The agent workflow seemed like a meaningful improvement over Cursor. It was easy to follow the agent's code exploration and "though process". The review workflow made it feel very natural to iterate on the agent's proposed solution.
The onboarding/trial felt weak somehow. I ran out of free-usage fairly quickly, and it fell over onto weaker models. I wasn't prompted to upgrade. For a new product, I'd expect to be able to trial it fully-featured, and then decide whether to upgrade.
There were some UI bugs, but nothing I couldn't live with for a new product.
The most important thing is going to be model accuracy. It's worth giving up on UX for better code generation.
In that sense, I actually stalled. Antigravity had momentum, but then stalled and failed over to weaker models. I had to get back to work, so I switched back to Cursor + Claude Code.
Extremely. And, the agent "thought process" is very discoverable.
It seemed to grok our moderate Rails monorepo pretty smoothly.
There's a planning mode (default) that I think did a standard job at. I didn't notice any guardrails
It felt very fast until I ran out of credits and decided to just stop working.
The agent workflow seemed like a meaningful improvement over Cursor. It was easy to follow the agent's code exploration and "though process". The review workflow made it feel very natural to iterate on the agent's proposed solution.
The onboarding/trial felt weak somehow. I ran out of free-usage fairly quickly, and it fell over onto weaker models. I wasn't prompted to upgrade. For a new product, I'd expect to be able to trial it fully-featured, and then decide whether to upgrade.
There were some UI bugs, but nothing I couldn't live with for a new product.
The most important thing is going to be model accuracy. It's worth giving up on UX for better code generation.
In that sense, I actually stalled. Antigravity had momentum, but then stalled and failed over to weaker models. I had to get back to work, so I switched back to Cursor + Claude Code.
Extremely. And, the agent "thought process" is very discoverable.
It seemed to grok our moderate Rails monorepo pretty smoothly.
There's a planning mode (default) that I think did a standard job at. I didn't notice any guardrails
It felt very fast until I ran out of credits and decided to just stop working.
For a very long time I had all these ideas that I couldn't act on. Antigravity has allowed me to just explore and create again, which is a wonderful feeling. The recent addition of skills has exploded my workflow productivity, it literally handles my inbox and calendar for me now. I've been a Google fanboy for a long time and I love what they're doing here.
Still has some minor UI quirks and small bugs as expected with a new tool, but nothing that stops the workflow. I’m excited to see more built-in skills added.
I’ve used Cursor and Claude Code extensively, but they always felt like advanced autocomplete. I chose Antigravity because it bridges the gap between "writing code" and "getting things done." The autonomous skills, handling my real-world emails, calendar, and project setup, along with the deep planning it applies to project architecture, make it the only tool that actually feels like a teammate instead of just a text editor.
Incredibly well. It’s the only tool I’ve found that actually researches the dependencies across the whole codebase before proposing a change, rather than just guessing.
It has completely replaced my previous setup. The transition was smooth, and it handles my local environment much better than typical cloud-based AI tools
As a solo dev, it has significantly reduced my "self-review" time. The implementation plans it generates allow me to catch design flaws before the code is even written
Some UI quirks and minor bugs, but very powerful overall. I really like how it handles implementation plans and walkthroughs. I think it will be great a couple months from now.
Mostly needs some bugfixes and UI polish. E.g. sometimes one of the panels obscures an Accept/Reject button, so it's hard to see the app is waiting for user input. And when allowing certain commands, I haven't been able to make this stick yet. There's an allowlist for commands in the settings, but this doesn't seem to work for me.
Cursor never quite resonated with me, but I need to give it another shot. Claude Code is great with Opus 4.5, but way too expensive. Antigravity is very close to Claude Code, and as a paid Google AI Pro subscriber I haven't bumped into any limits yet. I actually like Copilot quite a bit as well.
Python code seems pretty solid, but it sometimes misses some minor things, so definitely still needs a thorough review.
Haven't tried; I'm a solo dev.
N/A
I tried AntiGravity and am very satisfied, as it allowed me to be highly productive and tackle complex tasks efficiently. I am also very grateful to the team for making this possible, and I’m excited for future updates and what’s to come. 😊
I think the biggest potential for improvement probably lies in enhancing the Agents. I’m not currently using the Agent Manager (CTRL + E), because, to be honest, I usually handle one task at a time, as the tasks are very complex and need to be reviewed and understood by me.
Because it integrates Gemini3 in a free tier.
The agent always tests his own code and works until the test is passed.
My primary language is TS. Very acccurate. Problem is not accuracy. It is making the best design choices.
You can not the agent APIs without a network connection.
Cursor
Claude Code
Github Copilot