiaohongshu / Vibecoding Aesthetics + Creator Economy Collection
- dontbesilent AI Self-Media Course / 13-lesson systematic methodology
- Shensi Circle complete author entry portal
- Shensi Circle / AI Coding Company Case Studies (Cursor/Lovable/Bolt/Devin/Replit)
- Claude Code Workflow (Cross-Author Topic Synthesis)
- AI Agent Architecture and Multi-Agent Orchestration (Cross-Author Topic Synthesis)
- Vibe Coding and Indie Development (Cross-Author Topic Synthesis)
- Claude Skill Engineering (Cross-Author Topic Synthesis)
- Shensi Circle / AI Agent Company Case Library