
I recently took on a new project: building a crossword generator. Instead of starting from scratch, I decided to test out Windsurf AI in VS Code. It was an interesting five-hour sprint that gave me some working code and a few key takeaways about AI-assisted development.
My biggest lesson was that you have to break down the problem. Giving the AI a single prompt to “build a crossword generator” was overwhelming and unproductive. The successful approach was to tackle smaller, more specific tasks, like generating the grid or placing a word.
The code it produced was functional, but it wasn’t perfect. I quickly learned that the AI is a great starting point, but it doesn’t always produce performant or elegant code. A developer’s critical eye is still essential. I spent a good amount of time reviewing and refactoring the generated code with the AI to make it more efficient.
Debugging is still a part of my workflow, but it’s a different kind of debugging. I’m now less focused on finding errors from a blank page and more on refining and optimizing the AI’s output. It makes me wonder if there’s a way to feed my debugging process and code improvements back into the AI automatically.
The end result is a functional crossword generator (code on my GitHub) and a clear takeaway: AI is a powerful accelerator, but it’s a partnership that still requires your guidance.