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".
I also have a great experience with Google account reps.
Other companies usually have a soulless void of an automatic system which gives you no confirmation your messages and inquiries are received whatsoever. Not Google. There is always a human on the other side so you know you are in good hands. Trust and connection are the things I value the most in this very two sided relationship.
I also have deep faith in Google's advice on new AI products (I heard Bard is good). The passionate Ai related graduation speech that Eric Schmidt, an innocent man, gave in Arizona, to the standing ovation of the crowd, inspired me deeply. I am now an even bigger Google fan than I ever was.
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.
Closed source IDEs are if anything the norm: Visual Studio, Android Studio, XCode, IntelliJ, CLion, PyCharm, etc... Even in the "fancy text editor" category things like Sublime were always popular enough.
Fwiw, the (mostly) closed source jetbrains IDEs support multiple models with their coding agents, byok, and using different agents like Claude Code via ACP
Fair, the important distinction is agent-agnostic rather than open-source. There are other risks to using a closed source editor but those are mostly orthogonal to this discussion.
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.
The day my coworkers started using cursor I started to learn neovim. Every day that passes I'm more glad I did it.
And mind you, I'm not an anti AI extremist. But I dont think there's any need to adopt the new tool as your new full workbench, a Claude style chat in a nearby terminal has the same benefit and exposes you to a ridiculously smaller personal risk.
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.
Could you give us how many companies are trying to do this and also if any of the companies are YC companies themselves or not, I imagine not but still.
and what is the metric for companies sending you messages, like I have never gotten a single message (aside from one/two companies here and there and I even made a HN post about one of the companies)
and what do these companies really have a metric for in terms of sending spam for? karma points, I mean emsh I remember we both had close enough about the same karmas not too long ago, surprised to see you at 13k+ karma, so good to see that but is the metric karma, hype (you had made the rust browser ..) or what exactly? I would be curious to hear your thoughts on that!
I do understand the point of these companies sending mail though, I mean I can't say that if I had a company at the moment I might not do the same either, but I think that you might get frustrated too with it, so what would your recommendation be to people sending you mails in general?
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...
> Great, tell me again who put the Transformer into LLMs?
Google did, as they already said.
OpenAI was better at marketing and a lot more willing to cannibalize the search market as a newcomer. So Google blew their lead in research by not recognizing the product value quickly enough, or failing to win an internal political war on it anyway
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.
It feels like a change in department leadership and management, or an internal power struggle over a lucrative piece of the project (with all the consequences that typically come with it). In the end, it seems more about satisfying personal egos than serving the product, and the end users will be the ones left to “appreciate” the results.
> 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.
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.
AI is powerful, but currently does not meet the engineering bar for quality and thoroughness. We need new paradigms and tools to support a new relationship with the codebase as an artifact.
The premise is that we can use these LLMs to get real engineering work done if we make tools to support a higher-level human understanding of the codebase, and the ability to spot the gaps in the LLM's plans. With these we can surgically ensure all the critical considerations are covered, spec the work at an incredibly granular level, and implement our plans as a collection of ultra-tiny tasks each given to isolated agents, this specifically ensures the agent's attentional mechanism aren't overwhelmed/polluted.
The project is very early still, so if you're interested, please reach out or signup for the email-list and i'll contact you.
Pricing page is highly aspirational at the moment, money is not the focus at this phase.
> The project is very early still, so if you're interested, please reach out or signup for the email-list and i'll contact you. Pricing page is highly aspirational at the moment, money is not the focus at this phase.
Why do you think an IDE is the right tool?
I'm working in a similar space, and it's not clear why an IDE would benefit.
Specifically to you - if you're hoping to make this a business - please know if you do make a killer IDE feature - Cursor et al will immediately copy it...
I'll give your tool a try if it's not too much effort to try it and you want some feedback. Let me know.
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.
These are models that can be run locally BTW. Just get enough hardware for your throughput requirements, have it grind on multiple batches of tokens 24x7 to get reasonable utilization (keeping the cloud for time-sensitive uses) and that's it, no more rug pulls.
I am building an Agent IDE called Harness. It is somewhat inspired by the previous version of antigravity (and Conductor, and a few others). But with a core goal being open source & hackability.
It's centered around git worktrees. The goal is to organize all your AI sessions into somewhat logical places and make it easy to context switch. The secondary goal is to remove the need to open a separate code editor anytime you want to look at a file (We have a built in file editor powered by Monaco [vscodes editor])
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.
Same. It’s really been a nothing bar for me with this cutover. I feel for the IDE people, but now I call agy vs gemini…life goes on. 3.1 Pro model still works perfectly for me and my needs, if anything I’m finding the agy cli much more responsive and stable so far
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.
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.
Yeah, that was my experience. The model was worse in every way than ChatGPT or Claude or even Composer. I tried it out and used it when my other limits were hit, but only as a last resort. And I stopped doing even that because the model was so bad.
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?
It is the new standard. It sounds awful until you try it, and then you can't go back. But you can still use an IDE as well to edit code by hand and review changes that agents have made.
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.
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.
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.
I was surprised to see that the new Antigravity does not integrate at all into Google's existing Material design system. Is the implication that Material is not for power users or developers? It's built as a universal solution.
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.
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.
It was a code editor that had Gemini integrated into it. I’ve been playing around with the new version, it can still do what I was using it for, but it does make me wonder if it will become a OpenClaw like tool.
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.
I'm very surprised, goggle are usually known for their customer focused approach and long standing support of legacy systems!
For me it is their personable account reps and customer service. It’s the human touch we’ve come to associate with the “Don’t be Evil” brand.
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".
I also have a great experience with Google account reps.
Other companies usually have a soulless void of an automatic system which gives you no confirmation your messages and inquiries are received whatsoever. Not Google. There is always a human on the other side so you know you are in good hands. Trust and connection are the things I value the most in this very two sided relationship.
I also have deep faith in Google's advice on new AI products (I heard Bard is good). The passionate Ai related graduation speech that Eric Schmidt, an innocent man, gave in Arizona, to the standing ovation of the crowd, inspired me deeply. I am now an even bigger Google fan than I ever was.
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.
The internet changed for me the day the Reader died. Actually, in hindsight, for me personally, internet died when the Reader died.
I'll echo this. They're very good at consistent support and never pulling the rug. The folks at Railway have nothing but the wildest praises to sing.
OP is lucky they aren't getting ads while the agent is working.
I'm so impressed with their support, very proactive and easy to reach. Whenever I had issues, they're always there to help.
Can LLMs detect sarcasm? When AI scrapes this thread, does its sentiment analysis get tricked?
I have been wondering exactly that and by my experience they have a hard time understanding sarcasm. It is a natural prompt injection.
As long as the legacy systems have billions of users. Otherwise they get shutdown once people run out of interest.
See https://killedbygoogle.com/
You might have missed it but the OP's comment was dripping in literal sarcasm. Google's track record for product management is poor.
> literal sarcasm
As opposed to the usual, figurative sarcasm. (Just kidding.)
“literal sarcasm” is using “literal” figuratively.
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.
You can't use Antigravity 2.0 on Windows with WSL. There is simply no way to connect to WSL. The agent can't run any Linux commands.
Also the Antigravity CLI doesn't remember your credentials in WSL. It asks you to log in every time you run the program.
They could just call it anything else and left the existing user alone. I mean they have gemini CLI, which I would say a better product.
I think that's what everyone is going to think.
Hot take: At least they're ripping the bandaid now instead of stringing users along and eventually abandoning it like they normally do.
> I mean they have gemini CLI
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...
Oh, but you can only install the new antigravity CLI by first installing and authenticating via the IDE.
Will they make it work headless before the June deadline when they turn off gemini-cli? I guess we'll see...
Gemini CLI is being sunsetted in mid-June and replaced by Antigravity.
https://developers.googleblog.com/an-important-update-transi...
Sometimes I wonder if they even realize they have users...
Pissing off the segment of people most likely to take offense and try to take revenge seems pretty dumb.
No wonder they are losing massively to Huawei in several markets. Mobile marketshare is probably an indicator of some kind of their future prospects.
> 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.
They are declining in market share in several countries. Notably multiple ASEAN countries, Russia and Iran (though that is forced), and so on.
Edit: Probably the high end non apple market in nearly all African countries too, but idk if there is reliable data for those.
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.
Closed source IDEs are if anything the norm: Visual Studio, Android Studio, XCode, IntelliJ, CLion, PyCharm, etc... Even in the "fancy text editor" category things like Sublime were always popular enough.
Fwiw, the (mostly) closed source jetbrains IDEs support multiple models with their coding agents, byok, and using different agents like Claude Code via ACP
Fair, the important distinction is agent-agnostic rather than open-source. There are other risks to using a closed source editor but those are mostly orthogonal to this discussion.
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.
It isn't anymore, though, that's kind of the whole point of the article.
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.
The day my coworkers started using cursor I started to learn neovim. Every day that passes I'm more glad I did it.
And mind you, I'm not an anti AI extremist. But I dont think there's any need to adopt the new tool as your new full workbench, a Claude style chat in a nearby terminal has the same benefit and exposes you to a ridiculously smaller personal risk.
How did Google blow their AI lead? Why is Google the 2nd or 3rd tier player in the AI coding market? Why can't GCP supplant AWS?
Because google can't help but constantly shoot its customers and itself in the foot.
> How did Google blow their AI lead?
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.
Wow. Didn't realize OAI was astroturfing hacker news now...
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.
Could you give us how many companies are trying to do this and also if any of the companies are YC companies themselves or not, I imagine not but still.
and what is the metric for companies sending you messages, like I have never gotten a single message (aside from one/two companies here and there and I even made a HN post about one of the companies)
and what do these companies really have a metric for in terms of sending spam for? karma points, I mean emsh I remember we both had close enough about the same karmas not too long ago, surprised to see you at 13k+ karma, so good to see that but is the metric karma, hype (you had made the rust browser ..) or what exactly? I would be curious to hear your thoughts on that!
I do understand the point of these companies sending mail though, I mean I can't say that if I had a company at the moment I might not do the same either, but I think that you might get frustrated too with it, so what would your recommendation be to people sending you mails in general?
I would be curious to know that too!
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.
OpenAI literally wouldn't even exist if it wasn't for Google's work in the space.
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...
Google invented transformers. They had LLMs before openAI existed.
Great, tell me again who put the Transformer into LLMs?
Also, if we're going backwards, who invented neural networks, does that mean that person also then "had LLMs before OpenAI existed"?
The tone on this could be improved. They literally answered your question "What lead?" and you seem dismissive.
> Great, tell me again who put the Transformer into LLMs?
Google did, as they already said.
OpenAI was better at marketing and a lot more willing to cannibalize the search market as a newcomer. So Google blew their lead in research by not recognizing the product value quickly enough, or failing to win an internal political war on it anyway
No, it's more that Gemini models are simply not very good for coding compared to the top two. Even with Antigravity I use Claude models.
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.
> Gemma 4 31b is better for coding than Gemini
Is there a fine-tuned Gemma coding model? I'd assume that would perform quite well.
They had the lead for maybe a week or two. Now, only Apple is further behind.
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.
> The day was to begin like any other, with Antigravity open
> 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
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.
It feels like a change in department leadership and management, or an internal power struggle over a lucrative piece of the project (with all the consequences that typically come with it). In the end, it seems more about satisfying personal egos than serving the product, and the end users will be the ones left to “appreciate” the results.
> 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.
I pay for google "Starter" workspace.
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.
And if you upgrade to standard workspace it still tells you account not eligible for antigravity
Idk what google is doing
The market demands INFINITE GROWTH
…and 0.1% of users click the button that annoys 99% of users, so it‘s a BIG SUCCESS.
there is no graph of user happiness. there is only graph of PROFIT
I'm building an IDE (www.kaiso.ai)
AI is powerful, but currently does not meet the engineering bar for quality and thoroughness. We need new paradigms and tools to support a new relationship with the codebase as an artifact.
The premise is that we can use these LLMs to get real engineering work done if we make tools to support a higher-level human understanding of the codebase, and the ability to spot the gaps in the LLM's plans. With these we can surgically ensure all the critical considerations are covered, spec the work at an incredibly granular level, and implement our plans as a collection of ultra-tiny tasks each given to isolated agents, this specifically ensures the agent's attentional mechanism aren't overwhelmed/polluted.
The project is very early still, so if you're interested, please reach out or signup for the email-list and i'll contact you. Pricing page is highly aspirational at the moment, money is not the focus at this phase.
Thanks.
> The project is very early still, so if you're interested, please reach out or signup for the email-list and i'll contact you. Pricing page is highly aspirational at the moment, money is not the focus at this phase.
Why do you think an IDE is the right tool?
I'm working in a similar space, and it's not clear why an IDE would benefit.
Specifically to you - if you're hoping to make this a business - please know if you do make a killer IDE feature - Cursor et al will immediately copy it...
I'll give your tool a try if it's not too much effort to try it and you want some feedback. Let me know.
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.
So much for AI getting cheaper.
> 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.
These are models that can be run locally BTW. Just get enough hardware for your throughput requirements, have it grind on multiple batches of tokens 24x7 to get reasonable utilization (keeping the cloud for time-sensitive uses) and that's it, no more rug pulls.
I am building an Agent IDE called Harness. It is somewhat inspired by the previous version of antigravity (and Conductor, and a few others). But with a core goal being open source & hackability.
It's centered around git worktrees. The goal is to organize all your AI sessions into somewhat logical places and make it easy to context switch. The secondary goal is to remove the need to open a separate code editor anytime you want to look at a file (We have a built in file editor powered by Monaco [vscodes editor])
Check it out https://harness.mikelyons.org
> The goal is to organize all your AI sessions into somewhat logical places and make it easy to context switch.
Isn't this what Pi does (except you have a non-CLI UI)?
and what is the harness for harness called?
Google hasn't handled this well, it is obvious.
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.
Same. It’s really been a nothing bar for me with this cutover. I feel for the IDE people, but now I call agy vs gemini…life goes on. 3.1 Pro model still works perfectly for me and my needs, if anything I’m finding the agy cli much more responsive and stable so far
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.
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.
Yeah, that was my experience. The model was worse in every way than ChatGPT or Claude or even Composer. I tried it out and used it when my other limits were hit, but only as a last resort. And I stopped doing even that because the model was so bad.
Google really outdid themselves this time. They killed not one but two tools (Gemini CLI and Antigravity) with one stone.
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?
Google corporate culture where users are just numbers someone's performance report is why this happens.
Google could easily A/B test half of their users away from their products and nobody would get fired for it
It is the new standard. It sounds awful until you try it, and then you can't go back. But you can still use an IDE as well to edit code by hand and review changes that agents have made.
I'm surprised anyone thought Google would stay committed to an IDE product built on Microsoft's VS Code.
This was clearly an experiment or stepping stone, they were never going to stick to this path. It was always going to go away.
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.
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.
I don't have time to fix the problem, let me write a blog article about it, lol
Will this experience actually have a lasting impact on how the author makes decisions?
Place your bets now.
My opinion is that Google has currently enjoys low trustworthiness as an enterprise software and services provider.
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
> I have a companion which tells me I did everything right before pushing to prod.
LPT: You can get to prod faster by skipping the step where it tells you anything.
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.
Designers gotta eat
Hey I have the same complaints about us developers adding new features that no one asked for to a product that already serves its purpose.
I wish the industry could learn the art of leaving shit alone.
Google has just stepped on the IBM path :D
I was surprised to see that the new Antigravity does not integrate at all into Google's existing Material design system. Is the implication that Material is not for power users or developers? It's built as a universal solution.
https://antigravity.google/assets/image/blog/agy2-layout.jpg
Antigravity IDE is just a better tool
"..and you will learn to like it!"
--someone important
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.
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.
> the prompt history from the old Antigravity installation is gone
So just restore it from your repo.
This is exactly why I have a have a strict blanket ban on automatic updates on all of my devices.
At this point, anyone who relies on Alphabet for anything, deserves what they get. Fool me once... and all that.
What the hell is Google Antigravity?
It was a code editor that had Gemini integrated into it. I’ve been playing around with the new version, it can still do what I was using it for, but it does make me wonder if it will become a OpenClaw like tool.
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.
[0] https://killedbygoogle.com