AIコーディング: 「Vibe Coding」からプロフェッショナルへ / AI Coding: From "Vibe Coding" to Professional

本プレゼンテーションでは、Agile/Scrumのマネジメント思考を応用し、Coding AIエージェントの潜在能力を最大限に引き出す方法を探求します。参加者と共に、「Vibe Coding」―試行錯誤でプロトタイプを迅速に作成するが、コードの細部管理にはこだわらない手法―と、「AIを活用したプロフェッショナルプログラミング」―コードの検証、保守、そして最適なソフトウェア品質の保証を追求する手法―の違いを分析します。本セッションは、経験を共有し、意見を交換しながら、AIエージェント時代におけるソフトウェア開発の未来を共に形作る絶好の機会です。

Transcript

1. AI Coding: From "Vibe Coding" to Professional TechBase Vietnam NGUYEN TRONG PHUC
2. • Limited coding time due to support roles • Ideas often stuck at implementation My Story
3. • Enabled complex projects in one week • Made technical exploration accessible • Transformed prototype development speed AI Coding as Game Changer
4. • Fast start, chaotic growth • Context forgetting issues • Maintenance difficulties • Market hype vs. real-world limits AI Coding Concerns & Hype Gap
5. "Building software with LLMs without carefully examining the code it writes" - Simon Willison Reference: Willison, S. (2025, May 15). Building software on top of Large Language Models. Simon Willison's Weblog. https://simonwillison.net/2025/May/15/building-on-llms/ What is Vibe Coding?
6. Professional Approach with AI Trust the Process AI as Teammate
7. • Preparation • Planning • Development • Troubleshooting Practices
8. ~3000 words with diagrams Blog version
9. Preparation
10. • Reasoning AI • Requirements & Design • Internet search AI • Research & Best Practices • Coding AI Agent • Implementation Practice 1: Coordinating Multiple AI
16. • AI as a brainstorming assistant Practice 2: Prepare Detailed Requirements
17. "Preparation is the foundation of success"
18. Planning
19. • Each task ≤ 100-200 lines of code • Clear input/output • Minimal dependency Practice 3: Task Breakdown for AI
20. Sample: "Write tests first, then code, then run tests Focus on one small increment at a time Always explain reasoning before coding Follow TDD: Red → Green → Refactor Use TypeScript for type safety" Practice 4: Setup Implementation Rule
22. Development
23. Identify a small increment Practice 5: Edit-Test Loops
27. • Start new chat when context gets too long • Resync codebase frequently • Git commits as checkpoints Practice 6: Context Management
28. Troubleshooting
29. Level 1: Prompt Engineer 4 Levels of AI Mastery Level 2: AI Developer Level 3: AI Product Manager Level 4: AI Orchestrator
30. "Trust the process"
37. Process-Centric Human-Centric AI-Centric
38. "Trust the process"
39. "AI as teammate"
40. Happy coding! 楽しいコーディングを！
41. ~3000 words with diagrams Blog version