自上次研究 Claude Code 後,開始接觸到 AI 代理人程式開發,想當然耳也會想評估一下另一個鼎鼎有名的 GitHub Copilot,試著找出兩者的差異,而這本由陳會安撰寫的《Python 修仙之路:VS Code + GitHub Copilot》就成了我開箱 GitHub Copilot 的入門書。談起陳會安這位作者,真的算是位奇人。我 20 多年前初出社會、謀職時買的第一本電腦書,是講 ASP 網頁程式設計,作者就是他!除了是長青的資訊技術作者外,看到他掛名的著作能從 C 語言跨到 AI 工具,真心覺得這就是十八般武藝都略懂略懂的人吧。
就我來看,這本書的定位不太討喜。對於真正的程式開發者來說,介紹的內容都是日常工作中必備的基礎知識。從程式語言的基礎、程式碼的重構與優化、程式除錯以及版本控制,到最後的容器化部署,這些作業即使不使用 GitHub Copilot,對於工程師來說也不費吹灰之力。反過來說,如果書籍的目標讀者是完全不懂程式的門外漢,那這些內容又顯得節奏過於快速,造成知其然而不知其所以然的隱憂。不過,或許這正是 AI 代理人開發時代的特色,讓每個人都能在不了解細節的情況下,做出可以動起來的軟體,一切都像魔法般神奇。
在完讀這本介紹 GitHub Copilot 的作品後,大略體驗了這個工具的基本功能和能耐。感覺它是從原本程式開發 AI 助理的定位逐步調整,慢慢朝向 Claude Code 的AI 代理人模式前進。換言之,它本身具有能加速程式人員撰寫程式碼的傳統功能,再疊加上現在最熱門的下指令讓 AI 寫程式的技術,變成魚與熊掌都要的 AI 產品。可惜的是,畢竟 AI 代理的思維,感覺是 GitHub Copilot 後來才添加進去的,所以在整體的使用細節和功能完整性上,是落後 Claude Code 的解決方案。如果用汽車來比喻,它就像是油電混合車,既不像傳統方式要求工程師純手動、一行行刻出程式碼,又不完全讓 AI 代理人從無到有產製出近乎成品的專案。對於 IT 部門管理者和第一線的開發者來說,拿捏在什麼情境下要使用自動完成、什麼時候又該尋求 AI 代理,以及在工程師與這類 AI 工具協同工作的情況下,未來程式碼的質、量管理要如何調整、考核,這些都是要額外討論、思考的議題。
After researching Claude Code, I started getting into AI agent-based software development. Naturally, I also wanted to take a look at another famous tool, GitHub Copilot, and see what the differences are between the two. This book, *Python Cultivation: VS Code + GitHub Copilot*, written by Chen Hui-An, became my beginner's guide to trying out GitHub Copilot. Speaking of Chen Hui-An, I really think he is quite an interesting author. More than 20 years ago, when I first entered the workforce and was looking for my first job, the first computer book I bought was about ASP web programming. And guess what? He was the author! Besides being a long-time writer of IT books, I was also surprised to see that his books cover such a wide range of topics, from C language to AI tools. Honestly, I feel that he is the kind of person who knows a little about almost everything.
In my opinion, the positioning of this book is not very appealing. For real software developers, most of the topics covered in the book are basic knowledge that we need in our daily work. From programming language basics, code refactoring and optimization, debugging, and version control, all the way to container deployment, none of these tasks are very difficult for engineers, even without GitHub Copilot. On the other hand, if the target readers are complete beginners who know nothing about programming, the book moves too quickly. There is a risk that readers may learn how to do things without really understanding why they work. However, perhaps this is exactly what makes the age of AI agent-based development special. It allows almost anyone to build software that actually works without understanding all the details. Everything feels almost like magic.
After finishing this book about GitHub Copilot, I got a basic idea of what this tool can do and experienced some of its main features. I feel that it is gradually moving away from its original role as an "AI assistant for software development" and moving toward the AI agent approach of Claude Code. In other words, it still has its traditional features for helping programmers write code faster, while also adding the increasingly popular approach of simply giving AI instructions and letting it write the code. It is trying to have the best of both worlds. The problem is that, since the AI agent approach seems to have been added to GitHub Copilot later, it still falls behind Claude Code in terms of overall usability and completeness. If I use a car as an example, GitHub Copilot is like a hybrid car. It is neither like the traditional approach, where engineers manually write every line of code, nor does it completely let an AI agent build an almost finished project from scratch. For IT managers and frontline developers, deciding when to use traditional code completion and when to turn to an AI agent is now another issue to think about. At the same time, when engineers work together with these AI tools, we also need to rethink how the quality and quantity of future code should be managed and evaluated. These are all issues that require further discussion and thought.









