Part of the magic of their account rep strategy is how they keep them on your account for so long, you get to develop not just a rapport but a trust that they truly understand your business. It gives me faith that when they advise us on their new AI products, they're going to be a good fit.
Edit: I forgot to mention the curiosity and humility they bring to our calls. If I point out another vendors approach to a problem that we have, they always lean in and want to help improve their offerings from our feedback. They know it's not just enough to "be Google".
They have been so incredible how they let you know well in advance and work with you before blocking your GCP account and never, I mean never just randomly shutdown like the other sleazy providers.
This is a huge surprise, never thought I would see this in my life time.
I am especially impressed with how they keep supporting Google Reader for all these years despite the declining user base, because they care so much about the existing users.
I never really used the Antigravity IDE, but had it installed. The update also made me do a double take and wonder what the hell was going on.
It seems like Google is hitting the reset button on the product they call "Antigravity", existing users be damned. Fine, if you've never installed or used the previous version before... but for existing users the "bait and switch" is incredibly disorientating.
My take is they saw the market size for a general agentic tool as being larger and more significant than a specialised IDE. It shows a pretty large lack of respect for users in the later group though.
Gemini CLI (the open source cli) is being deprecated, and the recommended replacement is Antigravity CLI (which supposedly comes with the new Antigravity, not the IDE). shrug. Surely this will be maintained long term...
I was surprised people were so willing to jump to closed source IDEs just for access to coding agents. The trade-off you pay for tight integration between the IDE and the coding agent is lock-in because the barrier to switching IDEs is nontrivial.
Your coding environment stands a lower chance of disrupion when you use an open source IDE with a CLI agent. Yes it's slightly annoying to separate the agent from the IDE but the benefit is that it's much easier to switch between Claude Code, Codex, Gemini CLI (now antigravity CLI), etc which means you can more easily benefit from pricing and coding performance differences which seem to change monthly.
Fwiw, the (mostly) closed source jetbrains IDEs support multiple models with their coding agents, byok, and using different agents like Claude Code via ACP
Antigravity is just a vs code (more correctly: codeium) skin with Google telemetry and agent Integration. You can switch back to Microsoft's or cursor's flavor in minutes.
Google made its lack of interest in Antigravity IDE obvious from very early. Updates were few and far between and app-breaking bugs stuck around, despite tons of reports.
Google's lack of focus is astounding. They sprinkle random products here and there and seem to then tepidly pick the product surface that is doing least bad and then tepidly focus on that. Compare that to every other AI lab, large and small that knows its identity and shaped its products around that.
Perhaps it's a sort of resource curse. Google doesn't need any one of these products to succeed, and it shows.
But I have to say that I never understood the Antigravity IDE. I much prefer using Gemini CLI in combination with vscode. It works like a charm. Now, I'll do the same with Antigravity CLI and vscode. It works fine.
What lead? Maybe because I'm mostly using AI/LLMs for development, but neither Google, Anthropic, xAI or anyone else has ever been in the lead, OpenAI always had the best models in my mind, as long as you're comparing the "top" plans between all of them.
Besides, they all seem to shoot themselves in the foot, OpenAI included, seems the only thing that differs is how often and how big the damage is.
All the labs astroturf all the social media, HN is not unique and OpenAI wouldn't be the only ones. I even receive offers sometimes on my email put in my HN profile, asking me to post about their project in exchange for money.
Be skeptical of anything you read online, not just what you think is "obvious astroturf".
Wait what? Why don't I get emails like this too? /s
(on a serious note, do you feel comfortable naming and shaming such companies, this is sort of a serious accusation imo and if not then how much money they are trying to give. It would be an interesting discussion and feel free to mail me if its confidential, waiting for your response and have a nice day :-D)
Nah, maybe one day I do a collective public post of it, for now I just try to get their company and/or name first, then forward it to HN themselves so they can ban them and keep an eye out for them.
I probably wouldn’t say they always had the best model but for years OAI was definitely pushing the limits both on model quality and product offerings. It was not until the last year or so that Anthropic started punching above their weight.
> It was not until the last year or so that Anthropic started punching above their weight.
Anthropic's stuff been useful for the last two years I'd say, especially in the beginning of Claude Code, but as soon as the Codex TUI was available, I was daily-driving both of them, literally executing the same prompts for each of them and comparing the final results, and Codex simply writes better code in 9/10 cases (but still not always).
Claude Code has only been around for a year and change. At least for our internal tests 2 years ago Anthropic models started to at least become semi-useful but they still were not great, they struggled with structured output. Prior to that their alignment strategy made the products highly unhelpful in an API context. The past 6 months to a year is where Anthropic has really shined, they have model parity and sometimes taking the lead and more importantly their product offering on the consumer side has crushed it.
> Claude Code has only been around for a year and change.
We've been experimenting with "agent harnesses" way before that though, I'm sure the first time I tried building that sort of thing was in 2023 sometime with GPT3, and I'm like 80% confident I tried the same sort of TUI experience as CC from some random user before Claude Code even became public.
Who wouldn't exists if someone else didn't invent something else, which wouldn't exists...
We're all standing on the shoulders of giants here, I don't think one party is more responsible than someone else, unless you're specifically involved with the specific technology, then you can attribute it to them.
So yes, Google's researchers might have invented the Transformer, but OpenAI researchers invented GPT. Does it matter we credit "LLMs" more to one than the other? I don't think so, especially in this context it's highly irrelevant. Google didn't have the "LLM lead" before LLMs even existed...
Gemma 4 31b is better for coding than Gemini in my limited testing on a small C project single source file project, less than 1000 lines. Setting temperature to 0 gives better results for me. It seems like Gemini ignores the system prompt more and the default reasoning output seems more incoherent.
Apple may be behind, and even getting sued for false advertising around AI features, but at least they haven’t spent hundreds of billions of dollars with no indication of how they’ll make their money back.
Because their strategy wasn’t to become leaders but to be as good as it takes to erode the lead of others. They have the cash cow of search so they don’t rely on AI to succeed. All they need is to keep publishing new products/services to keep OpenAI from taking the initiative. Between that and the Chinese models all they have to do is wait for the bubble to burst at which point every major AI lab would go bust.
I had the same experience. I could not figure out how to use the IDE mode in the new version. Turns out this is a bug. It was not supposed to remove the IDE automatically, instead a user could click on "Keep the antigravity IDE" as shown in the Demo Video (at 1:09 in https://www.youtube.com/watch?v=6C0FjHoN3qE). Clean install and disabling auto update solved the problem.
Google Enterprise accounts are sunsetting AI Ultra in favor of consumption based pricing at the end of the month. It’s unclear how limits for AI Ultra might change for gmail users. Flash3.5 is much better at coding, but also more expensive the pervious flash models.
Or GLM or Kimi, Mistral is also surprisingly passable. Or just have to open the wallet and give money to OpenAI or Anthropic for the subsidized tokens.
> Google Enterprise accounts are sunsetting AI Ultra in favor of consumption based pricing at the end of the month.
This whole thing feels a bit like what GitHub did with Copilot, though.
Recently I started to get harassed to upgrade. Big button in gmail, large notifications on top of my mail in the mobile app etc. Also two other buttons to get me to turn on AI features I don't need.
I already pay a lot, I don't want to pay double just not to be harassed.
Having buttons to features that I would have to pay extra for is one thing. But having notifications and large buttons to upgrade when I am already a paying customer is harassment.
> The 2.0 update, it turns out, aggressively rewrites the default application paths to the point where it's impossible, at the time of writing, to have both versions of Antigravity installed and functioning at the same time.
Maybe it’s an OS difference but on my Mac when the new crappy antigravity updated, I got a very helpful dialog box explaining the changes and offering to download and install Antigravity IDE. Of course I did so and both run happily at the same time. Well, they did the one time I launched both, but now I’m back to just using the IDE.
Cursor did this IDE -> Agents transition very well.
Cursor still supports both the IDE and the Agents window, open at the same time, in the same project. I frequently use both and switch back and forth between them. They also link to each other from the top bar and right-click context menus so you can switch to one or the other seamlessly. Best of both worlds. Switch back to Cursor.
I want to Ask HN relating to this: What can be the motivation behind this change? Is this the preferred way of using AI coding tools nowadays? I've been using Antigravity mainly because of its tab completions. So I can work in code like in a traditional way and AI assists me. But it was a broken experience and now they are moving away from IDE based tool. The alternative is you write the prompt and it does everything. Is this the standard SW development workflow in 2026?
Yes, this is the standard model for the big frontier models. You don't need Gemini or Claude to do tab completions. A modest size local model can do that just fine. If that is all you are using AI tools for you are wasting money subscribing to Google.
The success of generalized agents is causing this change. Anthro and OAI have both made heavy pushes into expanding the use of agents past coding with Codex App and Claude Cowork. Maintaining an IDE is overkill in an age when most people are not coding by hand anymore.
This is how they want you to use AI-powered apps. The more ambiguity there is between you and the end result, the likelier you are to keep paying them to avoid friction.
The problem with AI products vs other rent-seeking is that AI is very expensive to build out and run… so they are desperate to push you into relying on it quickly.
It's not even good, honestly.
I was using it for couple weeks before dropping that 2 months ago. The model was not good and slow, the harness was not good, the IDE was subpar vscode clone.
If IDE still important for your Workflow, Trae of Cursor offer much better interface, harness and plan.
I had the exact same experience, on Windows had to purge everything and lost all my history, on Mac it was a one click upgrade and sign in again for the most part with history gone as well.
Overall the experience was pretty bad for what is expected from them and I'm wondering what the thought process behind this is, I dislike this single prompt box review workflow and is a reason I don't use any of the tui stuff and it's odd that they are leaning so hard to mimic CC when others like cursor are embracing the same workflow but still sculpting around the code. I want to edit as I'm working and have access to all my normal tools and fragmenting my work to this new vision and a separate text editor defeats the point.
For now I'll probably switch to using it as a fallback when I've exhausted my quota elsewhere and start to rely on it less before the next rug pull when I wake up and the IDE is gone. Aside, Gemini has been surprisingly good and I really liked their take on the implementation and review workflow.
It’s like Google Reader all over again. Because of all these changes, I had to cancel my Google Workspace Ultra plan and switch to a personal developer ultra plan to use Antigravity on a subscription basis, but I still have to use gemini webchat on the workspace, because there is no way to get total privacy from the individual plan. At least they prorate the cancellation and credit the unused time period.
Anyone ‘fully plugged into the Google ecosystem’ is going to end up being milked by corporate when shareholder pressure for revenue increase goes up. Same with the Apple ecosystem. Of course the language manipulation here is amusing - it’s not an ecosystem, it’s a company town where you have to do transactions in scrip that’s not transferable to another company town. Prison is not exactly the right word, either - you are free to leave, you just have to leave many of your assets behind when you do.
Sadly since couple of years or so ago we forgot about UX. Or quality in general. I have a companion which tells me I did everything right before pushing to prod. WCGW
Every time I update my JetBrains IDEs, they obliterate my lovely, tool packed UI and replace it with what looks like a minimalistic iPad app.
I have to reenable a “Classic UI”
plugin to fix it. This is annoying enough, but if they did something like the OP’s experience they’d lose a paying customer of 14 years overnight.
IDEs aren’t social media apps- they’re tools. Familiarity is not just important, it is VITAL.
Reminds me of the "dead dove do not eat" scene from arrested development. The surprising thing is not that Google is doing this, but that people are surprised by it.
you dont have to go look at the Google Graveyard [0] to understand that you might try a google product one day or month to have it either disappear or become a different product incompatible with the first the next month. They have been known for this for at least decades now.
Gemini CLI was fun for five minutes of testing until it tried to rewrite my whole code base.
Edit: I forgot to mention the curiosity and humility they bring to our calls. If I point out another vendors approach to a problem that we have, they always lean in and want to help improve their offerings from our feedback. They know it's not just enough to "be Google".
This is a huge surprise, never thought I would see this in my life time.
See https://killedbygoogle.com/
As opposed to the usual, figurative sarcasm. (Just kidding.)
It seems like Google is hitting the reset button on the product they call "Antigravity", existing users be damned. Fine, if you've never installed or used the previous version before... but for existing users the "bait and switch" is incredibly disorientating.
My take is they saw the market size for a general agentic tool as being larger and more significant than a specialised IDE. It shows a pretty large lack of respect for users in the later group though.
Uhhh, about that :)
Gemini CLI (the open source cli) is being deprecated, and the recommended replacement is Antigravity CLI (which supposedly comes with the new Antigravity, not the IDE). shrug. Surely this will be maintained long term...
Hot take: At least they're ripping the bandaid now instead of stringing users along and eventually abandoning it like they normally do.
https://developers.googleblog.com/an-important-update-transi...
No wonder they are losing massively to Huawei in several markets. Mobile marketshare is probably an indicator of some kind of their future prospects.
I don't like Google either, but I don't think this is a fair comparison.
It's easy for anyone to beat Google in China when the state has decided to block their servers.
Edit: Probably the high end non apple market in nearly all African countries too, but idk if there is reliable data for those.
Your coding environment stands a lower chance of disrupion when you use an open source IDE with a CLI agent. Yes it's slightly annoying to separate the agent from the IDE but the benefit is that it's much easier to switch between Claude Code, Codex, Gemini CLI (now antigravity CLI), etc which means you can more easily benefit from pricing and coding performance differences which seem to change monthly.
> This unexpected shift completely broke my preferred workflow
it might not have been so unexpected if you knew you were one of ~15 people that start their day with Antigravity
Google's lack of focus is astounding. They sprinkle random products here and there and seem to then tepidly pick the product surface that is doing least bad and then tepidly focus on that. Compare that to every other AI lab, large and small that knows its identity and shaped its products around that.
Perhaps it's a sort of resource curse. Google doesn't need any one of these products to succeed, and it shows.
Place your bets now.
But I have to say that I never understood the Antigravity IDE. I much prefer using Gemini CLI in combination with vscode. It works like a charm. Now, I'll do the same with Antigravity CLI and vscode. It works fine.
Because google can't help but constantly shoot its customers and itself in the foot.
What lead? Maybe because I'm mostly using AI/LLMs for development, but neither Google, Anthropic, xAI or anyone else has ever been in the lead, OpenAI always had the best models in my mind, as long as you're comparing the "top" plans between all of them.
Besides, they all seem to shoot themselves in the foot, OpenAI included, seems the only thing that differs is how often and how big the damage is.
Be skeptical of anything you read online, not just what you think is "obvious astroturf".
(on a serious note, do you feel comfortable naming and shaming such companies, this is sort of a serious accusation imo and if not then how much money they are trying to give. It would be an interesting discussion and feel free to mail me if its confidential, waiting for your response and have a nice day :-D)
Anthropic's stuff been useful for the last two years I'd say, especially in the beginning of Claude Code, but as soon as the Codex TUI was available, I was daily-driving both of them, literally executing the same prompts for each of them and comparing the final results, and Codex simply writes better code in 9/10 cases (but still not always).
We've been experimenting with "agent harnesses" way before that though, I'm sure the first time I tried building that sort of thing was in 2023 sometime with GPT3, and I'm like 80% confident I tried the same sort of TUI experience as CC from some random user before Claude Code even became public.
We're all standing on the shoulders of giants here, I don't think one party is more responsible than someone else, unless you're specifically involved with the specific technology, then you can attribute it to them.
So yes, Google's researchers might have invented the Transformer, but OpenAI researchers invented GPT. Does it matter we credit "LLMs" more to one than the other? I don't think so, especially in this context it's highly irrelevant. Google didn't have the "LLM lead" before LLMs even existed...
Also, if we're going backwards, who invented neural networks, does that mean that person also then "had LLMs before OpenAI existed"?
So much for AI getting cheaper.
For now, that's DeepSeek: https://api-docs.deepseek.com/quick_start/pricing/ (they have a discount until the end of the month, even after that they will have pretty good prices)
Or GLM or Kimi, Mistral is also surprisingly passable. Or just have to open the wallet and give money to OpenAI or Anthropic for the subsidized tokens.
> Google Enterprise accounts are sunsetting AI Ultra in favor of consumption based pricing at the end of the month.
This whole thing feels a bit like what GitHub did with Copilot, though.
Recently I started to get harassed to upgrade. Big button in gmail, large notifications on top of my mail in the mobile app etc. Also two other buttons to get me to turn on AI features I don't need.
I already pay a lot, I don't want to pay double just not to be harassed.
Having buttons to features that I would have to pay extra for is one thing. But having notifications and large buttons to upgrade when I am already a paying customer is harassment.
Maybe it’s an OS difference but on my Mac when the new crappy antigravity updated, I got a very helpful dialog box explaining the changes and offering to download and install Antigravity IDE. Of course I did so and both run happily at the same time. Well, they did the one time I launched both, but now I’m back to just using the IDE.
Cursor still supports both the IDE and the Agents window, open at the same time, in the same project. I frequently use both and switch back and forth between them. They also link to each other from the top bar and right-click context menus so you can switch to one or the other seamlessly. Best of both worlds. Switch back to Cursor.
Google could easily A/B test half of their users away from their products and nobody would get fired for it
This was clearly an experiment or stepping stone, they were never going to stick to this path. It was always going to go away.
The problem with AI products vs other rent-seeking is that AI is very expensive to build out and run… so they are desperate to push you into relying on it quickly.
Overall the experience was pretty bad for what is expected from them and I'm wondering what the thought process behind this is, I dislike this single prompt box review workflow and is a reason I don't use any of the tui stuff and it's odd that they are leaning so hard to mimic CC when others like cursor are embracing the same workflow but still sculpting around the code. I want to edit as I'm working and have access to all my normal tools and fragmenting my work to this new vision and a separate text editor defeats the point.
For now I'll probably switch to using it as a fallback when I've exhausted my quota elsewhere and start to rely on it less before the next rug pull when I wake up and the IDE is gone. Aside, Gemini has been surprisingly good and I really liked their take on the implementation and review workflow.
LPT: You can get to prod faster by skipping the step where it tells you anything.
I have to reenable a “Classic UI” plugin to fix it. This is annoying enough, but if they did something like the OP’s experience they’d lose a paying customer of 14 years overnight.
IDEs aren’t social media apps- they’re tools. Familiarity is not just important, it is VITAL.
--someone important
So just restore it from your repo.
Gemini CLI was fun for five minutes of testing until it tried to rewrite my whole code base.
[0] https://killedbygoogle.com