Why Building is Fun Again
(And Why I Ditched Vibe Coding for Cursor)
Ok so...building stuff is actually fun again.
Like genuinely fun. Not the “I spent 3 hours debugging nginx config” fun, but real “holy shit I just shipped a feature in 20 minutes” fun.
And the weird thing? I went full circle. Back to where I started.
The Vibe Coding Honeymoon
Over a year ago I started coding again. Re-learning basics, but this time with AI helping me out. Cursor was my day-one choice - perfect hybrid where I could write my own code, but when stuck or wanting speed, AI was right there.
But then...the vibe coding tools explosion happened. Lovable. Replit. Bolt. All these tools promising you could build apps without really coding.
And man, I fell HARD for Replit.
Here’s what hooked me: I could go from idea to live app in an hour. Not “coded in an hour then spent 3 more deploying” - actually LIVE. No nginx setup. No database config hell. No deployment pipeline fights. All that boring infrastructure stuff that AI back then just couldn’t help with? Replit handled it. Automated it.
I remember building full apps from 2-3 prompts. This was before they even had “Agent Mode” - that’s how good the basic experience was.
Cheap. Efficient. Handled everything - development, maintenance, deployment.
So yeah...I dropped Cursor. Why wouldn’t I? Started testing every vibe coding tool I could find. And they kept getting better. More integrations - Stripe, Supabase, AI APIs. It felt like the future was actually here.
When the Magic Broke
Can’t pinpoint exactly when, but something changed this year.
Replit started costing way more. And got slow. Really slow.
I remember building a QR code generator for like 5 bucks. Now with the autonomous AI Agent? Not even possible at that price. And worse - it got dumber. Can’t even get frontend right from one prompt anymore.
You might say “well that’s the model not the tool” but...I disagree. Hard.
Because models from 1.5 years ago were objectively worse than now. Yet I could do MORE, with FEWER tokens, for LESS money. That tells me something changed in how these tools architect things or use the models.
Don’t get me wrong - vibe coding tools still have value. They’re incredibly accessible. Someone with zero tech knowledge, some money and time can build real products. That’s genuinely cool.
But for me? It stopped working.
The Google Detour Nobody Asked For
I also tried Google’s Antigravity. Free Opus 4.5 and Gemini 3 Pro Preview? Hell yeah sign me up.
Except...I just couldn’t use it.
Once you’ve spent even 2-3 hours in Cursor, you get spoiled. The experience is SO much better. Antigravity felt like going backwards.
Google AI Studio with Gemini 3 Pro Preview was interesting enough to maybe prototype a frontend, then immediately move to Cursor to actually build. Some people like it. I don’t.
Coming Home to Cursor
So I came back to Cursor. Sheepishly. Like “hey old friend sorry I left you for shiny new toys.”
Timing was good though - they’d launched Cursor 2.0, recently 2.2.
And wow.
It’s not trying to be a 100% vibe coding tool. But the way AI is implemented, how agents work...you can tell it’s built by people who actually code. People who understand development. And it shows.
No matter which model I use, it’s really good. (Though for me: Opus 4.5 is my go-to for serious coding, Composer for smaller things. ChatGPT 5.2? No luck. Codex? Meh.)
How I Actually Use Cursor Now
My setup: Cursor + DigitalOcean App Platform. Yeah I know, pricey, not Vercel, but it works for me.
I use Agent Tab way more than expected. Because it actually delivers. Very few (sometimes zero) corrections needed.
Part of that is using Plan Mode more frequently. And debug features. Both showed incredible efficiency.
But the real game-changer? The agent that reviews changes before I commit PR to GitHub. It’s frighteningly good. I review everything it flags, and out of 10 times I only disagree twice. 80% accuracy on catching issues I would’ve missed. Seriously considering upgrading to Bugbot (higher tier).
One thing I can’t figure out: Cloud Agents. I think it’s like Claude code editor but in Cursor? Don’t know how to use it efficiently. When I’m mobile and get feedback like “hey I can’t see LLMs full names on mobile in LLMatcher” - I’d rather just open Claude iOS app and fix it there.
Oh and Cursor has built-in browser now where Agent can run dev environment to test. This is cooler than it sounds. Used to manually run dev servers every time I wanted to check something. Now? Don’t have to go anywhere. It’s all right there.
Real Talk: I Built LLMatcher With This
So here’s the proof this actually works...
I just built and shipped LLMatcher in about 10 hours. Total cost? Like $97 (domain, hosting, tokens, Cursor subscription).
What is it? A blind voting site where you compare AI model outputs side-by-side without knowing which model created them. After 50 votes you unlock personalized recommendations - basically “which AI actually works best for YOU.”
Why’d I build it? Because I was tired of “GPT vs Claude” debates. The best model depends on what YOU value. Your preferences for coding might be totally different than writing. Benchmarks don’t tell you that.
And I built the whole thing - frontend, backend, database, API integrations, deployment - with Cursor. Fast. Efficiently. Without fighting my tools.
That’s when I knew I made the right choice coming back.
(If you’re curious, try LLMatcher - takes like 15 minutes to unlock your personal AI toolkit. And honestly? The results surprised me. I thought I preferred different models than what the blind testing showed.)
Why This Actually Matters
The vibe coding dream isn’t dead. It’s just not for me anymore. Not right now.
What I learned: the best tool is the one that matches how you actually work. Not the coolest marketing. Not the most “magical” experience.
For rapid prototyping by non-technical founders? Vibe coding tools are still incredible.
But for someone who wants control, speed, and ability to understand and modify what’s being built? Cursor wins. Every time.
Building is fun again because I’m not fighting my tools. I’m working WITH them.
And honestly...that makes all the difference.
PS. How do you rate today’s email? Leave a comment or “❤️” if you liked the article - I always value your comments and insights, and it also gives me a better position in the Substack network.