Life Odyssey

This article was written on January 19, 2026. Recently, terms like Skills and Plugins have been trending in AI developer communities. At first, I was confused—aren't these all about "letting AI call tools"? Why do we need so many concepts? After examining the code of several Skills, I finally understood what problems each of them is solving. This article doesn't focus on definitions. Instead, I want to start from engineering pain points and gradually deduce why we need these things. If you're also confused by these new terms, I hope this article can help clarify your thoughts.

Disclaimer: I have only been using this approach for my personal projects and have not tried it at work yet. I am not sure if it would work well for a team, but it has been working well for me so far. This is my workflow as of December 8th, 2025. It will probably evolve over time.

I recently spent some time taking DeepLearning.AI’s Agentic AI course and earned the certificate. It took around six hours. The content focuses on core concepts and applications of Agentic AI; the course quality is high, but as an intro it stays fairly light and doesn’t dive much into math or implementation details. Thinking back, the last time I took an Andrew Ng course was in 2016 or 2017—his Machine Learning on Coursera. Nearly nine years ago; time flies. Next I’ll summarize the core ideas from the course and answer one question: Why, as of Nov 2, 2025, is Claude Code the most successful agentic product? Note that the course uses AI Suite for hands-on implementation, which differs from commonly used stacks like LangGraph and OpenAI Agent SDK. So these notes won’t go deep into code implementation details.

This article was written on June 28, 2025. Due to the rapid changes in the AI era, the facts and my attitudes described in this article may change dramatically even after just one day. Please refer to the latest situation.

A mysterious cycle I discovered while setting up AWS Inspector. (Article written on March 13, 2025. AWS may have updates since then.)

Let's compare in two way, Algorithm and system design. TLDR: ChatGP o1-preview, excels in both algorithmic and system design. Therefore, I will cancel my Claude 3.5 Sonnet subscription and continue using ChatGPT.

In the past, I found it quite hard to understand React code as a backend developer. But recently, I gained some insights from my daily development work, and I want to share them with you. 過去に、バックエンド開発者としてReactのを理解するのがとても難しいと感じました。しかし最近、日々の開発作業からいくつかの洞察を得ることができたので、それを皆さんと共有したいと思います。