A story about taking stock of just 5 technologies I've been interested in lately
In this note, I have written about my son's soccer, the apps I am building, and for a while now, the records of my experiment in 'letting AI handle management decisions' (AI Company Magazine). This time, as a third magazine, I would like to start talking about technology trends.
That said, I didn't want this to be just a summary of tech news. New things appear in tech trends almost every day, and honestly, I can't keep up with all of them. So, basically, I will make these commentary episodes by narrowing them down based on 'whether there is an actual use case for me as an individual developer.' Instead, I plan to post the results of small verifications of the technologies I'm interested in that I (or my development team) have actually tried, at a pace of about once a week. The goal of this magazine is not just to read about them, but to show the results of actually trying them out.
This first episode is a commentary. Out of the technologies I've been interested in lately, I've taken stock of five that I feel are particularly relevant to my product creation.
MCP (Model Context Protocol) has become more 'standard' than I thought
MCP, which appeared in 2024 as a common standard for connecting AI agents to your own tools and data, has reached a point where not only Claude but also major AI clients like ChatGPT and Gemini support it. The management has also been transferred from Anthropic alone to a neutral organization, and it is becoming positioned as an industry standard rather than a mechanism exclusive to some AIs.
The use case for me is simple: if I create one MCP server for my product (for example, the data of the app I'm building), it can be used in the same way from any AI client. For an individual developer, the cost-effectiveness of 'making it once and reusing it in various places' is quite important, so I would like to try this out myself in the near future.
With Claude Agent Skills, it seems we can reduce the problem of 'copy-pasting the same instructions every time'
I am also interested in 'Agent Skills,' which appeared as a feature of Claude Code. It is a mechanism that allows you to collect instruction manuals, scripts, and templates in one folder and have the AI read them only when necessary. Moreover, this itself is published as an open standard, and it is spreading in a way that it can be used with tools other than Claude.
When doing individual development, I often notice myself rewriting the same procedures into prompts every time I do a certain task. If I can write that down once as SKILL.md and reuse it, it seems like it will be subtly effective. Actually, the mechanism I am currently experimenting with, where I 'assign departments to AI and have them run like a company,' is based on a similar idea in the sense that each department has a file with rules, and I feel that this is a technology that is a direct extension of that.
AI coding agents have changed from 'completion' to 'entrustment'
This might be the change I feel the most myself. Until a little while ago, AI coding tools were centered on code completion and suggestions, but they are changing to an agent type that autonomously handles everything from planning and implementation to testing when you give them instructions. I saw a survey that said the percentage of developers using tools like Claude Code has also grown significantly over the past year.
I also feel that recently, I spend more time 'instructing what to build and reviewing what comes out' than 'writing code line by line.' For me, aiming to release 30 products a year, this change is quite significant in terms of handling volume.
Multi-agent orchestration—actually, this was my company itself
The 'multi-agent' configuration, where multiple AI agents are assigned roles and made to cooperate, also seems to have spread rapidly since around last year. There is a survey that says companies are running an average of a dozen or so AI agents.
Honestly, I laughed when I looked into this, but I realized that the experiment I recently started—'assigning departments to AI agents like secretary, management strategy, business development, marketing, etc., and having them run like a company' (the story I'm writing in the AI Company Magazine)—was just doing this trend at the scale of individual development. It's a story where I went to look up trends, only to find that I was already a practical example of them.
Vue 3.6's Vapor Mode seems to make existing code faster without rewriting it
One story about Vue.js, which I use. The new compilation method 'Vapor Mode' introduced in Vue 3.6 became feature complete in 2026, and there are reports that it runs quite fast because it does not go through the Virtual DOM. Moreover, it is appreciated that existing Composition API code can be left as is, and it seems it can be tried little by little on a component-by-component basis.
Honestly, I think this is the type of story where you won't get a real feel for it until you actually try it on a part of your own app and compare the speed. So, I'm thinking of making it my next verification candidate right away.
Summary and next episode preview
When I line them up like this, there are a lot of stories related to AI agents. Since I am currently in the middle of an experiment where I leave various things to AI agents, it might be natural.
For now, this ends with just being "interested," but in this magazine, I plan to post the results of actually verifying some of these at a pace of about once a week. Honestly, I don't know if it will go well, but I will record everything as it happens, including that. Look forward to seeing which one I try out next time and beyond.
#AIUtilization #PersonalDevelopment #TechTrends #ClaudeCode #MCP #MultiAgent #Vue
