🚀 Vibe-Coding Workflow: AI-Powered MVP Development

Build an MVP in hours, not months – guided by the latest AI coding agents

---

🎯 What This Does

Transform any app idea into working code through 5 AI-powered stages using the latest 2025 AI models:

#	Stage	Goal	Output
1️⃣	Research	Validate market & tech landscape	Research findings
2️⃣	Define	Clarify product scope	PRD (Product Requirements)
3️⃣	Design	Decide how to build	Technical Design doc
4️⃣	Generate AI Agent Instructions	Convert docs into agent blueprints	AGENTS.md + tool-specific config files
5️⃣	Build	Generate & test code	Working MVP

---

🏃 Quick Start

✨ New: Automated Web App Workflow ✨

Skip the manual copy-pasting!
We now have an interactive web app that automates the entire process for you.

⚡ Manual Workflow Guide (The Classic Way)

Step	What You Do	Time	Result
📚	Copy prompts → Answer questions	20 min	Research doc
📝	Define your app idea	15 min	PRD doc
🏗️	Choose technical approach	15 min	Tech Design doc
🤖	Generate AI instructions	10 min	AGENTS.md + agent config
💻	Tell the AI agent: "Read AGENTS.md and build the MVP"	1-3 hrs	Working MVP!

---

✅ Prerequisites

🤖 AI Platform (Required - Choose One)

Best Free Options

• AI Studio ⭐ - Gemini 3 Pro free tier with the full 1,048,576-token context (see model details)
• Claude.ai - Claude Sonnet 4.5 with 200K default / 1M beta context (see current versions)
• ChatGPT - ChatGPT 5.1 via the Responses interface (see OpenAI docs)

Premium Options (Better for Complete Apps)

• Claude Pro - Sonnet 4.5 with higher rate limits and 1M-token projects (pricing)
• Gemini Advanced - Gemini 3 Pro/Flash Ultra access with priority throughput (pricing)
• ChatGPT Plus - ChatGPT 5.1 with higher allocations (pricing)

💻 AI Coding Agent/IDE (Required - Choose One)

Terminal-Based Agents (Advanced)

• Claude Code ⭐ - Project-aware CLI with session memory.

  curl -fsSL https://claude.ai/install.sh | bash
  claude init # Generates CLAUDE.md; import AGENTS.md via @AGENTS.md

• Gemini CLI - Free & open source with Gemini 3 Pro.

  npm install -g @google/gemini-cli
  gemini login
  # Uses GEMINI.md; import AGENTS.md via @AGENTS.md

• Aider - CLI tool, requires explicit context loading.

  python -m pip install aider-install
  aider --read AGENTS.md

• GitHub Copilot CLI - Terminal interface for Copilot.

  npm install -g @github/copilot
  copilot --context AGENTS.md

• OpenAI Codex CLI - An open-source, terminal-based coding agent (TUI) that can read, edit, and run code in your repo. (docs)

  npm i -g @openai/codex
  codex exec "Summarize AGENTS.md"

Async/Cloud Agents

• Jules by Google ⭐ - Vertex AI-connected async agent.
• GitHub Copilot Agent - VS Code extension; enable chat.useAgentsMdFile to read AGENTS.md.

IDE-Based Tools (Beginner Friendly)

• Cursor ⭐ - AI editor that automatically reads AGENTS.md.
• Windsurf ⭐ - IDE by Codeium; reads AGENTS.md and .windsurf/rules.
• Cline - VS Code extension; reads AGENTS.md and .clinerules.
• Google Antigravity - Agent-first IDE; manually add AGENTS.md content to Knowledge Base.

No-Code Platforms (Easiest)

• Bolt.new ⭐ - Instant Next.js/Supabase apps.
• Lovable - Fullstack builder; add AGENTS.md to "Custom Knowledge".
• v0 by Vercel - UI composition; add AGENTS.md to "Project Instructions".

🛠 Basic Requirements

• Any modern browser
• 2-4 hours of time
• Basic computer skills (no coding required!)
• Optional: Node.js 22+ for terminal tools

---

📋 The 5-Step Workflow

1️⃣ Deep Research 🔍

Validate your idea with AI-powered market research • 20-30 min • Creates research-*.txt

What this does: Analyzes market opportunity, competitors, and technical feasibility using the latest AI models.

How it works:

1. Copy the entire part1-deepresearch.md file
2. Paste into AI Studio (for Gemini 3 Pro), Claude.ai (for Claude Sonnet 4.5), or ChatGPT (for ChatGPT 5.1)
3. Answer 5-6 questions tailored to your experience level
4. AI generates comprehensive research with:
   • Market analysis & size
   • Competitor breakdown
   • Technical recommendations
   • Cost estimates
5. Save output as research-[YourAppName].txt

💡 Pro Tip: Use Gemini 3 Pro for better research (1,048,576-token context window).

2️⃣ Product Requirements (PRD) 📝

Define exactly what you're building • 15-20 min • Creates PRD-*.md

What this does: Transforms your idea into clear, actionable product specifications.

How it works:

1. Copy part2-prd-mvp.md into a new AI chat
2. Attach your research findings when prompted
3. Answer questions about:
   • Core features (3-5 must-haves)
   • Target users
   • Success metrics
   • UI/UX vision
4. AI creates professional PRD document
5. Save as PRD-[YourAppName]-MVP.md

3️⃣ Technical Design 🏗️

Plan the technical architecture • 15-20 min • Creates TechDesign-*.md

What this does: Decides the best tech stack and implementation approach for 2025.

How it works:

1. Copy part3-tech-design-mvp.md into a new AI chat
2. Attach your PRD (required) and research (optional)
3. Answer questions about:
   • Platform (web/mobile/desktop)
   • Complexity tolerance
   • Budget constraints
   • Timeline
4. AI recommends optimal stack from:
   • No-code: Bolt.new, Lovable, Bubble
   • Low-code: Next.js + Supabase
   • Full-code: Your preferred framework
5. Save as TechDesign-[YourAppName]-MVP.md

4️⃣ Generate AI Agent Instructions 🤖

Create blueprints for your AI coding assistant • 5-10 min • Creates AGENTS.md + agent configs

What this does: Converts all docs into step-by-step coding instructions for AI agents.

How it works:

1. Copy part4-notes-for-agent.md into a new AI chat
2. Attach PRD and Technical Design documents
3. AI generates:
   • AGENTS.md - Universal instructions
   • Tool-specific configs (based on your choice):
     • CLAUDE.md for Claude Code
     • GEMINI.md for Gemini CLI & Antigravity
     • .aider.conf.yml for Aider
     • .clinerules for Cline
     • .cursorrules for Cursor
     • .windsurfrules for Windsurf
4. Save all files in your project root

5️⃣ Build with AI Agent 💻

Let AI build your MVP • 1-3 hrs • Creates working application

Setup by Tool Type

Terminal Agents (Claude Code, Gemini CLI)

# Claude Code
npm install -g @anthropic-ai/claude-code
cd your-project
claude init
# Add CLAUDE.md to project root
claude "Read CLAUDE.md and AGENTS.md, then build the MVP"

# Gemini CLI
npm install -g @google-gemini/cli
gemini login
# Add GEMINI.md to project root
gemini "Read GEMINI.md and AGENTS.md, then implement"

IDE Tools (Cursor, Windsurf)

1. Open your project folder in the IDE
2. Add configuration file:
   • Cursor: .cursorrules or .cursor/rules.mdc
   • Windsurf: .windsurfrules
3. Start with: "Read AGENTS.md and build the MVP step by step"

No-Code Platforms (Bolt.new, Lovable)

1. Go to platform website
2. Paste your PRD content
3. Say: "Build this MVP following the specifications"
4. Deploy instantly with one click

Essential Prompts for Building

Starting prompts by experience level:

Level	First Prompt
Beginner	"I'm new to coding. Read AGENTS.md and guide me step-by-step to build this MVP. Explain what you're doing."
Intermediate	"Read AGENTS.md and the docs folder. Build the core features first, test, then add polish."
Developer	"Review AGENTS.md and architecture. Implement Phase 1 with proper patterns and test coverage."

Follow-up prompts for all levels:

• "Show me the current progress vs requirements"
• "Test [feature] and fix any issues"
• "Add error handling and edge cases"
• "Generate README with setup instructions"
• "Prepare for deployment to [platform]"

---

📁 Final Project Structure

your-app/
├── docs/
│   ├── research-YourApp.txt
│   ├── PRD-YourApp-MVP.md
│   └── TechDesign-YourApp-MVP.md
├── AGENTS.md             # Universal AI instructions
├── CLAUDE.md             # Claude Code config (if using)
├── GEMINI.md             # Gemini CLI config (if using)
├── GEMINI.md             # Antigravity config (if using)
├── .aider.conf.yml       # Aider config (if using)
├── .clinerules           # Cline config (if using)
├── .cursorrules          # Cursor config (if using)
├── .windsurfrules        # Windsurf config (if using)
├── README.md             # Setup instructions (AI-generated)
├── .env.example          # Environment variables
└── src/                  # Your application code

---

🆕 What's New in 2025

Latest AI Models & Capabilities

Model Updates

• Claude Sonnet 4.5 – September 2025 release with 200K default / 1M beta context and top-tier architecture reasoning
• Claude Opus (2025 snapshot) – Handles massive monorepos and multi-hour planning sessions with improved tool use
• Gemini 3 Pro – 1,048,576-token input and 65,536-token output windows for deep research and synthesis
• ChatGPT 5.1 – Responses API with adjustable reasoning effort, faster tool orchestration, and lower latency

New Tools

• Claude Code - Anthropic's project-aware terminal agent with session memory and automated test orchestration
• Jules - Google's asynchronous coding agent that can work independently across Vertex AI projects
• Gemini CLI - Open-source CLI with direct Gemini 3 Pro/Flash access, live context streaming, and MCP integrations
• GitHub Copilot Agent Mode - Workspace automation for scoped plans, PRs, and deployment checklists
• OpenAI CLI (Responses API) - Terminal workflow for ChatGPT 5.1 tool-calling pipelines

Platform Updates

• Cursor - Version 2.1 adds improved plan mode, AI code reviews, and instant grep.
• Windsurf - Now supports AGENTS.md automatically for project context.
• Cline - Manages context in three layers (immediate, project, persistent) with auto-truncation.
• GitHub Copilot CLI - Version 0.0.365 adds --silent option for cleaner output.
• Bolt.new - Instant Next.js/Supabase deployments with scheduled automations.
• Lovable - "Custom Knowledge" tab for project-wide context.

---

💡 Pro Tips for 2025

Tool Selection Guide

Persona	Best Tool Stack	Why it fits	What to watch	Setup time
Complete beginner	Lovable • Bolt.new	Paste your idea, get a hosted app. Use "Custom Knowledge" for guidelines.	Daily credit caps; hosted code.	2–5 min
Learning hobbyist	Copilot Agent • Cline	Copilot edits files; Cline manages context layers for you.	Copilot requires subscription; Cline needs API key.	5–10 min
Experienced developer	Cursor • Windsurf	Cursor 2.1 has instant grep; Windsurf reads AGENTS.md automatically.	Usage-based credits; new IDE workflows.	5–10 min
Budget-limited builder	Cline • Gemini CLI	Free to install; Gemini CLI uses free tier API.	Less hand-holding; requires more prompting.	5–10 min
Need-it-today founder	Lovable • v0	Fastest path to MVP. Use "Project Instructions" for v0 context.	Watch credit burn; plan for security review.	2–5 min
Mobile-first product team	v0 • Flutter + Gemini	v0 sketches mobile UI; Gemini CLI helps with native code.	Mobile features are evolving.	45–60 min
Complex logic engineer	Claude Code • Windsurf	Claude Code has session memory; Windsurf handles deep context.	Claude Code is CLI-based.	10–15 min
Security/compliance lead	Cline (client-side)	All code stays local; persistent context is managed locally.	You define the policy.	10–20 min
Offline/privacy-focused dev	Gemini CLI • Cline	Works with local models (Ollama) or private keys.	Offline mode depends on hardware.	10–30 min
Open-source maintainer	Cline • Aider	Aider explicitly reads AGENTS.md; Cline merges rule files.	Terminal-first or VS Code extension.	5–10 min

Quick Picks (Plain English)

• Need an MVP tonight? Use Lovable or Bolt.new, then plan a follow-up pass for polish and security.
• Working inside an editor? Cursor or Copilot Agent can code for you—open Cline when you want to double-check every change.
• Huge legacy repo? Bring in Claude Code or Windsurf so the agent “remembers” more of your code at once.
• No budget? Stick with Cline + Gemini CLI and the generous free request limits.
• Mobile-first? Preview flows in v0 Mobile or build native screens with Flutter + Gemini’s suggestions.

Pricing Guardrails (Nov 2025)

• Free forever: Gemini Code Assist (6K code reqs/day), Copilot Free (2K completions + 50 chats), Windsurf Free (25 prompts).
• Best under $20: Windsurf Pro $15/mo (500 credits), Copilot Pro $10/mo (Agent Mode), Cursor Pro $20/mo (credit pool + parallel agents).
• Power tiers: Copilot Pro+ $39/mo (1.5K premium calls), Cursor Ultra $200/mo (20× credits), Bolt Pro 200 $200/mo (120M tokens).

MCP Watch

• Model Context Protocol (MCP) is a shared standard that lets your AI talk to extra tools (think CI pipelines, databases, ticket queues).
• Cline, Cursor, Claude Code, and Gemini CLI all plug into MCP, but treat those connections like production credentials.

Caution Flags

• Claude consumer accounts now offer opt-in data sharing; stick to org workspaces or API keys for sensitive repos.
• Replit Agent 3 can run 200-minute autonomous loops—enable guardrails so retries cannot nuke production data.

When Not to Use These Tools

• Native mobile or hardware builds: prefer Flutter + traditional toolchains with AI pairing.
• Regulated workloads needing SOC2/FedRAMP/HIPAA: lean on Copilot Enterprise, Cline with strict policies, or self-hosted stacks.
• Safety-critical or real-time systems: require deterministic, human-led engineering.
• Fundamentals practice: hand-code portions to avoid skipping core concepts.

Common Pitfalls & Solutions

❌ Pitfall	✅ Solution
Skipping discovery work	Run the Part I research prompt first so the PRD and tech design aren’t guesses
Letting agents ship code alone	Ask agents to show their plan, review the diff, and run tests before anything merges
Sticker-shock bills	Check credit dashboards weekly and keep a backup stack (Cline + Gemini) that stays free
Publishing auto-generated UIs without checks	Test accessibility, security, and performance before launch day
Building sensitive apps with personal accounts	Use business plans or API workspaces, switch off data sharing, and keep secrets in vaults
Forcing one tool to do everything	Mix and match (IDE + terminal + builder) so each tool covers what it does best

---

🚨 Troubleshooting

Quick Fixes for Common Issues

Problem	Solution
"AI ignores my documents"	Start with: "First read AGENTS.md, PRD, and TechDesign. Summarize key requirements before coding."
"Code doesn't match PRD"	Say: "Re-read the PRD section on [feature], list acceptance criteria, then refactor accordingly."
"AI is overcomplicating"	Add to config: "Prioritize MVP scope. Offer the simplest working implementation before optimizations."
"Lost track of progress"	Ask: "Update the AGENTS.md progress log and map remaining tasks to implementation phases."
"Deployment failing"	Request: "Walk through deployment checklist, verify env vars, then run the platform-specific health command."

---

🤝 Contributing

PRs & issues welcome! Help us improve:

• 🐛 Report issues with prompts
• 💡 Share your success stories
• 🔧 Add new tool configurations
• 📚 Submit example MVPs built with this workflow

---

📜 License

Released under the MIT License.

---

The best time to build your idea was yesterday.
The second best time is now. 🚀

Last updated: November 2025 | Created by the vibe-coding community