I Tried Vibe Coding the Same App with ChatGPT, Claude, and Cursor. Here’s What Actually Happened

Same prompt. Same feature. Three AI tools. The results were not what I expected.

three boxing rings side by side, each labeled ‘ChatGPT’, ‘Claude’, and ‘Cursor’ in amber/brown handwritten text, with a laptop glowing in the center.

So here’s how this started. I was building a small SaaS side project, a simple habit tracker with a dashboard, user auth, and a basic subscription flow. Not a toy app, but not a massive system either. The kind of thing you’d actually ship.

I’d been vibe coding for a while. Letting the AI take the wheel, reviewing what came out, shipping. It’s fast. It works. But I kept hearing conflicting takes on which AI is actually best for this workflow. Some people swear by ChatGPT. Others say Claude is the only one worth using. The Cursor crowd is practically a religion at this point.

So I did the only sensible thing. I built the same feature three times. Same prompt each time, same starting context, different tool. And I tracked everything: time to working code, bugs found in review, quality of tests, and the less measurable stuff, like whether the AI asked useful questions or just started generating and hoped for the best.

This is what I found.

The setup

The feature was a subscription upgrade flow. User on a free plan, wants to go Pro. Needs to handle: payment via Stripe, plan…