I've been using Cursor over a year now and have used it quite a bit in that time. From its early days when the chat felt like an extension tacked on top to now where chat seems to be the primary mode of interaction, it's come a long way.
Given the hype around Antigravity I decided to use it for a while. I've been so used to my workflow in Cursor that switching IDEs feels like trying to write with my left hand, but I wanted to see if the grass was actually greener; or at least different.
Spoiler: It's a mixed bag.
The Bad stuff
The onboarding wasn't exactly the "white glove" experience I was hoping for. I expected it to just pull in all my VS Code extensions and settings seamlessly, but a bunch of them failed to import. The worst part? It didn't even tell me. I only realized when I opened the extensions panel to see why my theme was different. It uses the Open VSX marketplace, which causes compatibility gaps, but not warning the user feels like a letdown.
Then I started coding. It feels like it doesn't automatically index the codebase in the background. Every time I started a new chat, it felt like it had to search the codebase from scratch. The agent would go "let me read the codebase to gather context", every single time.
The agent's interaction with tools also felt a bit raw. I tried to get it to check some TypeScript errors,
and
it ran npx tsc --noEmit in the terminal. Same for Eslint. It works, sure, but I wished it could
just tap into the
language server directly.
The browser agent relies heavily on executing JS for everything rather than interacting with the DOM directly via MCP.
Another friction point: The lack of tabs for agents. It's difficult to switch conversations even if things are running in the background. It seems they want to push us towards the Agent Manager but that ruins context for me.
The Good stuff
Google AI Pro plan limits are incredibly generous. I stopped worrying about "spending" credits. Do note that I got Google AI Pro plan for free through a promotion Google is running in India.
The Hell yeah! stuff
1. Video Analysis for Testing
This was huge. I was working on some tricky CSS animations. In Cursor, I'd have to describe what's happening. The Antigravity browser agent just... analyzed the video recording of the browser session. It's something I'd desperately wanted in Cursor. Antigravity's browser agent works perfectly for this usecase.
2. No "Modes"
I realized I've developed a subtle annoyance with Cursor's modes-switching to "Agent" to write, "Ask" to ask, "Plan" to plan and "Debug" to debug.
In Antigravity, the agent just does what it's asked. I tell it to "Make a plan," and it creates a plan artifact. I can review it line-by-line, leave comments, and iterate right there. It feels organic. It doesn't feel like I'm jumping lines at the receptionist's desk.
Conclusion
It's definitely rough around the edges. It feels like where Cursor was in Feb/March 2025. The onboarding needs love, and the UI for switching conversations is clunky. But the philosophy: generous limits, video analysis, and a mode-less workflow centered around artifacts is compelling.
I'm not uninstalling Cursor just yet, but if Google's models improve context size, following instructions and the browser agent begging me for permission every 5 seconds, Antigravity might be a better option overall given the price point and the additional benefits like cloud storage, Gemini in emails, workspace etc. To keep me on Cursor, they could get rid of the modes and make the agent do what it's asked to do. Better pricing like a $40 plan with 1k requests would be great as well. I currently run out of the included 500 requests in less than 2 weeks.