Cursor is counterproductive

I did recommend using Cursor, and I still stand by it. At the same time, in a world and industry that is rapidly changing, I am also constantly looking for reasons to change my mind. Yesterday I found one that I want to share with you.

Using Cursor (and similar tools) makes everything faster. Everything. The Agent mode runs with an idea, and implements everything it can think of and more. Even if 90% of it is wrong, it still writes the code, shuffles everything around and leaves you wondering what the hell just happened. You get so much code and logic to try and parse and understand that you give up and throw it all away, starting from scratch again. This has been referred to as the illusion of AI productivity.

I don’t believe in throwing everything away. I think instead this is the learning opportunity. Why did we get to a point where the code became useless? What hubris led us here?

While we wait for the machine to get smarter, for the context to get longer and more comprehensive, and for solutions that make sense of the big picture while getting the details right, I think we have to find the middle ground ourselves. Keep control of the context. Persist the truth as you know it, and guard it from tampering by keen machines. Use that truth as the starting point for your AI adventures, LLM discussions and coding frenzies. I think that also means taking a step back from the tools that promise to do everything for you, and look for - or build your own - tools that make you faster and better at learning, understanding and making decisions.