Will’s complete pipeline for non-technical PMs to go from “idea to runnable prototype”, breaking vibe coding into 7 copyable steps: ① Use 3 tiers of templates to turn a one-sentence product idea into structured input; ② Have AI act as a “requirements interrogator” to ask follow-up questions in sequence and force out a Grill Summary; ③ Use a “top-tier product architect” role prompt to expand the 5 answers into an investment-grade PRD; ④ Extract feature points from the PRD to generate 3 non-overlapping interaction DRD options, then convert the chosen option into a coding prompt that can directly produce a product; ⑤ Feed it to Figma Make (to produce an interactive demo) or Claude (to produce a real launch-ready product) to generate a prototype; ⑥ Feed screenshots back in to let AI spot errors and iterate on its own; ⑦ Publish to mobile (Claude outputs HTML → Netlify Drop drag-and-drop hosting → add to home screen as an App) + one-click reskinning with ReferoStyle. Each step provides reusable prompt skeletons + Why + examples + pitfalls. Three parallel methods (Method 1 Document-driven mainline / Method 2 Document-driven HTML output / Method 3 GPT+Figma natural language UI building). The audience is non-technical product managers who care about “when to use which step, how to copy the prompts” rather than code. Division of labor with [[newtype-ai-coding]]: that article covers tool stack selection and mindset (how to choose Cursor/Claude Code/Codex), while this article covers the concrete action chain and prompt templates for non-technical PMs to run through a prototype from scratch.
30+ selected Q&A entries from Huang Yihe (newtype)'s Planet community from January to July 2025. Beginner learning path (Learning by doing: try all three models → ChatWise/Cherry Studio → OpenRouter) / Network conditions (US residential IP / Claude Code subscription vs API) / Multi-platform strategy (Domestic: WeChat Channels 68k + Bilibili & Douyin 40k combined + Xiaohongshu 10k / Overseas: YouTube + X + Medium + Reddit + newtype.pro) / Three reasons for not joining WeChat groups (filtering / education stage / time and energy) / AI Coding tool stack (Cursor + Claude Code complement each other) / Reasons for being bearish on Workflow vs bullish on MCP / Prompt House design philosophy (2-step MCP tool call) / Video production workflow (A Roll + B Roll + Screen Studio + OBS) / Three reasons for not subscribing to ChatGPT / RAG vs Fine-tune decision-making / Note structure Input-Processing-Output / Three things that don't rely on self-discipline (sense of mission + physical stamina + not forcing yourself). This is a "Q&A scenario" supplement to [[newtype-mindset-super-individual]] and [[newtype-knowledge-system]]
Shensi Circle 39 AI Coding company case studies. Cursor ($50 billion valuation + training as product + chief designer + Composer)/ Lovable ($50 million ARR in 6 months + 6 growth playbooks + anti-traditional growth)/ Bolt / Cognition Devin / Replit (SaaStr zero-engineer case)/ Codex / GitHub Spec Kit / Open-source AI coding tools ($32 million)/ Readdy.ai (Chinese-founded / $5 million ARR in 4 months). Covers coding Agent business models, growth playbooks, technical inside stories, and the 80→100 make-or-break line
The evolution path of AI Coding recorded across 385 posts: from a product manager who does not write code → using Cursor with a monthly cost of 50,000 USD → switching to the Claude Code terminal workflow → restructuring the basic collaboration process in the AI era. It emphasizes that "AI Coding is a foundational capability, not something for writing code."
