0:00 You know, when I first saw the demos for Google's new IDE product called anti-gravity, I was genuinely excited. Not hype train excited, but actually interested because there are a few ideas in there that looked like real progress for what an agentic IDE could be. The ability to manage multiple separate projects with multiple agents working on them all from a single centralized window. That's something I've actually wanted. And the review process where it can automatically fire up Chrome and record a screencast and check the agents work. That's a genuinely clever approach to verification. So, I cleared some time this week and I gave anti-gravity a real try. And I have to be honest with you, what I found was dysfunctional. I almost decided not to even publish this video because I couldn't even complete the demo that I had planned. But in the end, I wanted you to see what I'm seeing,
0:46 both the good and the bad, because there are some legitimately good ideas here for pushing forward what an agentic IDE should be. And I have to give the team at Google credit for that. But if [clears throat] I'm honest, I'm also kind of surprised to see a product this rough around the edges ship from a company as large as Google. Not to mention the fact that they forked VS Code, a Microsoft project, to build it. That's an interesting choice as well. So today, you'll see my first look at Google Anti-gravity. Who knows, maybe some of these ideas might actually stick. Maybe they'll fix the bugs and this thing becomes actually usable. Or maybe Cursor might take a few cues from these genuinely good concepts here and execute them in a way that actually works. I want to hear what you think in the comments. Now, before I dive in, if you're new here, I'm Brian Castle. I help professional developers stay ahead of the curve when it comes to building
1:32 with AI. And every week, I send my builder briefing. That's a free 5-minute read to help you stay sharp. Get yours by going to buildermethods.com. And subscribers can get in on my live workshops. This month, I'm covering my AI first design process. So, you can join me live or grab that recording if it's already passed. Okay, let's get into it. All right. So, let's fire up anti-gravity and we'll see what we're working with here. Now, when I open it up, I'm looking at what they call the editor view, which is very, very similar to cursor and VS Code. As I said, it is a fork of VS Code. But the other main view is what they call their agent manager. And I would say that between these two views, the agent manager is where Google Anti-gravity is really trying to innovate and really push their
2:20 vision for what an agentic IDE should be. And so really the whole point of using anti-gravity would be to use it in this view in in the agent manager because my first impressions of the editor view again are just exactly the same as you know VS Code or cursor. Uh, I'm not really seeing a whole lot that's like different or a good reason to use this view. And to be honest, I I really do wish that this agent manager view and the and the editor view were sort of merged in the same window because as I'll talk about here, I am pretty interested in some of their ideas for how an agentic IDE should be laid out here in the agent view. Um, but I do like how in cursor I could easily flip between them in a single window. Now
3:07 before I get too deep into the unique aspects of uh anti-gravity you know since this is a VS code uh fork that that means that you can import your settings and your key bindings and your extensions. So for example so if I open up the user settings you know by default when I when I just fire up anti-gravity for the first time this is empty. So I haven't customized any user settings yet. But if I go over to cursor and I open up the user settings there, which you know, as you know, cursor is also a VS Code fork. So I can grab all of my user settings, this JSON, and just drop it in here. And once I save that, it's going to ask me to restart because, you know, it changes the uh the layout a little bit. So there we go. I mean, that
3:53 that changes some of the sizing, a little bit of the colors, and I've got some other settings in there. The other thing that I'll want to do is get my key bindings in there. So, I'll I'll go to the keyboard shortcuts JSON. And I have a bunch of custom key bindings that I've set up in cursor and VS Code. Um, and I'm going to drop these in my keyboard shortcuts JSON in anti-gravity. And I'll do that. So, now I have all my all of my common keyboard shortcuts that I'm used to from my days of working with VS Code and cursor, which is my everyday IDE at the moment. So, you know, like for me, I like to use like command T to pop up the the terminal, command L to open up the chat, uh, commandB to open up my file explorer, and I've got a whole bunch of other key bindings that I use every
4:39 single day, all the time. And in fact, I did an entire video about how I try to stay fast and stay on the keyboard without touching the mouse in cursor. I did that a few months ago, so you can check that out on my channel. All right, so I've got my custom settings and my key bindings imported into Google Anti-gravity the way that I'm used to having them from VS Code. So that's kind of nice. But let's get into some of the more unique features that anti-gravity has to offer. And what I'm always trying to look for when I'm evaluating a new tool is how useful is this compared to what I'm currently using. Now I would be coming from cursor. So everything that I evaluate here is going through the lens of like is this actually better than cursor for my particular workflow when
5:26 it comes to building products every day using AI. So let's open up that agent manager and I can do a command E to open that up. And you know, one of the very first things that popped out to me about about anti-gravity that seems different and maybe better, something that I'm I'm really intrigued by is the idea of having multiple workspaces here in the sidebar. You know, I I work on several different projects from, you know, from day to day. I'm working on my builder methods site every day. Of course, I've got my personal site. I'm always spinning up new apps and working on things. So, it's really nice to have different workspaces because when I am using cursor, what I'm often doing is just opening up, you know, multiple cursor windows and throwing them across my different uh desktops and flipping
6:12 around with different windows for each individual workspace. I kind of like that from a mental model because like every project is completely separate from a workspace standpoint, but I really like the idea of at least trying to consolidate all of my different projects and different agents working on different projects in a single sidebar like this. So that's kind of interesting. And here in this agent manager, we have this, you know, centralized chat, which is very similar to the new cursor 2.0 uh agent view or whatever they call that. here in the dropown, I can open up any of the the workspaces that I have opened over here. And so that would sort of So if I want to like flip over to my builder methods project, I would do that. And now I'm I'm going to kick off an agent with that one. Or if I wanted to go to this uh
6:58 little demo app where I was listing out baseball teams, I can do that. Now I'm going to come back to interacting with different workspaces and agents here in just a sec. But one other new concept that I really kind of like about what anti-gravity is is proposing here is this idea of the playground because this really matches something that I do almost every single day, which is quickly spinning up new, often temporary projects to use as demos or to test something out, but not really a project that's going to turn into something. Or maybe it will. So rather than going through the whole process of like actually spinning up a whole new project and creating a GitHub repo for it and and creating a workspace for it, I can just start coding. I can just start
7:43 hacking on something in the playground. I can just hit that. And I was Let's go ahead and have Gemini 3 spin up a simple demo app. I'll use dictation for this. Please spin up a front-end client side only uh vit and react app. also use shad CN components and Tailwind CSS. Let's create a simple to-do list manager with checkable check boxes and uh the ability to add new items to the list. I know, I know it's just a to-do list app. It's super simple, kind of boring, nothing exciting here. Uh but the purpose of this video and this demo is really to just kind of kick the tires on Google Anti-gravity and see what these features are all about. So, let's just get that going. The other thing, by the way, is,
8:29 you know, Google is providing some token usage for for Gemini 3 when you're using Google Anti-gravity, but I have noticed that just with a little bit of use earlier today, I' I've already kind of burned through that token limit and I had to wait about an hour until I can start using it again. So, that's why I'm just going to try to keep it simple here and keep it pretty light on on the agent token usage. And that was just a simple uh request. And I can see agent taking unexpectedly long to load. All right. What happens if I hit this button, open editor? Okay. So that that takes me back to, you know, the editor view and and I'm still looking at this temporary, if you will, uh, playground app, but I don't really know what's happening. So, you know, I' I've been running into a lot of these little clunky UX issues
9:14 when I'm just trying to play around with anti-gravity. Like I I I just got to be honest. My first run experience throughout today and yesterday playing around with it just doesn't really click too much. All right, let's try to get this going. And you know, I can't even seem to stop it either. I don't know. Should we try the old quit it and reopen it? Let's try that. All right, so I'm going to go back to the playground. I'll just try this again. I'm just going to say hello. Is this thing on? Okay, so after a little bit of a wait, it did respond. So it it looks like we're we're up and running. how you know and this is clearly a bug because this little MLB teams page was a test playground that I started earlier and I had already moved that into my workspaces but now it's calling this new playground by the same name which is confusing and doesn't seem
10:03 correct to me but we'll keep rolling with it. Let's let's just try to get it to actually spin up an app that I can build something on. Spin up a client side only app with Vit React Shad CN and Tailwind CSS. Make it a simple to-do list manager with checkable tasks and titles for each task. And you know what? Let's actually try out the planning mode before we actually build that and take a look at that. One thing I've noticed, I don't know if it's just like today and it's sort of high traffic, but Gemini 3 seems to be working even slower when I'm running it in anti-gravity compared to when I been when I've been using Gemini 3 in cursor. Yeah. No, this this totally
10:49 crapped out. Okay, here we go. I think we are finally actually working. Let's see how it does. Third times a charm here. I mean, maybe it's because it's a playground. Let's just try creating a new workspace, I guess. All right. So, uh, normally when I start up a new project, I would use the terminal to do that. Maybe to make a directory or to clone a GitHub repo or something. Uh, in this case, for a super simple demo, I would want to open up a terminal and and and make a new directory. Unfortunately, in the agent manager view, I can't just do that off the bat. So, if I try to hit uh command J, which is supposed to open the terminal, I can't do that. And I think because we can't actually open up a terminal until I'm already in a conversation, like an active conversation with an agent. Um, which is
11:36 kind of baffling if you ask me. So, for example, if I'm in here, so this was something that I worked on earlier. I I added a feature to my builder methods app. Since I already have an an active conversation going in one of my workspaces, I can open up the terminal now. And of course I'm of course it it assumes that I'm in my builder methods directory. So you know here I I would be able to you know cd out of it. But again just like in terms of like context. Okay. I like I like the idea of being able to flip between multiple workspaces and even conceptually the idea of having a playground all in one all all in one window here. But if it requires actually like talking to an agent before I'm even able to open up a terminal and you know
12:22 either start up a new folder or change the the git branch that I'm on. That just doesn't really make any sense to me. You know, of course I could just like get go back to the editor view and again I'm still in one of my workspaces. I'm not in my like general agent manager view here. So just the the context and the management of these different windows, you know, like we still have multiple windows happening. So, it sort of defeats the purpose of having this one centralized uh manager. Anyway, let's let's at least try to get something going here, right? So, I'm going to uh sort of CD to my to my temp folder and I will Okay, so I've just created a folder and now let's open up a workspace for that. Okay, here is demo app. Now, we're in an actual workspace. I would have liked to start this off as
13:08 a playground and then convert it to a workspace, but that just didn't work here. So, I'm going to go ahead and create a demo app. And again, like I can't even open up the terminal here until I start talking to it. So, let's just try this one more time. And let's do that in planning mode and and we'll see it create a plan. Okay. So now that I'm talking to it, now I would actually have the ability to open up the terminal and if I if I had git initialized, I could, you know, change branches if I wanted to, which again, like that workflow just doesn't really work for me cuz I would want to change the or or start a new branch before I have the agent working on something. But the way that anti-gravity is set up, I can't actually open up the terminal and create a new branch until the agent has already started working on something. Very
13:55 strange. Okay, so it looks like it has started to create the implementation plan. One other concept that I do like about Google Anti-gravity is this idea of artifacts. If I so I could pull open this sidebar here and this is a really interesting concept for an agentic IDE and cursor doesn't really have something quite like this where you know we have these artifacts like the like planning and there and sometimes when we get into the review stage there are like media videos or pl of like browser recordings or browser screenshots those would show up as artifacts. So I do like the idea of having these like planning documents and review documents as well as like a list of file changes having that
14:40 separate from my files explorer. You know, typically in cursor I would be saving plans or saving specs as markdown files in the codebase itself, but I do like the idea of having them like visually separated into this into this artifacts sidebar. So that's a really cool idea. I kind of wish that cursor would pick pick that up. Now, something that's also interesting, like I thought I was still in planning mode, but it appears to have just gone ahead and started uh implementing the app, which is which is fine, but that's not what I expected from a user experience. So, you know, like there are just a lot of rough edges, and I'm pretty surprised to to see that kind of thing from a huge company like Google, to be totally honest. Again, one of the unique ideas
15:26 that anti-gravity is bringing to the table here is the idea of being able to comment on, you know, highlighting individual lines and leaving a comment and bringing that comment into your conversation with the agent. Again, that's something I haven't really seen. Um, you know, cursor does have the ability to sort of like highlight and and do some like inline agent coding, but this is this is more of a conversation with your agent and it makes it really easy to reference specific details that might be off or need some course correction in your plan. And hopefully later I can get to a point where where I have a visual review thing that I can comment on as well if we ever get there. I'm going to accept this. I would have liked that to be a little bit more like alert like this is waiting for my attention. All right, so it's still working on this. It's been
16:12 several minutes and again I did a version of this demo using cursor earlier today sort of in prep for this and it handled spinning up a simple app like this in a matter of minutes. Even using Gemini 3, it it just seems to be just moving a lot slower when I'm using anti-gravity. One thing I do like about the feed here is that the the interface seems a little bit more intuitive. I like how these are like collapsible and they and they and they and they animate a little bit. I've got this little arrow here. So, that that does make it a little bit easier for me to like visually scan what's going on compared to some of the other tools where I'm just seeing a feed of text fly by and I and it's sort of hard to to get a read on what's happening. And you know what? Going back to the issue with the agent just go going ahead with the proceeding of of of doing the plan. I do see this
17:00 setting up here. Oh, I just lost it. Oh, yeah. Over here, agent decides. looks like always proceed agent decides or request review. So I thought that that would that would relate to just like asking my permission to you know install things or run commands. I didn't know that this would also result in making a plan and then having the agent proceed to building that plan. Um that again that just seemed like a little off to me compared to what what the normal workflow I think should be when we're talking about spec driven development, right? I want to think about the spec, make a plan, think about the plan, review the plan, and then I want to be the one to say go. Um, and so even if I'm giving it permission to run commands and stuff, I still want to be the one to initiate like, yes, we're going to go ahead and build this now. And I would
17:46 say at this point, like it's been almost 10 minutes of of waiting for it to spin up an extremely simple demo application, a to-do list app. Oh, okay. So, here's where, you know, the interface kind of got in its own way. Like, something was like this is waiting for me to accept, but I didn't even see that because it was tucked away. See, it was tucked away inside of this expand thing. I And I only happened to notice that one step requires user input. I mean, that should be alerting me in my face. That's saying that's saying like this is not making progress until you open this up and click expand. I don't know. I'm trying to stay positive. Trying to find the positive in in this. Let's see if I can make this always. I'll say always proceed. I'll accept that one and maybe
18:33 that'll make the future requests just go through. Oh, [laughter] I hit the limit. We haven't even finished. Oh, man. You know, speaking of selecting models, this is almost a dealbreaker. I say I would say it is a dealbreaker for me. It's just the lack of selection here with the models. Obviously, Google is going to give us Gemini 3 Pro as as an option. Uh they do have Claude Sonnet 4.5, but they don't have Claude Opus 4.5, and I I have a Claude Pro account and and I would be able I would like to be able to use that in my IDE, but apparently that's not an option. I do sometimes use GPT5, but that's also not available here. And I did go into the settings to try to see if like in cursor where you know you can
19:20 add and remove all sorts of different models um that we don't seem to have that option here in in in Google Anti-gravity at least not right now. So, you know, pretty limited. But let's see if we can actually get this simple demo app up and running. Hopefully by the end of this video, let's see if we can have Claude Sonnet 4.5 pick up the pieces here. I'll just say continue. Now again, I did set it on always proceed, but it's still asking me for permission to to to run some things. So, okay. Okay. So, so Anti-gravity has [clears throat] launched its Chrome browser, and you can see the uh the the blue glow around it. Of course, I'm looking at an error. It's not actually the the simple app is just not actually working yet. Let's see if it can work through the issues that it's
20:06 that it's finding there. But this browser is being controlled by uh I guess this is sonnet 4.5 through anti-gravity. Okay. So this just looks like a complete fail if you ask me. It it g it says that it it's complete. I see we have a walk through and this is what features supposedly were implemented and it gave me a screenshot of it not working. I mean what what are we talking about here? So, I mean, what if I just run this myself? Whoa. Okay. I thought the same I thought clicking that would Wow. I can't clear. Okay, that's not even clearing the terminal. Let's at least
20:54 try to get it working. What is this? That [laughter] That is not a Tailwind class. So, I don't know what's happening there. Um, of course it seems to be using Tailwind 3, not Tailwind 4. Okay, so now we're up and running. I would have liked to to show the actual review process that anti-gravity offers. That was one of the things that I was most excited about when I when I watched the uh the preview videos for anti-gravity was the fact that it should be able to open up your app and use it and record its own screencast to to review and and and you know show that it's working in this case. It tried to do that but it clearly was not working. Um and yeah, so now one
21:40 more aspect of of this agent manager view which again I I like I like it in concept. I like the idea of being able to have access to all my workspaces and be able to flip between them, which is something I actually do in my in my workflow day-to-day. But I do often want to see the actual files, right? I want to see I want to be able to pull open a file and either edit code myself or at least just review how the code and how the project is organized even if I'm not the one coding it. Right? So if I'm here in agent manager view instead of opening up separate window like I I could open up this this demo app I can click open editor and that's going to open up the traditional VS codebased editor here but
22:25 that is already a separate window and that sort of defeats the purpose of having a singular window to manage multiple workspaces. Oh, my question is, can I actually open up files and look at look at my project files while I'm looking at the uh agent view or the agent manager? Um, the answer seems to be sort of. I can, you know, I can open up single single files here. Um, so, you know, I can use the search if I if I know the name of the file, I can I can find it and then open it up and then, you know, I can look through it. And again, we have that really nice commenting interface. So, that's really cool. What I'm missing is the file explorer. You know, if I go back to the actual code editor here, I could
23:11 actually see my project. I can see the folders. And even though I might not click through this very often, I do often look at it to see the structure and also to like find a lot of times I'm looking for sections of code that I don't even know what the file is named and I just have to sort of like visually look for it. Um, and I it's it's nice to know that I can open up single files, but I still want to be able to just look at my at my file exp I still just want to be able to look at my file explorer when I'm in this agent manager view, if you will. So, that's Google anti-gravity. And look, I wanted to like this. The ideas here are genuinely good ones. multi-RO management, multiple agents in one window, that screen recording review process. These are real
23:57 innovations that push forward what an agentic IDE should be. But ideas only matter if the execution is there. And right now, it's just not. Maybe they'll ship fixes and this thing becomes something worth using. And I hope that they do. But if you're looking for an agentic IDE that actually works today, the one that I use as my daily driver, then you need to see what the team at Cursor just shipped with their 2.0 version. It's a much better take on an AI first code editor. And I did a full review of Cursor 2.0 and the features that I'm actually using in my daily workflow. So, right after you hit subscribe on this channel, I'll see you over there next. Let's keep building.
