LogoAI Practice Knowledge Base
LogoAI Practice Knowledge Base
Homepage
Prompt to Product

Part 1: Foundation

What is Vibe CodingWhy AI ProgrammingSolo Founder MindsetIndie Hacker Tailwinds in the AI Era

Part 2: Discovery

Google Trends Demand MiningNew-Word StrategyLong-Tail Mining & One-Keyword-One-SiteReddit/X Pain-Point MiningProduct Hunt Competitive ResearchMVP Definition & Boundaries

Part 3: Tooling

Tool SelectionMCP Complete GuidePlaywright Browser AutomationAI Coding Practical Tips

Part 4: Methodology

From Vibe Coding to Spec CodingOpenSpec Hands-on GuideMBRY Prompt FrameworkAI Is Not a Chat Box

Part 5: Prompts

Prompt ArsenalComplete AI Coding Rules Guide

Part 6: Launch

Budget-Friendly Tech StackOn-Page SEO BasicsLink Building & Directory Submission

Part 7: Monetization

Stripe & International PaymentsPricing StrategyThe 80/20 Principle

Part 8: Marketing

Social Media & Build In PublicCold Start 100 UsersEmail List Newsletter

Part 9: Pitfalls

Anti-Patterns Guide
AI Self-Growth System

Part 0: The Laboratory

The Laboratory: Environment Setup

Part 1: Mindset - Understanding Compound Growth

What Is an AI Self-Growth SystemLinear vs Compounding GrowthFlywheel Effect ExplainedMaxwell's Demon Philosophy

Part 2: Engine - Four Core Modules

Content FactoryAutomated DistributionData Monitoring SystemFeedback Loop: System Self-Evolution

Part 3: SEO Factory - Traffic Compounding

pSEO Basics and PrinciplesKeyword Matrix DesignDrip Release StrategyJust-in-Time Release (JIT)Internal Linking AutomationSite Matrix and Fingerprint Isolation

Part 4: Social Leverage - Interception and Downscaling

Social Listening SystemHot Topic TransformerAuto-Reply InterceptorContent Format Arbitrage

Part 5: Viral Growth - User-Powered Distribution

Viral Product DesignShareable Result PatternLow-Friction Conversion DesignGamified Sharing Mechanism

Part 6: Knowledge Arbitrage - Becoming the Authority

Information Gap ArbitrageAggregation as a ServiceTrend Prediction EngineData Moat

Part 7: Portfolio Strategy - Scaling Systems

Portfolio StrategyUnified PassportCross-Promotion EngineAsset Reuse Engine

Part 8: Automation Endgame

Automation EndgameMonetization StackBuild to SellNext S-Curve

Part 9: Case Studies

Case Study: SEO Factory in PracticeCase Study: Viral Tool in PracticeCase Study: Matrix in Practice

Part 10: Human Advantage

Human AdvantageDark ForestFinal Manifesto
Counterintuitive Facts反直觉事实:终极选题规划 (No.068-100)

Writing Protocol

Canonical PromptArticle Template

Sample Articles

Counterintuitive Facts (1): How Do You Prove You're Not a Brain That Just Popped Into Existence From the Void?Counterintuitive Facts (2): Why Are All the 'Good Ones' Never on the Market?Counterintuitive Facts (3): Rules You Don't Understand Were Often Paid for in CorpsesCounterintuitive Facts (4): Why Do Elites Who Advocate 'Open Marriage' Stay Faithful Themselves?Counterintuitive Facts (5): Your Anger Is a Parasite Reproducing in Someone Else's BrainCounterintuitive Facts (6): Why Do You Prefer Fake Things? Because Real Things Aren't Stimulating Enough AnymoreCounterintuitive Facts (7): Why Does a Gazelle Stop to Jump Up and Down When It Sees a Lion, Instead of Running?Counterintuitive Facts (8): Most of Humanity's Greatest Achievements Are Evolutionary 'Waste'Counterintuitive Facts (9): 'For the Good of the Group' Is the World's Biggest LieCounterintuitive Facts (10): You're Sitting in an Office, But Your Body Thinks You're Fleeing FamineCounterintuitive Facts (11): Why Is the Most Rational Strategy at the Negotiating Table to Make Your Opponent Think You're Insane?Counterintuitive Facts (12): Why Would a Group of Smart, Good People Collectively Walk Into Disaster?Counterintuitive Facts (13): 'Everyone Knows' and 'Everyone Knows That Everyone Knows' Are Completely Different ThingsCounterintuitive Facts (14): How Did Kindness Survive in This Cold Universe?Counterintuitive Facts (15): Why Is Everyone Richer But More Anxious?Counterintuitive Facts (16): The Most Effective Threat Is a One Time Thing: You Only Get One ChanceCounterintuitive Facts (17): Why Would You Rather Lose Money Yourself Just to Make the Person Who Earned More Suffer?Counterintuitive Facts (18): Why Is More Expensive Waste Paper Worth More?Counterintuitive Facts (19): You Think the Universe Is Perfect Only Because You Haven't Died YetCounterintuitive Facts (20): Stupidity Is More Dangerous Than Evil Because Stupidity Cannot Be RefutedCounterintuitive Facts (21): If He Doesn't Pay for His Mistakes, His Advice Is GarbageCounterintuitive Facts (22): Why Does the Boss Always Promote the Stupidest Person?Counterintuitive Facts (23): Why Do Experts Lead the Persecution of Those Who Tell the Truth?Counterintuitive Facts (24): The Better Things Get, the Closer You Are to DeathCounterintuitive Facts (25): Why Do Dictators Who Ruin Their Countries the Most Often Live the Longest?Counterintuitive Facts (26): Making Money and Creating Wealth Are Completely Different ThingsCounterintuitive Facts (27): Free Things Are Often the Most ExpensiveCounterintuitive Facts (28): Why Do Smart People Also Go Down Dead End Roads?Counterintuitive Facts (29): Even If You're an All Around Genius, You Still Need Someone Who's 'Useless'Counterintuitive Facts (30): Why Is the Seat Next to You Half the Price of Your Ticket?Counterintuitive Facts (31): Why Is Effort Meaningless?Counterintuitive Facts (32): Why Do the Rich Get Richer and the Poor Get Poorer?Counterintuitive Facts (33): Why Does a Group of Smart People Become One Idiot?Counterintuitive Facts (34): Why Does a Consumer's Sneeze Cause a Factory Earthquake?Counterintuitive Facts (35): Why Are Keyboard Letters Arranged Randomly?Counterintuitive Facts (36): The Demon Everyone Is FeedingCounterintuitive Facts (37): Why Does Every Great Organization Eventually Become a Zombie?Counterintuitive Facts (38): The Emperor's New Clothes Happens Around You Every DayCounterintuitive Facts (39): Why Does Zuckerberg Dress Like a Computer Repair Guy?Counterintuitive Facts (40): When You Measure Something, You Destroy ItCounterintuitive Facts (41): How to Turn Lies Into Truth?Counterintuitive Facts (42): You Live in a Map Without TruthCounterintuitive Facts (43): You Have No Idea What You Actually WantCounterintuitive Facts (44): Why You Should Never Trust 'Average Returns'Counterintuitive Facts (45): The Older Something Is, the Less Likely It Is to DieCounterintuitive Facts (46): Why Does Your Room Always Get Messy on Its Own?Counterintuitive Facts (47): The Demon Who Died But Is Still Charging YouCounterintuitive Facts (48): Why Isn't Tomorrow's Sunrise News?
AI Wealth Truth

Chapter 1: The Hidden Physics of Wealth Distribution

AI Wealth Truth (01): Why Wealth Inequality Follows the Second Law of ThermodynamicsAI Wealth Truth (02): Why Can Randomness Create Extreme Inequality?AI Wealth Truth (03): Why 'Fair' Markets Make Inequality WorseAI Wealth Truth (04): Why the ZIP Code You Grow Up In Predicts Your Income Better Than IQAI Wealth Truth (05): Why the Role of Luck Is Systematically Underestimated by 90%AI Wealth Truth (06): Why 'Equal Opportunity' Is Mathematically ImpossibleAI Wealth Truth (07): Why the Poor's 'Irrational' Decisions Can Be the Optimal ChoiceAI Wealth Truth (08): Why the 'Middle Class' Is a Postwar Historical AnomalyAI Wealth Truth (09): Why Economic Growth Has Nothing to Do With Your Wage GrowthAI Wealth Truth (10): Why Technological Progress Makes Ordinary People PoorerAI Wealth Truth (11): Why Trickle-Down Economics Never WorkedAI Wealth Truth (12): Why Inflation Is a Hidden Wealth TransferAI Wealth Truth (13): Why Rising Housing Prices Make Society PoorerAI Wealth Truth (14): Why Financialization Shrinks the Real EconomyAI Wealth Truth (15): Why You Will Never 'Beat the Market'

Chapter 2: How Your Brain Makes You Poor

AI Wealth Truth (16): Why Your Brain Was Not Designed for Personal FinanceAI Wealth Truth (17): Why Higher Prices Can Make You Buy MoreAI Wealth Truth (18): Why Saving Small and Spending Big Is a Nervous-System BugAI Wealth Truth (19): Why You Pay More for FreeAI Wealth Truth (20): Why Losing 1 Hurts 2.5 Times More Than Gaining 1AI Wealth Truth (21): Why You Always Buy at Market Tops and Sell at Market BottomsAI Wealth Truth (22): Why Gut-Level Investment Decisions Can Be BetterAI Wealth Truth (23): Why Experts' Forecasts Can Be Worse Than RandomAI Wealth Truth (24): Why Your Intuition About Low-Probability Events Is Catastrophically WrongAI Wealth Truth (25): Why Sunk Costs Drain Your WealthAI Wealth Truth (26): Why You Overpay for "Optionality"AI Wealth Truth (27): Why Poorer People Are Easier to ScamAI Wealth Truth (28): Why Casinos Are Designed That WayAI Wealth Truth (29): Why Finance Apps All Look the SameAI Wealth Truth (30): Why the "Rational Man" Assumption Is Wrong at the Root

Chapter 3: Engineered Poverty: How Systems Extract You

AI Wealth Truth (31): Why Minimum Payments Are Banks' Most Profitable InventionAI Wealth Truth (32): Why "Interest-Free" Installments Often Mean You Pay 20% MoreAI Wealth Truth (33): Why Insurance Actuaries Live Ten Years Longer Than YouAI Wealth Truth (34): Why "Principal-Protected" Products Guarantee You LoseAI Wealth Truth (35): Why Bank Deposit Rates Are Almost Always Below InflationAI Wealth Truth (36): Why Pension Systems Are Ponzi Schemes Destined for InsolvencyAI Wealth Truth (37): Why Dollar-Cost Averaging Returns Are Often Exaggerated by 10xAI Wealth Truth (38): Why Medical Bankruptcy Is the No.1 Personal Financial KillerAI Wealth Truth (39): Why Higher Education Is Turning Into a High-Stakes BetAI Wealth Truth (40): Why "Buying a Home Is a Must" Is a Constructed IdeaAI Wealth Truth (41): Why Wage Growth Almost Always Lags Housing PricesAI Wealth Truth (42): Why Taxi Licenses Can Be Worth Hundreds of ThousandsAI Wealth Truth (43): Why Metacognition Is the Real Marker of Class StratificationAI Wealth Truth (44): Why Some Cities Keep "Purchase Restrictions" Without Increasing SupplyAI Wealth Truth (45): Why the System Does Not Want You to Understand These Things

Chapter 4: Wealth Black Holes of the Network Era

AI Wealth Truth (46): Why the "Free" Internet Costs You Tens of Thousands of DollarsAI Wealth Truth (47): Why Recommendation Algorithms Make the Poor Poorer and the Rich RicherAI Wealth Truth (48): Why Every "Viral" Hit Has Someone Harvesting ValueAI Wealth Truth (49): Why Live-Stream Shopping Prices Are Not Actually CheapAI Wealth Truth (50): Why "We Don't Sell Your Data" Is the Biggest LieAI Wealth Truth (51): Why "User Growth" Matters More Than ProfitAI Wealth Truth (52): Why Every "Viral Hit" Is Carefully Designed HarvestingAI Wealth Truth (53): Why You Are the One Who Ultimately Pays for Platform "Subsidy Wars"AI Wealth Truth (54): Why "Private Traffic" Is a Bubble About to BurstAI Wealth Truth (55): Why the Crypto Bubble Is Exactly Like the 17th-Century Tulip ManiaAI Wealth Truth (56): Why NFTs Are Not "Digital Ownership" but "Digital Tulips"AI Wealth Truth (57): Why "Metaverse Real Estate" May Be the Most Absurd Speculation in HistoryAI Wealth Truth (58): Why Retail Investors Who Rush Into Every "Tech Revolution" Die FirstAI Wealth Truth (59): Why FOMO Has Been WeaponizedAI Wealth Truth (60): Why "Deep Work" Is Becoming a Class Privilege

Chapter 5: Wealth Redistribution in the AI Era

AI Wealth Truth (61): Why AI Makes "Skills" Less ValuableAI Wealth Truth (62): Why Capital Still Wins in "Human-AI Collaboration"AI Wealth Truth (63): Why the Biggest Asset in the AI Era Is "Attention Sovereignty"AI Wealth Truth (64): Why "Data Labor" Is Not Recognized as LaborAI Wealth Truth (65): Why AI Chips Are Worth More Than AI AlgorithmsAI Wealth Truth (66): Why OpenAI's $7 Trillion Chip Plan Is a Power GameAI Wealth Truth (67): Why "AI Democratization" Is a LieAI Wealth Truth (68): Why AI Makes "Taste" the Last MoatAI Wealth Truth (69): Why "Personal Brand" Matters More Than Companies in the AI EraAI Wealth Truth (70): Why "One-Person Companies" Have an Advantage Over Big CompaniesAI Wealth Truth (71): Why the "Interface Layer" Is Always More Valuable Than the "Implementation Layer"AI Wealth Truth (72): Why AI Makes "Vertical" More Valuable Than "General"AI Wealth Truth (73): Why "Speed" Matters 10x More Than "Perfection" in the AI EraAI Wealth Truth (74): Why Real AI Dividends Mostly Belong to Capital OwnersAI Wealth Truth (75): Why "Technological Unemployment" Is Totally Different This Time

Chapter 6: Game Theory, Information Theory, and Wealth Warfare

AI Wealth Truth (76): Why the "Market for Lemons" Hurts Honest PeopleAI Wealth Truth (77): Why "Signals" Matter More Than "Ability" for Your IncomeAI Wealth Truth (78): Why Interviews Are a Game Where Both Sides LieAI Wealth Truth (79): Why Referrals Are 100x More Effective Than Cold ApplicationsAI Wealth Truth (80): Why the "Anchoring Effect" Is Worth Millions in NegotiationAI Wealth Truth (81): Why "Silence" Is the Strongest Weapon in NegotiationAI Wealth Truth (82): Why the Principal-Agent Problem Lets You Get Extracted in Every RelationshipAI Wealth Truth (83): Why Incentive Compatibility Is the Key to Designing Any SystemAI Wealth Truth (84): Why the Tragedy of the Commons Is Replaying on the InternetAI Wealth Truth (85): Why the Prisoner's Dilemma Explains Most Social ProblemsAI Wealth Truth (86): Why First-Mover Advantage Can Be a CurseAI Wealth Truth (87): Why "Slow Variables" Matter More Than "Fast Variables" for Your FateAI Wealth Truth (88): Why Feedback Delay Makes You Unable to LearnAI Wealth Truth (89): Why Complex Systems Make Experts' Forecasts WorthlessAI Wealth Truth (90): Why Black Swans Are Becoming More Frequent

Chapter 7: Ultimate Cognition: The Philosophy and Nihilism of Wealth

AI Wealth Truth (91): Why You Are Playing a "Finite Game" While the Rich Play an "Infinite Game"AI Wealth Truth (92): Why Money May Be Humanity's Biggest "Consensus Illusion"AI Wealth Truth (93): Why "Economic Growth" May Be a Game Near Its EndAI Wealth Truth (94): Why GDP Growth Did Not Make Humans HappierAI Wealth Truth (95): Why "Success" Might Be a Carefully Designed Social ControlAI Wealth Truth (96): Why the Richer You Are, the More Anxious You Can BecomeAI Wealth Truth (97): Why "Lying Flat" Might Be a Rational ResistanceAI Wealth Truth (98): Why "Meaning" Cannot Be Bought With MoneyAI Wealth Truth (99): Why the Richest People Often Give Away Most of Their WealthAI Wealth Truth (100): If Wealth Is Ultimately Meaningless, Why Pursue It?
X (Twitter)

Spec Coding

从 Vibe Coding 到 Spec Coding

前面几章讲的都是怎么用 AI 编程。这一章讲的是怎么用对 AI 编程。

区别在哪里?用 AI 写代码谁都会。打开 Cursor,随便说几句,AI 就开始写了。但写出来的代码是不是你想要的?能不能直接用?需不需要反复修改?这些问题的答案,取决于你用的方法。

2025 年的 AI 编程世界里,有两种截然不同的方法论:Vibe Coding 和 Spec Coding。

什么是 Vibe Coding

Vibe Coding,可以翻译成"氛围编程"或"凭感觉写代码"。它的做法是:你有一个模糊的想法,告诉 AI,AI 开始写代码,写出来不太对,你再补充一些信息,AI 再改,来来回回,最后勉强凑出一个能跑的东西。

这种方式有几个问题。首先,需求是散落在聊天历史里的。你说一句,AI 理解一点,过几轮对话 AI 把前面的需求忘了。其次,AI 做的很多技术决策你没有参与,可能选了不合适的方案。最后,代码写完了,但你也不太清楚它具体实现了什么,文档也没有。

对于一个人随便折腾的小项目,Vibe Coding 问题不大。大不了就是一坨屎山代码,反正只有你自己用。但在团队里,如果每个人都这么干,最后就是一座巨型屎山。直到某一天 CTO 受不了了,痛定思痛说:这个项目我们重构吧。

Vibe Coding vs Spec CodingVibe Coding 氛围编程想到哪写到哪需求散落在聊天历史里AI 自作主张做技术决策适合个人小项目Spec Coding 规范驱动开发先写规范再写代码需求文档化、可追溯人类参与关键决策适合团队和正式项目

什么是 Spec Coding

Spec Coding,规范驱动开发,是完全相反的思路。核心原则是:在写任何代码之前,先把规范写清楚。

这里说的规范不是那种几百页的需求文档。它可以是一个简单的 Markdown 文件,描述你要做什么、为什么做、怎么做、验收标准是什么。关键是形成一个明确的、可执行的蓝图,然后 AI 根据这个蓝图来写代码。

为什么要这样做?因为 AI 写代码的能力和理解需求的能力是不对等的。它可以很快地写出一堆代码,但如果需求不清晰,写出来的代码只是"看起来像那么回事",实际上可能完全不是你想要的。

前期工作准备得越充分,AI 写代码就越能够一次性到位。这个道理其实很简单,但很多人不愿意做。觉得写文档太繁琐,不如直接让 AI 动手。结果就是后期反复修改,浪费更多时间。

软件工程早就告诉我们一件事:完成一个项目会产生大量文档——需求文档、设计文档、测试文档、部署文档。代码只是其中一个环节。AI 时代没有改变这个本质,只是把写文档和写代码的过程都加速了。

规范驱动开发的工作流

一个典型的 Spec Coding 工作流是这样的。

首先是定义阶段。你用自然语言描述你要做什么。比如"我要给网站加一个用户评论功能"。这时候你和 AI 讨论需求细节,来回几轮对话,把评论功能具体包括什么、不包括什么、有什么特殊要求都聊清楚。AI 帮你把这些整理成一个 requirements.md 文件。

然后是设计阶段。需求清楚了,开始讨论怎么实现。评论功能需要改哪些数据库表?API 接口怎么设计?前端组件怎么组织?这些技术决策不是 AI 一个人做的,是你们一起讨论的。讨论完形成一个 design.md 文件。

接着是任务拆分阶段。把设计拆成具体的任务。任务 1 是创建评论表,任务 2 是写后端 API,任务 3 是前端评论组件……形成一个 tasks.md 文件,每个任务有明确的边界和验收标准。

最后才是实现阶段。AI 按照任务列表一个个完成,每完成一个就标记一下。你 review 代码,确认符合预期,再继续下一个。

这个流程看起来繁琐,但实际上比 Vibe Coding 更高效。因为大量的返工被前期的规范工作消除了。需求聊个三五轮,代码一次到位——这是真实体验。

Spec Coding 工作流📝Define 定义需求讨论requirements.md→🏗️Design 设计技术方案design.md→📋Tasks 任务拆分成可执行步骤tasks.md→💻ImplementAI 按任务写代码code files

工具支持:GitHub Spec Kit

2025 年 GitHub 官方发布了一个叫 Spec Kit 的开源工具包,专门支持规范驱动开发。

Spec Kit 的设计思路是把规范变成"可执行的蓝图"。它定义了四个阶段:Specify(定义目标和用户成果)、Plan(规划架构和约束)、Tasks(拆分成可测试的任务单元)、Implement(AI 逐步实现并接受 review)。每个阶段都需要验证通过才能进入下一个阶段。

使用方法是通过一系列斜杠命令。比如 /speckit.constitution 用来定义项目的核心原则、编码规范、安全策略;/speckit.specify 用来创建具体功能的规范;/speckit.plan 用来制定技术计划;/speckit.tasks 用来生成可执行的任务列表。

Spec Kit 的亮点是它可以和各种 AI 工具配合使用——GitHub Copilot、Claude Code、Gemini CLI 都可以。而且它强调规范是一个"活文档",会随着项目演进不断更新。

这个工具特别适合从 0 到 1 的新项目,尤其是企业级团队需要清晰度、合规性和可追溯性的场景。

项目地址:github.com/github/spec-kit

工具支持:OpenSpec

另一个值得关注的工具是 OpenSpec,由 Fission AI 开发。

和 Spec Kit 相比,OpenSpec 更轻量、更敏捷。它的设计目标是让规范驱动开发能快速融入你现有的工作流,特别适合改造已有项目(所谓的"brownfield"项目)。

OpenSpec 的工作流程更简单:提议(propose)一个变更,AI 审核并实施,然后归档(archive)。每一个 proposal 都是一个独立的文档,记录这次变更的目标、方案和验收标准。

安装方法很简单:

npm install -g @fission-ai/openspec@latest

然后在项目目录里执行:

openspec init

它会问你用什么 AI 工具,然后在项目里创建一个 openspec 文件夹,并给你一段提示词让你发给 AI。AI 读了这段提示词之后,就知道怎么按照 OpenSpec 的规范来工作了。

我自己的实践是这样的:如果我想给项目加一个新功能,不再是想到一出是一出,而是先让 AI 生成一个 OpenSpec proposal。有时候我有很多想法但不着急实现,就先让 AI 把这些想法都做成 proposal 放着。后续实现一个,就归档一个。

OpenSpec 配合 2025 年最新的 Claude Opus 4.5 模型,基本上需求聊三五轮,代码就能一次到位。这种体验确实让人直呼惊艳。

项目地址:github.com/Fission-AI/OpenSpec

AWS Kiro:IDE 级别的 Spec 支持

2025 年 7 月,AWS 发布了一个新的 AI IDE 叫 Kiro。它直接把规范驱动开发内置到了 IDE 里,不需要安装额外的工具。

Kiro 的设计理念是:你只需要用自然语言描述你想做什么,Kiro 会自动帮你生成三个文件。第一个是 requirements.md,包含用户故事和验收标准。第二个是 design.md,包含技术架构、序列图和实现注意事项。第三个是 tasks.md,是具体的任务清单。

然后 AI 根据这些规范文件来写代码,每完成一个任务就在任务清单上打勾。你可以看到进度,可以随时介入 review。

Kiro 还有一个叫 Hooks 的功能,可以设置触发条件。比如每次文件保存后自动跑测试,每次提交前自动检查安全问题。这让规范驱动不只是写代码的规范,也包括了工作流程的规范。

另外 Kiro 支持 MCP(Model Context Protocol),可以连接外部数据源和工具。和 AWS 服务的深度集成也是它的卖点。

当然,你不一定要用 Kiro。即使用 Cursor 或者 Windsurf,也可以手动在项目里创建 docs 文件夹,按照 requirements.md、design.md、tasks.md 的结构来组织规范文档。AI 看到这些文档自然就会按照规范来工作。工具只是辅助,方法论才是核心。

Spec Coding 工具对比GitHub Spec Kit官方出品,功能完整四阶段流程斜杠命令驱动✓ 适合从 0 到 1 新项目✓ 企业级合规需求OpenSpec轻量敏捷提议-实施-归档快速融入现有项目✓ 适合从 1 到 n 迭代✓ 现有项目改造Kiro IDE (AWS)IDE 级别集成自动生成规范文档Hooks 自动化✓ 开箱即用✓ AWS 深度集成

更轻量的方法:PDD

如果觉得 Spec Coding 还是太重,有一个更轻量的变体叫 PDD(Prompt Driven Development),提示驱动开发。

PDD 的核心是四个步骤。

第一步是 Context Curation,喂料。不要直接问 AI。先收集这次任务需要的上下文——相关的文档、类型定义、数据库 schema——用 @ 符号喂给 AI。上下文越准确,AI 的输出就越靠谱。

第二步是 Intent Definition,下指令。写伪代码和注释,而不是让 AI 自己想。比如"写一个函数,处理用户登录,参考 @AuthService 的逻辑,返回 JWT token,处理密码错误和账号被锁定两种异常"。这比直接说"帮我写登录功能"精确得多。

第三步是 AI Generation & Iteration,生成与迭代。AI 生成代码,你跑一下,有错误把 error log 甩回给 AI,让它自己修。来回几次直到跑通。

第四步是 Expert Review,专家审核。代码跑通不等于代码正确。你作为"架构师"要检查安全性、性能、业务逻辑是否正确。AI 不会替你做这个判断。

PDD 比 Spec Coding 轻很多,适合规模不大的任务。它的核心思想和 Spec Coding 是一致的:先想清楚再动手,而不是让 AI 瞎猜。

落地的挑战

规范驱动开发虽然好,但落地有困难。

最大的挑战是习惯问题。很多开发者觉得"明明我用 AI 一句话就能解决,为什么非要先写规范文档"。对于简单任务确实如此。但对于复杂任务,先写规范反而更快。这个转变需要亲身体验才能理解。

另一个挑战是维护成本。如果规范文档和代码脱节,规范就变成了形式主义。代码改了文档没改,下次 AI 读了过期的文档,写出来的代码更有问题。解决方法是把更新文档作为每次任务的一部分——代码改完,规范文档也要相应更新。

团队协作的挑战更大。Spec Coding 不只是一个工具的问题,是整个开发流程的变革。它需要配合 Git 工作流的改造、Code Review 标准的升级。团队里有人坚持写规范,有人坚持 Vibe Coding,最后的代码质量还是会参差不齐。

Spec Coding 的本质不是工具,而是流程变革。 工具免费,但"让工具在团队里跑通"的经验,才是最昂贵的。

我的实践建议

根据这些原则来选择方法:

个人小项目,时间紧,随便折腾——用 Vibe Coding 没问题。反正只有你自己看代码,效率优先。

个人正式项目,打算长期维护——建议用 OpenSpec 或类似的轻量规范。每次改功能先写一个 proposal,实现完归档。这样几个月后你还能知道当时为什么这么做。

团队项目,多人协作——必须用规范驱动开发。可以用 Spec Kit,可以用 Kiro,也可以自己定义规范模板。关键是团队达成一致,每个人都遵守。

企业级项目,需要合规和审计——全流程规范驱动。需求、设计、任务、代码都有文档可追溯。这时候 Spec Kit 或者 Kiro 的价值最大。

还有一个建议:从小处开始。不要一上来就想着改变整个团队的工作流。先自己试试,在一个小功能上用规范驱动开发的方式做一遍,体验差异,再考虑推广。


下一章讲如何在真实项目中落地 AI 编程,包括遇到的坑和解决方案。工具选好了,方法论懂了,接下来是实战。