iTranslated by AI
2025 Engineering Year in Review
As is tradition, I'm writing a reflection on my engineering activities this year. This year, in particular, felt like about three years' worth of events because major changes were happening every month in the AI coding space.
Last year's (2024) reflection can be found here.
Speaking and Writing
As I mentioned in last year's reflection, I had been working since around October 2024 with a team focused on utilizing generative AI within the company, and I officially transferred there this January. The team is called the "AI Yatteiki (Getting It Done) Team" [1]. Our team's primary duties involve supporting the introduction of AI-related tools and R&D activities, so I think it's equivalent to what other companies might call an "AI Promotion Department."
Since I've specialized in CI/CD and GitHub Actions for the past several years, this marks my first time changing my area of expertise in quite a while. Because of that, I didn't think I'd have many opportunities to present externally this year, but as it turned out, I had quite a few.
First, I held a Dify workshop for internal new graduate training to utilize generative AI and presented on case studies of GitHub Copilot usage at an internal study session. Since Cybozu releases its training materials externally every year, you can view the materials at the links below. Perhaps due to the current trends, it seems they went a bit viral shortly after being published.
I was accepted for a CfP and spoke at TSKaigi 2025, a conference for the TypeScript community, about the AI SDK we use in our internal ChatGPT clone service developed and operated by the AI Yatteiki Team. Looking at the list of talks from last year's TSKaigi, it seemed the organizers were intentionally ensuring diversity in fields, so I deliberately aimed for the AI slot with this CfP. Whether that strategy paid off is just a guess, but it might have been accepted because stories about developing AI apps in languages other than Python were still rare at the time.
From summer onwards, I mainly presented on how to use GitHub Copilot. This year, I began to see a huge number of blogs and presentation slides online about generative AI coding tools like Claude Code. While I had more opportunities to listen to others' presentations at study sessions, I sometimes found it difficult to judge the truth of whether the AI actually behaves that way when explained only with text.
Therefore, I switched to a style where I minimize slides and use a live demo format to show how I actually use GitHub Copilot. This is because I believed that, especially with generative AI, showing it in action is the most convincing approach. The downside of this style is that the presentation record doesn't remain as slides, but I supplemented this by summarizing the scripts of the presentations as blog posts.
My last presentation of the year was at the Cybozu Tech Meetup in November, where I spoke not from the perspective of an AI tool user, but from the perspective of someone supporting AI adoption within a company.
GitHub Stars
As usual, I continued my routine of checking github.blog/changelog every morning and sharing it on X. While updates related to GitHub Actions have truly decreased, updates related to GitHub Copilot occurred almost every day. Since they directly impact my work, I checked the Changelog every day without getting bored at all.
During this time, GitHub staff were looking for people interested in the GitHub Stars program, so I went to talk with them. It was also a period of transition for GitHub Stars in Japan, with events like the GitHub Stars program introduction being held. Because of this, I secretly thought, "Maybe there's a chance for me," and to my surprise, I actually got that chance.
I want to thank those who nominated me for Stars, as well as my previous and current employers for giving me the opportunity to share information externally so freely. I used to jokingly introduce myself as "GitHub's agent," but I've truly become a GitHub agent now lol.
For now, becoming a GitHub Star hasn't changed my activities much, but I'll likely continue focusing on sharing information related to GitHub Copilot through next year and beyond.
Due to the nature of my work, I've tried out all the major AI coding tools besides GitHub Copilot, such as Cline, Cursor, Claude Code, Gemini CLI, Codex CLI, and Antigravity. While Cursor and Claude Code seem to be gaining dominance in the world, as of the end of 2025, the combination I rely on most is VSCode + GitHub Copilot. I'll talk more about that another time.
Donations
Continuing from last year, I’ve kept up my donations to OSS. For my stance on donations, please refer to the 2023 reflection article.
This year as well, I made some donations, mainly to people active in the JS/TS community and tools I use frequently.
OSS Activities
Compared to last year, the density of the green grass (contributions) has increased, but most of it was activity in private repositories. I’ve been continuously improving a hobby project I gave a sneak peek of during my TSKaigi 2025 presentation—a combination of a Kintone-specific Chrome extension and a local LLM—and I spent almost the entire year writing code for that.
I’ve tried out all the major AI coding tools that appeared in 2025 on this private repository. Since it directly relates to my work, and testing new tools is simply fun, I spent most of my time this year exploring AI coding tools rather than on traditional OSS activities.
As a publicly shareable result, I summarized a method for reusing GitHub Copilot settings for other AI coding tools like Claude Code at the end of the year. This might be useful for people like me who switch between multiple AI coding tools.
I’m not sure what 2026 will bring, but it feels like the evolution in the AI coding tool space has calmed down slightly. If I find some breathing room or get bored with following the evolution, I might want to restart work on something like Kesin11/actions-timeline.
I have notes on various things I’ve wanted to do for a long time, such as displaying Composite Actions on the timeline. Since I’ve accumulated know-how on GitHub Copilot and Claude Code, it shouldn’t take much time to make it happen as long as I have the motivation (though the theory that the person who needs to get motivated is the biggest bottleneck holds strong).
-
This is the official name, so it's even printed on my business cards. ↩︎
Discussion