💻 Daily AI Coding Setup
- IDE: Cursor / Windsurf / VS Code + Cline
- CLI: Claude Code / Aider
- Browser AI: ChatGPT / Claude (architecture chat)
- Local: Ollama (offline backup)
📋 Project Setup ⭐
.cursorrules / .windsurfrules
# Project Conventions
- Language: TypeScript strict
- Style: Airbnb + Prettier
- Test: Jest, AAA pattern
- Avoid: any type, jQuery
- Pattern: functional components, hooks
- API: REST with Zod validation
# When making changes
1. Always run tests
2. Update related docs
3. No console.log in commits
4. Conventional commit messages
🎯 Task-Based Workflow
Feature Implementation
1. Describe feature (Markdown)
2. AI: Architecture proposal
3. Human: Review, approve
4. AI: Skeleton files (no logic)
5. AI: Implement file by file
6. Human: Test each
7. AI: Generate tests
8. Human: Run, fix
9. AI: PR description + commit
Bug Fix
1. Reproduce + paste error
2. AI: Hypothesize root cause
3. AI: Add diagnostic logging
4. Run, gather evidence
5. AI: Propose fix
6. Human: Review, apply
7. AI: Regression test
8. Verify
Refactor
1. Identify code smell
2. AI: Multiple refactor proposal
3. Compare trade-offs
4. AI: Stepwise refactor (small commit)
5. Run tests after each step
6. Verify behavior unchanged
🧠 Prompt Pattern for Coding
"Plan-then-Code"
"Don't write code yet. First:
1. Restate the requirement
2. List clarifying questions
3. Propose architecture
4. List files to create/modify
5. Identify risks
Wait for my approval before coding."
"Test-First"
"Before implementation:
1. Write failing tests covering all cases
2. Wait for me to approve test cases
3. Then implement to pass all tests"
"Conservative"
"Make minimum change to fix issue.
Don't refactor unrelated code.
Don't add features not requested.
Preserve existing patterns."
🔍 Context Management
- Cursor:
@file, @folder, @docs
- Windsurf: Cascade reads relevant files
- Claude Code: project-wide aware
- Big repo → focus on subdirectory
- Past PR-গুলো reference হিসেবে
🎯 Effective Prompt Tips
- File-এ
// TODO: implement X comment → AI fill
- Function signature + docstring → body generate
- Test fail output paste → fix
- Stack trace + relevant file → debug
- "Like the pattern in @file.ts"
⚠️ AI Coding Mistake
- বুঝা ছাড়া accept
- Hallucinated package install
- Outdated API use
- Test না লিখে blind trust
- Architecture decision বিনা reflection
- Security vulnerability miss
- Performance issue overlook
📊 Productivity Metric
Before AI:
- 1 feature/day
- 30% time debugging
- 2 PR/day
With AI (well used):
- 3-5 feature/day
- 40% time on architecture/review
- 5-8 PR/day
- More test coverage
- Better documentation
🎓 Learning while Coding
- "Explain why this approach" বলুন
- Alternative চান comparison-সহ
- "Teach me the pattern"
- Code review feedback হিসেবে use
- Unfamiliar library → AI tutorial
👥 Team Adoption
- Shared
.cursorrules in repo
- Prompt library team-wide
- Best practice doc
- Pair-programming-with-AI session
- Code review expectation update
- Quality bar maintain
🏆 Senior vs Junior Use
| Junior | Senior |
|---|
| Code generate | Architecture brainstorm |
| Syntax help | Trade-off analysis |
| Bug fix | Code review |
| Test write | Edge case identify |
| Doc lookup | Library evaluate |
💡 Golden Rule ⭐
AI = junior pair programmer। Strategy আপনার, execution accelerate। Code merge-এর আগে সবসময় বুঝুন।