AI 에이전트가 자면서 코딩한다, Ralph Wiggum 기법 실전 가이드
AI 코딩 에이전트를 루프로 실행해 자율적으로 개발하는 Ralph Wiggum 기법. Google 엔지니어 Addy Osmani의 실전 구현 방법을 소개합니다.AI 에이전트가 자면서 코딩한다, Ralph Wiggum 기법 실전 가이드
AI 코딩 에이전트를 루프로 실행해 자율적으로 개발하는 Ralph Wiggum 기법. Google 엔지니어 Addy Osmani의 실전 구현 방법을 소개합니다.The Next Two Years of Software Engineering
https://addyosmani.com/blog/next-two-years/
#HackerNews #SoftwareEngineering #Future #Trends #TechInnovation #AddyOsmani #EngineeringInsights
Quy trình lập trình cùng AI cho năm 2026 từ Addy Osmani (Google):
1. Prototype: Dùng Claude 3.5 Sonnet hoặc GPT-4o để dựng khung nhanh.
2. IDE: Ưu tiên Cursor/Windsurf để tận dụng tính năng Composer/Flow giúp sửa lỗi và refactor theo ngữ cảnh.
3. Review: Dùng AI giải thích các thay đổi phức tạp trong Pull Request.
4. Testing: Tự động hóa việc viết unit test và kịch bản kiểm thử.
Tư duy quan trọng: AI là cộng sự (Copilot), không phải người thay thế.
"It's becoming a pattern: Teams use AI to rapidly build impressive demos. The happy path works beautifully. Investors and social networks are wowed. But when real users start clicking around? That's when things fall apart."
#AddyOsmani, 2024
https://addyo.substack.com/p/the-70-problem-hard-truths-about
It's been a pattern for about 20 years. Just replace "AI" with blockchains, or frameworks, or whatever the new hotness was at the time. I can't find it right now, but @baldur wrote a great piece about this;
"The key insight I've gained from working with these tools is that the future isn't about AI replacing developers - it's about AI becoming an increasingly capable collaborator that can take initiative while still respecting human guidance and expertise."
#AddyOsmani, 2024
https://addyo.substack.com/p/the-70-problem-hard-truths-about
In others words, as @pluralistic's puts it, enabling a human to be a centaur. Rather than being forced into being a reverse-centaur in service to the Trained #MOLE;
Further down, the hype train reaches top speed;
"But look at newer features like Anthropic's computer use in Claude, or Cline's ability to automatically launch browsers and run tests. These aren't just glorified autocomplete - they're actually understanding tasks and taking initiative to solve problems."
#AddyOsmani, 2024
https://addyo.substack.com/p/the-70-problem-hard-truths-about
You don't need a Trained MOLE to automate that. It's a simple linear task that can be automated using standard scripting methods.
(1/2)
"This gap will likely narrow as tools improve. But for now, the most pragmatic approach is to use AI to accelerate learning, not replace it entirely."
#AddyOsmani, 2024
https://addyo.substack.com/p/the-70-problem-hard-truths-about
There are much more effective learning aids already available. Also Osmani sneaks in a huge and highly questionable assumption; that more #MOLE Training will result in improved tools. There's good reason to believe that the current tools are the best that can be produced with generative models.
(1/3)
"The very thing that makes AI coding tools accessible to non-engineers - their ability to handle complexity on your behalf - can actually impede learning. When code just "appears" without you understanding the underlying principles ...
This creates a dependency where you need to keep going back to AI to fix issues, rather than developing the expertise to handle them yourself."
#AddyOsmani, 2024
https://addyo.substack.com/p/the-70-problem-hard-truths-about