My Wild Ride Through AI Development Tools (Or: How I Landed Back at Cursor)
Testing Google AI Studio, Antigravity IDE, and why I keep coming back to the same answer
Hey there!
So I’ve been on this journey lately... testing out different AI development tools. And when I say testing, I mean REALLY diving in and building actual things with them. Not just playing around for a few minutes and calling it a review.
This isn’t about those vibe coding platforms like Replit or Lovable (though I built a Shopify store with Lovable recently and that was wild). This is about the tools that sit somewhere in the middle - the ones that help you actually CODE but with serious AI assistance behind the scenes.
The Three Contenders
I ended up trying three different tools over the past few weeks:
Cursor - the one everyone’s talking about
Google AI Studio - surprisingly capable and underrated
Google Antigravity IDE - Google’s answer to Cursor
Let me walk you through what I found with each one...
Google AI Studio: The Underrated Champion
I’m going to start here because Google AI Studio genuinely surprised me.
I stumbled into it because I wanted to try Gemini 3 Pro for app creation. And honestly? It’s REALLY good at one-shotting apps. I’m talking about giving it one solid prompt and getting back something that actually works - not just a mock-up, but a functional app with nice visuals that you can immediately interact with.
What makes it interesting:
Gemini 3 Pro is legitimately impressive for development work
You can build, test, AND deploy all in one place (no need to spin up separate infrastructure)
GitHub integration is built right in
You can make apps public immediately
File downloads work seamlessly
It’s incredibly cost-effective
The catch? It’s not as flexible when you need to do more complex stuff or switch between different AI models. Google defaults to Gemini (which makes sense for them), but sometimes you want GPT or Claude for specific tasks.
Still, if you have a clear idea and want to prototype something FAST without worrying about deployment... Google AI Studio is absolutely worth your time. It’s underrepresented in all the “best AI coding tools” lists, and that’s a shame.
Google Antigravity IDE: The “Almost There”
Next up was Google’s Antigravity IDE. This one positions itself as a more serious competitor to Cursor - a full development environment with AI baked in.
My take? It’s like Cursor... but from maybe a year and a half ago.
The concept is solid. The agent-based approach makes sense. But when I actually tried to build with it, I kept running into execution problems and missing features that I just didn’t have in other tools.
It feels like Google is TRYING to compete directly with Cursor, but they’re still playing catch-up. Which is fine - it’s in development, things will improve. But right now? It’s not quite there yet for serious work.
Cursor: Why I Keep Coming Back
And this brings me back to Cursor.
Here’s the thing - I’d used Cursor before, maybe 8-10 months ago. It was good then. But when I came back to it recently? Holy hell, it’s TRANSFORMED.
The big changes:
Cursor 2.0 introduced a whole new agents tab with seriously helpful features
It’s way more agent-based than it used to be - you can actually delegate entire coding tasks
You can choose your model (I’m using Claude 4.5 Opus for most development work)
The layout is flexible - switch between editor mode and agent mode depending on what you need
Built on VS Code, so it’s immediately familiar if you’ve used VS Code
What I really appreciate: Sometimes you just need to code something manually because it’s FASTER than explaining it to an AI. Cursor gets this. It gives you the flexibility to jump between “I’ll do this myself” and “AI, handle this for me.”
Example: I wanted to switch the AI model in an app from Gemini to GPT. If you know where to look in the code, it’s literally a few seconds. But if you ask an AI agent to do it? Sometimes it gets confused, starts checking docs, second-guesses itself... and suddenly what should take 5 seconds takes 2 minutes.
Cursor gives you both options. You’re not FORCED into an agent-only workflow.
The Real Workflow Insight
Here’s something I learned that nobody really talks about: knowing how to code still matters, even with AI tools.
Not because you need to write everything from scratch. But because understanding code lets you make judgment calls about when to use the AI and when to just... fix it yourself.
The best workflow I’ve found is switching between modes:
Let AI handle the heavy lifting (new features, boilerplate, structure)
Jump in manually for quick fixes and tweaks
Use AI again for testing and refactoring
This is why Cursor works so well for me. It’s built for this hybrid approach.
(Side note: I also found this great “Building with Cursor“ tutorial on Notion that’s super helpful for new users. Six easy steps with videos. Worth checking out if you’re just getting started.)
Tools Are Getting Wild... and That’s Exciting
The pace of change here is INSANE. Like, genuinely wild.
I was using Cursor less than a year ago and it felt like a smart code editor. Now it feels like having a junior developer pair-programming with me. That’s a massive leap in a short time.
Google is clearly investing heavily in this space too - both AI Studio and Antigravity show they’re serious about competing. And honestly, that competition is GREAT for us developers because it means these tools will keep getting better and cheaper.
Speaking of tools getting more powerful... if you haven’t checked out what you can do with Claude’s Model Context Protocol, that’s another rabbit hole worth exploring. Giving AI assistants “hands” to interact with your actual systems changes everything.
Where I Landed (For Now)
For serious development work: Cursor is my go-to. The flexibility, model selection, and mature feature set make it the best choice when I’m building something real.
For quick prototypes: Google AI Studio is fantastic. One prompt, functional app, deployed. Can’t beat that for speed.
For experimentation: I’ll keep checking back on Antigravity IDE as it matures. Google has the resources to make it competitive - it just needs more time.
And next up? I’m diving into deployment workflows. Probably going with Vercel because it’s stupidly easy to use, though I do have a DigitalOcean server I’ve been meaning to utilize more...
But that’s a topic for another post.
What are YOU using for AI-assisted development? Are you team Cursor? Team GitHub Copilot? Something else entirely? I’m always curious to hear what’s working for other people.
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.