Hi, I'm Hattori from coiai. This time I want to share the effective ways of using AI tools, and the shifts in development style, that I've noticed after trying out a bunch of different AI tools this week .
1. Splitting coding duties between "Cursor" and "Claude"
The two coding LLMs I mainly used this week were Cursor and Claude .
- Cursor's evolution: its recent "Composer" model is extremely capable . Set up a documentation directory with a requirements file like "Requirement.md," run it in plan mode, and it writes code step by step following the requirements .
- Splitting by language: I've noticed each LLM has its own strengths and weaknesses depending on the language .
- Cursor (Next.js / TypeScript): extremely strong with web-oriented languages, but it tends to produce errors in Swift .
- Claude (Swift / Java): delivers high accuracy in Swift and Java, exactly where Cursor struggles .
- Gemini: notably strong with Python .
I used to rely on the autonomous AI engineer "Devin," but these days I've mostly switched over to Cursor for the better balance of cost-performance and accuracy .
2. Blazing-fast slide generation with "Marp" × "LLM"
Since LLMs are so good at handling text, document creation can get dramatically more efficient with the right approach .
My favorite discovery lately is combining Marp, which converts Markdown into slides, with an LLM . Just have the AI write a requirements document in Markdown, then format it into Marp, and a several-dozen-page proposal or spec document is done in an instant . Gemini's canvas feature is great too, but for generating large slide decks in one go, this "manage everything as text" approach is extremely powerful .
3. Networking knowledge with the note-taking tool "Obsidian"
I'm also making use of Obsidian, another note-taking tool that manages everything in Markdown . By linking notes together with hashtags and visualizing the connections, you can see where your own interests are gravitating, almost like looking at a "city" of connected ideas . It pairs well with AI too, making it ideal for the work of connecting knowledge together .
4. The engineer's joy is shifting from "writing" to "making"
AI used to give you the feeling of "building something together," but now it's entering a stage where it "does the whole thing completely" .
Because of this, the pure joy of programming itself might fade a little, but I feel that in its place, **the joy of building things at breakneck speed is what's taking over . Going forward, the emphasis will keep shifting away from "writing code" and toward "requirements definition" — correctly instructing AI on what to build**, and that shift will only get bigger .
A note from us Alongside XR development and web app development, coiai also builds "on-premise AI" that keeps the risk of confidential information leaks to a minimum. If your company wants to safely have AI learn from your internal documents, please get in touch through "coiai.net" .