メカタフ

Hello World

Writing Software in 2025

Most linkedin posts I see speak about “vibe coding” - but my friend Zachry Huang prefers the term “agentic coding”. I agree with his definition: the human designs how the program should behave and the ai implements.

It doesn’t take much to hit the “Aha!” moment where an LLM writes functional code for you and you realize “Hey, I can do this too.” There’s no unseeing it once you experience it. Fire up Replit, Vercel V0, or Claude Artifacts and see how far you can go in a couple of prompts.

The barrier to entry for building has never been lower. Paradoxically, the standard for shipping good product is much higher. Everything standing between you and your vision is:

  • Clear articulation of what you need (always an underrated skill)
  • Choice of inspiration – finding similar projects on GitHub has replaced choosing frameworks. Think designer’s mood board, but in markdown not Figma
  • Tokens – the cost of back-and-forth with your LLM of choice

Now it’s your move.

Introducing AIcademy

A good metaphor for Agentic Coding is that you act like Curator rather than producer. We’re preparing DJ sets, not composing symphonies. Some DJs still beatmatch manually using vinyl. That’s what using Cursor IDE feels like – an overly complicated use of old tools to remix existing patterns. Don’t get me wrong, Cursor and similar IDEs are fantastic products. They’ve pioneered paradigms like “rules” (detailed coding principle prompts) that we all use now. But the Rekordbox for AI-assisted development doesn’t exist yet.

There’s a massive tailwind in 2025 for people willing to tinker and create their own workflows. The tools are there – GitHub Actions, Claude API, conventional commits, semantic versioning. The challenge is orchestrating them into something coherent.

AIcademy.so is a collection of my heuristics and best practices for repository engineering in this new paradigm:

  • Every repository is AI-maintainable by default
  • Documentation serves both human and AI readers
  • Mechanical tasks stay mechanical
  • Context flows seamlessly from issue to implementation to release

The convergent method I’m developing treats repositories as living systems that can be operated on by both humans and AI agents. The documentation at aicademy.so will keep evolving as I learn.

Stay tuned! T