Document Once, Let AI Do the Rest

The key point is that things have become much simpler than they used to be. The first time I do something, I do it myself and document the process carefully in a markdown (.md) file. I write explicit instructions, capture best practices, and include concrete examples. That initial effort isn’t wasted time—it’s how I build the understanding needed to tell whether an AI agent’s output is good or bad.

Once that foundation is in place, everything changes. The next time I work on the same problem, I simply feed the markdown file to tools like Claude or Cursor and let them handle the execution. The AI does the work, but I stay in control because I understand the domain well enough to evaluate the results.

I don’t know about you, but I read a lot, and that habit probably played a big role in shaping this approach. It’s also what led me to build ider-UI, a component library for our studio. Unlike traditional libraries, it’s not made of code. Instead, it’s essentially a collection of markdown files documenting best practices across different design domains, with clear instructions and concrete examples.

In a way, it’s a knowledge-first system. You do the thinking once, write it down clearly, and after that, AI becomes a force multiplier rather than a black box.