Skip to main content

Development Process

Development with AI​

AI is actively utilized throughout the development process.

  1. As of August 2025, Claude Code performs the best.
    • Initial development and code structure design are done together with Claude Code.
  2. During early development, I give AI significant autonomy and delegate tasks. The goal is to build features very quickly. Once the structure is somewhat established, I impose constraints and maintain the codebase while working.
  3. After the structure is relatively well-established, I also use Cursor and Gemini CLI.
    • Once the folder structure is well-organized, even lower-performance AI can handle smaller tasks.
    • To avoid relying solely on Claude Code and to save costs.
    • I'm currently exploring better ways to use Gemini CLI.
  4. For FE, I utilize free plans of services like v0 and Lovable.
    • Not just for code, but to solve UI design challenges.
    • I'm researching other useful AI tools.

Test Code​

I work with at least a minimal test code and process in place. Especially with AI, accessibility to testing has improved.

  1. For backend, I parse test scenarios written in markdown documents to execute tests.
    • I was inspired by the test code in ty.
    • Currently, I have it set up to perform E2E tests and type validation in Go.
  2. For frontend, I utilize Storybook and Playwright.

Git + History management​

  1. During initial development, I skip much of the process.
    • I found that rules I create are easily broken, and being too caught up in process actually reduces productivity.
  2. After the code stabilizes, I move the code to GitHub and document important details in this documentation and GitHub Discussions.
    • GitHub Issues are more common, but I personally prefer Discussions. 😊
      • Clearly, GitHub Discussions still lack support. Some open source projects have abandoned Discussions in favor of Issues.1
    • Rather than writing code details extensively, I focus on documenting important concepts, ideas, and considerations.
    • However, there's an issue with the CLI not supporting Discussions, so I'm considering solutions.2
  3. I use Linear for branch and issue management.
    • With the free plan, if you consistently clean up tickets, you can effectively use it unlimitedly.
    • In my experience, there haven't been many cases where I needed to review history in detail for personal projects.
    • I aim to use PR conventions. The format is type: [ticket] description.

Productivity​

  1. I actively utilize MCP and CLI.
    • It increases AI productivity, and creating CLIs for frequently performed tasks reduces unnecessary resource waste.
    • I create and use personal MCP and CLI to be usable anywhere and to improve development skills.
  2. I always consider automation and generalization.
  3. Occasionally, I manually organize and refactor.
    • While it may seem to reduce productivity in the short term, building a more robust structure increases productivity later.
    • It's not a one-time thing; it needs to be done consistently.
  4. I aim for multitasking.
    • With AI tools, it's become possible to handle multiple tasks at once, much like running auto-hunt in a game.
    • I reduce directly handling simple tasks and instead perform higher-level work like idea planning, architecture design, and reviews.
    • However, this should be adjusted appropriately considering personal preferences.

Footnotes​

  1. https://github.com/backstage/backstage/issues/20985 ↩

  2. https://github.com/cli/cli/discussions/4212 ↩