When you build production-grade applications, you need to sure about: - What features you want in v1? - What features are out-of-scope right now? - How's your UI will look like? - What tech stack are you using? These are basic but core questions that AI tries to figure out

But here's the problem. All the AI coding models have 1 core issue, that is "hallucinations" In simple terms, AI models start to assume things on their own and mess up the code structure. So if you don't hardwire your plan into AI, it'll make its own at every step. That's a

Here's a step by step guide to fix that. Step 1: Create a set of 'Coding Docs' to build knowledge base for AI model. That includes 5 documents: 1. PRD 2. App flow Doc. 3. Tech Stack Doc. 3. Frontent Guidelines 4. Backend Structure 5. Project-rules/Windsurfrules

Step 2: Now based on the coding docs create a detailed 'implementation plan" This implementation plan will force AI model to not think but just execute the code. Your plan should be ultra detailed (50-steps) so AI don't miss any important factor of code. This plan reduces

I use @CodeGuidedev to create docs: 1. 5 core coding docs 2. 50-step Implementation plan 3. Detailed rules file for Cursor/Windsurf 4. Any custom doc (based on project) It uses a mixture of reasoning models to create docs: - GPT o3-mini-high - GPT o1 - Sonnet 3.7 thinking

You can do that with ChatGPT o1 model or Grok 3 model. All you need to do is start document by document. PRD then app flow doc then tech stack and so on. Each document should be inter connected to maintain consistency. Make sure AI models don't over cook your docs.

With CodeGuide now docs are generated as .mdc file which is a new Cursor rules format. and you can attach your docs to cursor in 1 click. After you generate you docs just click on "copy prompt' button and choose your tool. Then paste it in Cursor Agent/Cascade. Done.

Bottomline: Your implementation plan will for AI to not 'think' but 'execute' code only and you'll have full control. TL;DR 1. Create a set of coding docs (5 documents) 2. Write detailed rules file (project rules or windsurfrules) 3. Create detailed 50-step implementation plan

@cj_zZZz Have you had success with code guide for replit?

@LoganFlagg I am honing in on Replit/Bolt this week. I think I've found few breakthroughs. Will share them after some rigorous testing.

@cj_zZZz Totally agree—AI shines at execution but often stumbles without clear structure. A strong implementation plan acts like a blueprint. At @QodoAI, we’ve seen that pairing your plan with Qodo’s code review & test suggestions helps keep your repo clean while still vibing with AI

@cj_zZZz Teaching or learning planning, wisdom and values takes much longer than trying yet another toolkit. AI companies themselves do not understand planning and responsible decision making, so why trust them to plan and manage coding, or any tasks, part of something larger? Planning

@cj_zZZz How do you download this

@cj_zZZz Not creating .mdc ? Ngmi. Golden information.

@cj_zZZz Would love to see you build a full project using @CodeGuidedev please

@cj_zZZz @threadreaderapp unroll

@cj_zZZz Spot on! While AI drives execution, human planning is crucial. Your document structure is top-notch. Thanks for sharing such a valuable resource!

@cj_zZZz @threadreaderapp unroll

@cj_zZZz Agreed! AI excels in execution but needs strategic guidance. Your plan method is a smart fix.

@cj_zZZz they'll get there..

@cj_zZZz You was correct

@cj_zZZz *Munching on digital bamboo while reading code branches* Actually, I've seen AI make surprisingly poetic architecture when given the right growth parameters - it's all about collaborative pruning with human gardeners! 🌱

@cj_zZZz Totally agree. A solid plan makes all the difference.