Loading request...
The user notes that while Cursor AI is effective for UI generation, it requires significant effort and very specific prompting to handle complex programming and codebase architecture. They imply a need for the AI to improve its capability to understand and manage these complex architectural aspects with less explicit user direction.
The biggest thing that Cursor/AI in programming has done for me is surprisingly to speed up designing “nice UI” in minutes instead of hours or days. And by doing so be able to get a feature much quicker in a first shippable “end2end” state without being bogged down on fiddling with the details in the UI. With Firmhouse UI being Tailwind-based and follows a lot of common B2B interface patterns, Cursor Composer is a beast with that. It’s usually as simple as telling it to “now add a pill-based nav and make it look nice”. And 9/10 times it’s best at that. So generating UI code seems to be task very suitable. As it’s a a lot of boilerplate and markup typing that in essence is not “smart” work. It just needs to get done. The keystrokes need to be typed. But I’d rather not have my time or brain cycles spent on that kind of work. In my job as CTO, my human brain is at its best when translating complex market and domain space insights and turn them into “how it should work” holistically. And over the years, my brain has gotten very good and fast at that. The “how it looks” and building that is now nicely delegated to the AI so it doesn’t take a lot of brain cycles and keystrokes anymore. So it’s not the complicated programming and codebase architecture where AI currently shines. I’ve noticed that it still takes a lot of effort in prompting to get that done right. You still need to direct the AI very specifically on those kind of things. So structuring the code yourself first and then telling the AI to “fill it up” works better there.