So I have been using Cursor for coding in my office and personal work, and I have used it extensively. Interestingly, my longest streak last year was 62 days. Has productivity increased? Yes. Have I learned to use AI efficiently? Yes. Have I mastered the art of commanding AI to write code? Yes.
However, I would list out the rules and some of the best ways I feel about writing code with AI:
Learn the existing code before making even a single line of change manually.
Find the exact changes you wish to make, and be clear about the requirement.
If it’s a feature, plan, plan, plan. Plan every single thing so you don’t have to write, write, write. Also, sometimes I ask the AI to explain what it understands from my prompt. That helps a lot in understanding what we can expect.
Add some skills so AI understands the way the existing code is written.
This normalization is the saddest part, as it keeps creating normalization for things that don’t even need it.
This is one of the biggest issues I have noticed. AI focuses on reaching the goal, but not on how to look for the best way to reach the goal. So either add some rules to check the utils or similar folders for what can be reused.
The idea of using AI is not bad, but how to harness it and deploy the code that you actually need to add is the question. Also, sometimes it’s not just about deploying code fast, faster, fastest. It is also about adding code that is planned, structured, and easily readable—not creating something you regret in the next hour, or even months later.
It’s more like harnessing a power from any superhero movie, like Wanda, but you need to know what you have created and what its consequences are, and not go mad with power to the point that you don’t know what you are coding at the end of the day.
Keeping this article short and sweet, so you all can go back to code more.