It Know Everything About Programming — But Not How to Program: My Experience with Cursor AI
AI tools like Cursor AI are changing how we code. They seem to know everything about programming — but sometimes, they still don’t know how to actually program.
After spending a lot of time working with Cursor AI, I wanted to share a few honest thoughts. I’m excited about what’s possible, but I also think we’re still far from letting tools like this replace real programmers.
First, where Cursor AI shines:
But here’s where things get tricky — and why we’re not close to replacing programmers yet:
Where Cursor AI struggles — and why humans still matter:
The bottom line:
Working with Cursor AI is a lot of fun — but also a bit like working with a really smart but unpredictable intern. Sometimes it nails it, sometimes it breaks things in ways you didn’t expect.
Recommended by LinkedIn
Why I’m still optimistic:
Even with all these issues, the progress is amazing. I fully expect tools like this to get dramatically better in the coming months and years.
But for now?
If you’re trying Cursor AI or similar tools, my advice: enjoy the speed, but stay in the driver’s seat. The future is coming fast — and I’m excited to see where this goes
I'm curious to hear from others — what’s been your experience with AI coding tools like Cursor AI, Copilot, or others?
Have they made you faster? Or caused unexpected headaches? Let's share notes!
#AI #GenerativeAI #SoftwareEngineering #Coding #AIAssistant #CursorAI #CodeGeneration #FutureOfWork #Developers #TechInnovation
Nir Melamoud, this was a sharp and honest piece. What stood out is not just the tooling review, but the underlying tension between code execution and system reasoning. Cursor AI does not fail because it lacks syntax. It fails because it lacks abstraction, recursion, and judgment. The question for me is not whether it helps engineers write faster, but whether it enables better systems thinking across the stack. Does it reduce noise? Does it create fewer reversals? Does it reinforce or erode intentionality in architecture decisions? The real edge is not speed. It is the preservation of meaning as code compounds. Curious how you see this evolving once teams move from solo exploration to coordinated build with AI support.
We use AI as code review tool with some success, it does ok job of pointing more obvious things out and, what is more important, no offense can be taken 😅
Nir Melamoud love this article! Really highlights what AI will be good for but actually where it fails which no one talks about because of the 'AI hype'. Really surprised with it missing some of the basics, eye opener!
Thanks for sharing, somehow I would have expected greater reliability and predictability from AI even at this stage of maturity. Also intrigued by the reckless decision making....
Try Claude-Code... It's very expensive but it's gonna blow your mind