Claude Code vs. Cursor vs. Google Antigravity vs. Lovable
· 2 min read
Claude Code vs. Cursor vs. Google Antigravity vs. Lovable.
I've paid for all four AI coding tools. This is how they compare for a product person who builds real products with AI agents.
Cursor
Used: April 2025 to November 2026, my main tool for 1.5 years.
Great at: maturity. I watched it grow from a VS Code copilot into an agent for everything. Pashtpan bot, a content analyzer (no good name yet) and memo-ri were all built in it. There are more, and I'll cover each one in upcoming posts.
Underrated: you stay close to the code. Many times I caught the agent touching services it didn't need to, only because I saw the diff.
Best for: building inside your code and learning how your product works.
There's a reason SpaceX paid $60 billion for it.
Google Antigravity
Used: about 3 months, right after launch.
Great at: the idea. An agent-first IDE where you manage agents, not files.
Weak at: almost everything else back then. Bugs, wrong answers, missing features. Next to Cursor, it felt like a toy.
Fair note: that was a while ago, and Google has shipped a lot since. I'm pretty sure it's much more polished now.
Best for: Google's agent-first take, much changed since I tried it.
Lovable
Used: small projects, and I still use it at work.
Great at: keeping its promise. It's built for non-technical people, and they can ship a real product with it.
Limit: the same promise keeps it simple on the surface. That said, it does a lot more today than when I started.
Best for: turning an idea into a live product without touching code.
Claude Code
Used: the last two weeks, on the Max plan.
Great at: quality. Skills, connectors, a desktop app, memory. Cursor has all of these too. The difference is how good the result is.
Best for: top results from an agent connected to your whole workflow.
Why I switched: at renewal, external models were billed separately, and my $20 plan became $36. Then checkout got stuck in a Grok account sync loop for 10 minutes. I subscribed to Claude instead, and Claude Code moved my whole setup in about an hour.
What Cursor and Claude Code both taught me: the model is only half of an AI agent. The harness around it is the other half, and both teams nailed it. Put Opus 5.5 inside a homemade agent, and you'll feel the difference in a minute.
My ranking today:
1. Claude Code
2. Cursor
3. Lovable
4. Antigravity (as I remember it)
Which AI coding tool do you use, and what would make you switch?
