yeah. My rule of thumb has been to use LLMs to compress the distance between an idea and working software, but never outsource my understanding of how each piece or component fits together.
For example, when I'm working on something, I intentionally don't use an LLM to break down the problem. I take the time doing that myself, including defining the implementation details and my style preferences. Then I use plan mode in tools like Cursor or Claude to generate a detailed technical design.
Before I hit build, I read through the plan carefully and make sure I understand every part of it. I want the file structure, component boundaries, and overall organization to be exactly how I intend them to be.
Ultimately, the goal is to make sure "I own the solution, not the AI"
For example, when I'm working on something, I intentionally don't use an LLM to break down the problem. I take the time doing that myself, including defining the implementation details and my style preferences. Then I use plan mode in tools like Cursor or Claude to generate a detailed technical design.
Before I hit build, I read through the plan carefully and make sure I understand every part of it. I want the file structure, component boundaries, and overall organization to be exactly how I intend them to be.
Ultimately, the goal is to make sure "I own the solution, not the AI"