Claude 3.7 in Cursor is eager — sometimes a bit too eager. Like a programmer who's had one espresso too many. Sure, the code works, but often at the cost of unnecessary complexity — adding extra if-statements "just in case," even when simpler patterns would suffice.

Some might call this "defensive programming." Perhaps folks from the Java world even appreciate this style. I prefer to call it "unnecessary bloat." Yes, it might boost the percentage of working code (and thus look better in benchmarks), but complexity matters — a lot. Not just for aesthetics (though I admit holding the minority view that clean code is valuable in itself), but because overly complex code makes future changes harder and more error-prone, whether done manually or by Claude 3.7 itself.

For example, I asked Claude to move CSS into a separate file instead of embedding it in a Python string. It did exactly that—but then, just to be safe, also inlined the CSS anyway.

Another common pattern: when I request removing unnecessarily long imports, Claude often hesitates, leaving them in place "just in case." Unfortunately, this happens frequently.

Still, I don't mean to imply that Claude 3.7 is bad. There's a delicate balance between doing too little and doing too much—not only for AI but also for humans.

#claude #cursor #ai #llm