#Cursor (vibe coding) experiments

Been experimenting with #Cursor over the past few weeks for a small personal project - a web-based calendar that displays different types of regional events from my home state of Bihar. Here are my thoughts from the trenches:

Stage 1: Idea to MVP

Cursor is a game changer when it comes to getting your idea off the ground. I started with just HTML/CSS/JS - no server-side logic - and was amazed at how quickly I could turn an idea into a working prototype. Before tools like this, probably only 1 out of thousands of people with an idea might’ve even reached MVP stage. Now? You can build something real in a weekend. Not that I am building anything on the level of starting something real, just playing around!!

The AI-native dev experience makes this phase feel magical. Hats off to the Cursor team - what they’ve built for that initial lift is truly impressive.

Stage 2: MVP Bugs

But then... the real dev journey begins. Once I wanted to go beyond the basic month view and add week/day views, a dialog on every day click, things started breaking.

• Cursor introduced bugs in calendar navigation
• Fixes in one view broke the others
• In one case, the app even went into an infinite loop - I had to kill the IDE Eventually, I had to step in, debug manually, and take over. AI couldn’t stitch it all together at this stage.
• And once you start seeing bugs aren't getting fixed by AI, how can you improve further without fixing these bugs first ? Seems like a dead end for vibe coding.

Observations

We’ve all groaned at the idea of fixing someone else’s code - and let’s be honest, AI-generated code feels like that sometimes 😅

Sure, we’re still in early innings of AI-first dev tools. There’s massive potential, and I’m optimistic. But one thing is clear to me: developers aren’t going anywhere. The bar has just shifted - we step in later, and when we do, it’s at a higher level of complexity.

Boilerplate generation has been around for a while. What Cursor does is exponentially next-level. But for now, AI can start the race - humans still finish it. This is depicted in the article main image. Yes, the Y axis is coding speed and X axis is time. So, I feel like the speed of coding exponentially increases at the start of project with the help of AI but then decreases when developers come in and add advanced features and start fixing bugs which can't be fixed by AI. Even at this stage, AI could help implementing small feature sets but in the end, developers will need to be more involved as time progresses !

Would love to hear how others are using tools like Cursor or CodeWhisperer or Copilot. Where do they shine, and where do they stumble?

#AI #DevTools #Cursor #WebDevelopment #CodingWithAI #SoftwareEngineering