It is a vs code fork. There were some UI glitches. Some usability was better. Cursor has some real annoying usability issues - like their previous/next code change never going away and no way to disable it. Design of this one looks more polished and less muddy.

I was working on a project and just continued with it. It was easy because they import setting from cursor. Feels like the browser wars.

Anyway, I figured it was the only way to use gemini 3 so I got started. A fast model that doesn't look for much context. Could be a preprompt issue. But you have to prod it do stuff - no ambition and a kinda offputting atitude like 2.5.

But hey - a smarter, less context rich Cursor composer model. And that's a complement because the latest composer is a hidden gem. Gemini has potential.

So I start using it for my project and after about 20 mins - oh, no. Out of credits.

What can I do? Is there a buy a plan button? No? Just use a different model?

What's the strategy here? If I am into your IDE and your LLM, how do I actually use it? I can't pay for it and it has 20 minutes of use.

I switched back to cursor. And you know? it had gemini 3 pro. Likely a less hobbled version. Day one. Seems like a mistake in the eyes of the big evil companies but I'll take it.

Real developers want to pay real money for real useful things.

Google needs to not set themselves up for failure with every product release.

If you release a product, let those who actually want to use it have a path to do so.

  - Gemini 3 Pro (High)
  - Gemini 3 Pro (Low)
  - Claude Sonnet 4.5
  - Claude Sonnet 4.5 (Thinking)
  - GPT-OSS 120B (Medium)

That's not exactly really where I hoped my career would lead. It's like managing junior developers, but without having nice people to work with.

2024: every day a new Chrome fork browser is announced

2025: every day a new AI IDE vscode fork is announced

Anthropic and OpenAI are investing a lot into this space and are now competing directly with companies like Cursor. Cursor's biggest moat at the moment is their tab completion model, which doesn't exist in the Anthropic's and OpenAI's current offerings and is leagues ahead of Github Copilot's.

Antigravity is a VSCode fork that adds both Google's own tab complete and an agent composer, similar to products like https://conductor.build/. Assuming that Google doesn't shoot themselves in the foot (which they seem to like doing), we'll see if wrappers like Cursor / Windsurf / Cognition can compete against the big labs. It's worth noting that the category seems to be blurring, since Cursor has trained not only their own tab complete model but also their own agent model.

> Model quota limit exceeded. You have reached the quota limit for this model.

The software of the future, where nobody on staff knows how anything is built, no one understands why anything breaks, and cruft multiplies exponentially.

But at least we're not taken out of our flow!

Too early in my career to not give a shit and retire, but too late be excited about these things and eager to learn. What a time...

I'm going to need an AI summary of this page to even start comprehending this... It doesn't help that the scrolling makes me nauseous, just like real anti-gravity probably would.

It really seems like it's just standardizing into a first-class UI what a lot of people have already been doing.

I don't think I'm the target for this - I already use Claude Code with jj workspaces and a mostly design-doc first workflow, and I don't see why I would switch to this, but I think this could be quite useful for people who don't want to dive in so deep and combine raw tooling themselves.

I'm not sure many engineers will welcome this "promotion".

- It's VS Code

Like clockwork!

Looks like I'll wait to see if Google cares about putting the polish into a VSCode fork that at least comes close to what Cursor did.

It seems to streamline my existing Claude Code workflow with a much better UI. The tab complete seems the best I've experienced and the text/image selection, adding comments and iterating on a plan is genius.

Depressing to see everyone here unable to see the forest for the trees.

> You can verify your code quality at a glance, then ship with absolute confidence.

Proclaiming absolute confidence after a glance leaves me with scant confident in the merit of the confidence.

Trying to understand how this is anything net new in the space.

Additionally... Google Code was shut down in 2016? I have zero confidence in such a user hostile company. They gave you a Linux phone, they extended it, and made it proprietary. They gave you a good email account, extended it and made it proprietary. They took away office software from you via Google Docs, so now you don't even own the software they do.

No thanks.

> Fork VS Code, add a few workflow / management ideas on top.

> "Agentic development platform"

I'm Jack's depressed lack of surprise.

Please someone, make me feel something with software again.

- Nano Banana => Mockup

- Antigravity/IDE => Comments/note

- Gemini => Turn to code

- Antigravity/IDE => Adjust/code

All on the same platform so can maximum automate / "agentic"

Antigravity would be a world-changing technology. This isn't.

I used to love leaving that site open on public PCs and watching the reactions that resulted :)

What's most astonishing is that I can't seem to find what actual platforms it works for. I don't doubt the LLM's can write code in almost any language and for almost all frameworks, with varying success.

But which languages/platforms/framework will the IDE work for technically, having compilers etc built in? I don't care if an LLM can help me with the code, if I then can't compile it within the same IDE!

They have a "full stack" use case here, which doesn't even suggest what this stack consists of? https://antigravity.google/use-cases/fullstack

Am I going crazy or are they just handwaving the _actual_ development tasks in all this?

Why would you not at least link it to the pro and ultra accounts

at least you could upsell the pro subs to ultra. Millions of claude code and codex users who are into agentic coding is your servicable market paying attention today.

Now I'll delete antigravity and go back to codex / claude code / cursor ...

very interesting times; i'm glad to see browser automation becoming more mainstream as part of the ai-assisted dev loop for testing. (disclosure: started the selenium project, now working on something similar for a vibe coding context)

I remember a previous story months ago about Gemini that had Google PMs trying to hype their product, but it was all question about how nobody knows how to get Gemini API keys with any number of paid subscription.

On top of that how long until it’s https://killedbygoogle.com/ ?

Antigravity enables developers to operate at a higher, task-oriented level by managing agents across workspaces, while retaining a familiar AI IDE experience at its core. Agents operate across the editor, terminal, and browser, enabling them to autonomously plan and execute complex, end-to-end tasks elevating all aspects of software development.

via: https://www.linkedin.com/showcase/google-antigravity/about/

https://www.youtube.com/watch?v=YX-OpeNZYI4

https://www.youtube.com/watch?v=rKQ9b4UMpGQ

https://news.ycombinator.com/item?id=45967787

Nice that it's built-in, Claude Code needs an MCP for this at least.

> User Feedback: Intuitively integrate feedback across surfaces and artifacts to guide and refine the agent’s work.

I wish they'd just let me edit the implementation plan directly instead of me having to explain the corrections. Claude Code has the same weakness. Explaining the corrections is slower than editing the plan manually, and it still keeps the incorrect text in context as well.

> An Agent-First Experience: Manage multiple agents at the same time

Sounds nice in theory but I assume you can run multiple agents for 5 minutes or so and then you're out of credits.

As a claude code user I'm not really sold on this product.

Really reflects how companies are prioritizing hype and adoption over product quality.

(now off to download it...)

The problem is the rate limiting is both aggressive and has no option to pay to bypass.

Also call it "antigrav". Less of a mouthful

Pressing the "Submit" button on their "Google Antigravity for Organizations Interest Form" (https://antigravity.google/interest-form) doesn't actually do anything for me (tried Firefox and Chrome) -> their metrics will indicate that there's no interest from organizations -> the product will be killed in a year.

</snark>

Anyway, not a great first impression. I guess I'll try again in a few months.

Then I installed it and it was a VSCode fork.

Additionally, there are issues setting up accounts (Singapore VPN solved that for me), no support for Workspace users, only a free tier that requires data sharing, no additional rate limits for paying Pro or Ultra customers, etc. Even worse, Gemini CLI currently does NOT provide Gemini 3 Pro for Ultra Business customers despite paying over € 260,- per month, which is frankly ridiculous.

Will be honest, I was speculating that the reason for the multi month delay between the first A/B tests of Gemini 3 class models and the final release was so they'd have all their dugs in a row. Have some time to test everything, improve tooling, provide new paid subscriptions and/or ensure existing ones get access to everything day one, but they didn't.

Gemini 3 Pro seems very interesting (to early to say), but compared to every other recent launch by OpenAI (5, 5.1, Codex variants), Anthropic (Sonnet and Haiku 4.5), even Kimi (K2 Thinking) and Z.AI (GLM-4.6), this is by far the least organized launch of any frontier lab.

A buggy IDE which is unusable for paying customers, no CLI access for Ultra business (and none at all for Pro of any kind), etc. is frankly embarrassing when considering what competitors manage to provide the day a model launches.

What have they been working on these last two months besides going on X and posting "3" every couple of days? Why is there no paid Antigravity tier, no way to use Workspace accounts, etc? Before launching in this state, I feel it'd have been better to delay a bit more if it was absolutely needed.

Also, correct me if I'm wrong but isn't this the fourth or fifth IDE built by Google for LLM assisted coding? What happened to IDX and Firebase Studio and aren't they also based on VSCode?

> “Google Antigravity's Editor view offers tab autocompletion, natural language code commands, and a configurable, and context-aware configurable agent.”

Is it a typo or was there a reason to add configurable twice?

The browser extension is really cool and it provides a needed tool for the agent to use. It used the extension to show the page that it updated in the task document (the task doc is great too). However it showed me a page and did it was done, when it was clearly not done and not what I asked for.

I was expecting weaker tooling and a better model. I got good tooling and a not very good model.

Maybe 3.1 will deliver?

> Come join us! Programming is fun again! It's a whole new world up here!

I ran into a neat website and asked it to generate a similiar UX with Astro and it did a decent-ish job of seeing how the site handled scrolling visually and in code and replicating it in a tidy repo.

Those quota limits brought me back down to earth quickly.

"Novel agent-first form factor" feels very buzz-wordy. Does it refer to an actual feature?

I really miss the days of the professional casualness and naturalness of something like the "mother of all demos" [0]. Like, can you imagine the guy wearing a turtleneck and going, "but wait!" and acting surprised after every sentence? It would NOT have been the same demo.

[0] https://www.youtube.com/watch?v=yJDv-zdhzMY

But honestly Google software seems so buggy. The management class took over there a long ago and are quietly ruining the company.

That's a huge advantage, it's means all the obvious stuff will just work. LSPs, debuggers, version control, customisation.

As much as I like Emacs, it's an insane pain to make all these things work.

If your value prop is agents on a codebase, there's no point in trying to reinvent those. They have basically been solved.

I believe it is aimed at investors. Thus it will be forgotten the minute it stops influencing stock price.

Thus there is no need to take it literally as a developer tool - it's not.

Okay, but is it configurable? Also, can you configure it to write DRY code?

If Google has forgotten how to do Software, than the future doesn't look bright.

I have to close 4+ after just a few minutes of poking around

I assume that Copilot will have this model soon...

The task was to put create a header, putting the company logo in the corner and the text in the middle.

The resulting CSS was an abomination - I threw it all away and rewrote it from scratch (using my somewhat anemic CSS knowledge), ending up with like 3 selectors with like 20 lines of styles in total.

This made me think that 1: CSS and the way we do UI sucks, I still don't get why don't we have a graphical editor that can at least do the simple stuff well. 2: when these model's don't wanna do what you want them to the way you want them, they really don't wanna.

I think AI has shown us there's a need for a new generation of simple to write software and libraries, where translating your intent into actual code is much simpler and the tools actually help you work instead of barely allowing to fight be all the accidental complexity.

We were much closer to this reality back in the 90s when you opened up a drag and drop UI editor (like VB6, Borland Delphi, Flash), wrote some glue code and out came an .exe that you could just give to people.

Somewhere along the way, the cool kids came up with the idea that GUIs are bad, and everything needs to go through the command line.

Nowadays I need a shell script that configures my typescript CDK template (with its own NPM repo), that deploys the backend infra (which is bundled via node), the database schema, compiles the frontend, and puts the code into the right places, and hope to god that I don't run into all sorts of weird security errors because I didn't configure the security the way the browser/AWS/security middleware wanted to.

That seems bad.

No thanks...

—No one, ever.

Lotta people mining science fiction for cool names and then applying them to their crappy products, cheapening the source ideas.

I'm concerned that the new role of "manager of agents" (as google puts it) will be a soul destroying brain dead work and the morale won't be good.

If I write "float exp(float base, float exp){"

Then that is the source code and the rest is generated. Mixing it all up is as dumb as uploading a compiled binary or bytecode to git.

Especially annoying when you are working with other people and you can't tell what they actually wrote and know about.

I guess it must have been the GPL which isn’t compatible with their AI agents.

Oh, wait I was meant to take this announcement seriously?

I'm going to treat this like Kiro, and just use it until they start charging for it and then probably switch back to VS code with its built-in agent support.

Eventually they're going to do a rug pull, and instead of paying $10 a month for tons of AI code request, it's going to be two or $300 for that. The economics just aren't there to actually make a profit, hopefully before the rug pool happens local models on normal hardware will be fast enough.

⌘F only shows 1 result. and 0 in the comments here!!

I mean, google doesn't have the greatest track record.

Also, why does that site's scroll behavior is so weird? Just use the browser's default for Ford's sake!

And of course I would need to look at all the implications of spying, being locked out of google account and absence of support that are google amo. No time for that. Not for them.

Wow was google researching some kind of anti-gravity device behind the curtains for real and then dropped it out of nowhere?

Ah damn, yet another ai-assisted-something. Crap.

Why would I even bother getting mildly invested in this when the product launch/promotion incentive structure at Google is so well known?

The people at Windsurf who worked on this must be laughing at us driving on their Lambos and Ferraris.

They glued slop together, shipped this and now are in Tahoe drinking Martinis watching the sunset from their private chalets.

It’s just google’s attempt at cursor. Nothing to see here.

What about a demo that shows how this can be used to fix for example https://github.com/emscripten-core/emscripten/issues/24792?

But for writing code in some domain I am good in, they are pretty much useless.. I would spend a lot longer struggling to get something that barely functions from them VS writing it myself, and the one I write myself will be terse and maintainable + if it has bugs they will be like obvious ones, not insane ones that a human would never do.

Even just when getting them to write individual functions with very clear and small scopes.

I know there's a "free plan with generous rate limits" but it's obvious that they're losing money there.

Console error:

> Loading module from “https://antigravity.google/main-74LQFSAF.js” was blocked because of a disallowed MIME type (“text/html”).

I dont know what i expected tbh

opencode with it's superior feature set and ability to use any model provider i want is....

superior

why would you even bother with google at this point?

Google at its finest

Ah Google misconfigured their web server:

> Loading module from “https://antigravity.google/main-74LQFSAF.js” was blocked because of a disallowed MIME type (“text/html”).

Edit: And a couple minutes later, it is now working. Guess Google is reading HN.

And now they can’t even ship a desktop app without forking VSCode? Look, I get it. There’s this huge ecosystem. Everyone uses it. I’m not saying it’s damning or even bad to fork it.

But why is this being painted as something revolutionary? It’s a reskin of all the other tools which are variations on the same theme, dressed up in business speak (an agent-first UX!). I’m sure it’s OK. I downloaded it. The default Tokyo Night theme is unusable; the contrast can’t be read. I picked Vim bindings, but as soon as I tried to edit a file I noticed that was ignored.

What happened? Is this how these beautiful, innovative companies are bound to end up?