ChatGPT Dots Is Insane... But Gemini’s NEW Argon Is EVEN Bigger

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: ChatGPT Dots Is Insane… But Gemini’s NEW Argon Is EVEN Bigger · Channel: Riley Brown · 24:20 · Watch on YouTube

Topics: AI agents, Task and workflow automation, Software engineering with AI, AI models, Open source and software ecosystems

Summary

Riley Brown reviews four major AI-agent updates: Anthropic’s Sonnet 5.5 and Opus 5.5, OpenAI’s new agent features, Google’s announced Gemini 4 Argon, and updates to Grokbot. He praises Opus 5.5 and demonstrates how OpenAI’s Dot can coordinate tasks, create local coding sessions, and work with documents in GPT Space. He is cautiously optimistic about Argon, noting that public access is limited and that Google has yet to establish a clear flagship app for using its models. The video also covers Grokbot’s proactive suggestions, Slack integrations, team bots, and computer-based traffic routing, before promoting Brown’s new vlog channel.

Key points

Brown says Sonnet 5.5 is cheaper and reportedly stronger at agentic coding than Opus 5.5, but he still prefers Opus for demanding work.
Claude’s Projects feature lets a main agent delegate tasks to separate threads, including cloud sessions and sessions running locally on the user’s computer.
OpenAI’s Dot is presented as a personal agent built into the ChatGPT app that can handle tasks and launch Codex sessions.
GPT Space provides a place for documents, pages, sites, and images that users and agents can work on together.
ChatGPT plug-in extensions let users apply their ChatGPT account and models inside third-party apps such as Magic Path.
Google announced Gemini 4 Argon for cyber defenders, with broader access planned; Brown says he is interested but skeptical based on past Gemini models.
Brown argues that Google needs to clarify which of its many apps should serve as the main platform for its AI models.
Grokbot updates include proactive suggestions, editable email drafts, team bots in Slack, and an option to route web traffic through the user’s computer.
Brown introduces a second YouTube channel for informal videos about experiments, internal software, 3D printing, and hardware projects.

Chapters

Start	Chapter	Summary
00:00	AI agent updates and Anthropic’s new models	Brown introduces the week’s major AI news and compares Sonnet 5.5 with Opus 5.5, including his experience using Claude Projects.
04:40	OpenAI Dot and the ChatGPT app	He demonstrates Dot handling email, Slack, and a landing-page task by launching a local Codex session.
08:48	GPT Space and ChatGPT extensions	Brown shows how agents can collaborate on pages in GPT Space and how ChatGPT extensions work inside third-party apps.
15:01	Gemini 4 Argon	He discusses Google’s Argon announcement, limited early access, his skepticism about Gemini, and the lack of a clear Google flagship app.
19:11	Grokbot updates	Brown covers Grokbot’s proactive suggestions, Slack email drafts, team bots, and routing traffic through a user’s computer.
22:48	Wrap-up and new channel	Brown recaps the week’s updates and describes his new vlog channel focused on experiments and behind-the-scenes projects.

Quotes

Opus 5.5 is still the goat right now. — 04:10

Anyway, that is Dot. — 08:16

I’m skeptically excited. — 18:09

My goal with this channel is to educate. — 23:20

Concepts and entities

Anthropic, Claude Sonnet 5.5, Claude Opus 5.5, Claude Projects, OpenAI, ChatGPT Dots, ChatGPT Spaces, Codex, Magic Path, Google, Gemini 4 Argon, Grokbot, Muse, Riley Brown

Full transcript (original language)

00:00 What an insane week in the world of AI agents. Apparently, Google is on the verge of releasing the best model in the world called Argon. And only cyber defenders have access. And this comes one day after open AAI dev day where they announced 20 different updates to their Chad GBT app. One of which is their brand new agent platform called DOTS. But a day before dev day, Enthropic released Claude Sonnet 5.5. just 6 days after releasing the best

00:31 model in the world, Opus 5.5. We’re going to discuss all of these today and more because I took everything going on in the world of AI agents this week and distilled it down to the four most important things that you need to know and how they can be useful for you. Okay, so we’re going to dive straight into update number one and that is that Anthropic released Sonnet 5.5. And so if you are using claude whether you are on the normal claude or claude code you

01:03 will see a new model which is sonnet 5.5. It is the most efficient for simple tasks or at least that’s what it says in the app. If you go to claude code you can very easily use sonnet 5.5 to build any app you want. Basically if you guys watch my videos I don’t like to talk too long about model cards. I just wanted to show you this really quickly because this is the new Sonnet 5.5 performance. the model that they just announced. And you can see here that it is apparently better at agentic coding than Opus 5.5.

01:35 And this is pretty cool because Claude’s sonnet is half the price of Opus. Opus is $4 per million tokens in and $20 per million tokens out compared to 2 and 10. So this model is even cheaper than Claude Opus 5. But let me tell you, I have been using Opus 5.5 for the past week for over three hours a day. Sometimes up to like five or six hours per day. And I’ve actually never come close to my limits. If you look down

02:05 here, like my weekly limits have always been pretty low. My 5h hour limit, the closest I got was like right here. And I’ve just been running Opus 5.5 all week. Now, I want to give my thoughts on this because I’ve spent around 4 hours using Sonnet and then I’ve spent around like 30 hours using Opus. Opus 5.5 is still the best model maybe in the world, but for sure for the price. Opus 5.5 is a monster. And specifically, I have been using Opus 5.5 in their new projects

02:38 feature. Okay, I’m actually a little bit shook by this. I seem to be the only person on YouTube talking about the new projects feature which gives you kind of this main agent that can delegate into all of these different threads. I think this interface for especially for building apps and like doing knowledge work is incredibly awesome. Like I can very easily just send an agent off to like create a deck and it will just go off and it’s perfectly integrated with design mode. So, like this is the internal tool that I’m building right

03:08 now for my entire company. And it can just go off and it just designs every page of the app and you can just like tell your main agent, hey, I need you to create a deck which explains my website and how it works. And it’ll just fire off a new thread and by default, it’s going to be Opus 5.5 Medium. And it allows me to just keep this main thread right here. And it’s not super long. And then it’ll just spin off a new thread. Now it’s saying uh native note explainer deck and it’ll create the deck in this

03:39 thread and I can see all of the artifacts in the library right here. You can spin off cloud threads. So if you see this little computer right here, this one’s actually running locally on my computer. So if you need a thread to like use your computer or use your GitHub or Verscell or if you vibe code on your computer using cloud code, it can do that normally if you spin off one of these like local threads. But the rest of these can be done in the cloud. So I’ve been making changes to my app just using my phone and I can spin off

04:10 cloud sessions and it’s incredibly easy. And so in summary, when it comes to Claude, I would say Opus 5.5 is still the goat right now. use Opus 5.5, especially if you are on a $100 per month plan or the $200 per month plan. I think it’s just a great price. Sonnet adds more usage. So, if you’re using a $100 per month plan or the $20 per month plan, you might want to use Sonnet because it’s still a really good model and it is cheaper than Opus. But, if you can afford it, just use Opus. If you have the $200 per month plan, I’ve yet to hit limits. The only people who are hitting the limits are people who are

04:40 like insane. Uh like they just are coding massive projects all day long and they’re running many sessions at the same time. I’m not usually like that. I only run a couple sessions at once. So with Opus, you’ll be fine. So the second big update this week in the world of AI agents is of course OpenAI dev day. And they released a lot. And so I don’t want to talk about everything they released. I made a video on that previously. What I do want to discuss are the three most

05:11 important things that you should know about the recent updates to their chat GBT desktop app. And so if you go to your chat GBT desktop app or the ChatGBT web app or the iOS app, you’re actually going to notice a lot of things are different. Specifically, we have access to a new model which is GBD 6.1 Soul. You also have access to a new sidebar and the main new feature is dot. So, here is Bluey. This is my own little agent that operates on a computer very

05:42 similar to Grockbot, similar to Muse, except we have access to OpenAI’s models, which are actually on the frontier, unlike the other two I just mentioned. So, I could call Bluey. >> Hey, Riley. Good afternoon. >> Yo, Bluey, how’s it going, man? Dude, I thanks for all the help this morning. I we did a 35minute call talking about everything that I needed to do. We sent emails. I need you to send another email real quick and you know the summary you generated me earlier. Can you please

06:13 send that to Emily? And then also I want you to send a Slack message to her as well just to make sure she sees it right away. >> Okay, let me get that set up. I’m checking that summary now. I’ll take the summary to mean the Slack and video status update I just sent you. >> Exactly. >> And I’ll prep the email and a short Slack heads up. Now, I put the email and Slack drafts in our chat. >> Go ahead and send that >> latest summary. >> Just go ahead and send it. And then >> Yep. Sending now. >> Also, after you send it, I want you to do something for me. I want you to

06:43 redesign my landing page on uh agentnative.inc, right? I have a landing page there that kind of shows the world of different tools. I want you to uh spin up a new codec session and make this a local codec session and spin up some options for a landing page for me there. Yeah. Yeah. And I want you to do it >> just do it on a chat GBT sites as well and let me toggle between the different options for the landing page. >> Got it. Let me line all that up. >> Thanks, Blueie. >> Of course. Quick update before you send.

07:14 Emily already sent the dev day cut to OpenAI at 3. The draft’s upload blocker is outdated. Want me to update both messages to say it’s sent and awaiting review? Then send. >> Just send her the same thing that you sent me. What you’re saying makes sense. Just make sure she gets this information. Just send them both now. >> Okay, I’ll send them now. They’re going out now with the correction that the dev date cut is with OpenAI and awaiting review. >> Cool. Um, are you working on the other codeex task I gave you? >> Yeah, I’m on it. So, it’s prepared, but I haven’t confirmed the local codec

07:45 session is running yet. The local session hasn’t started yet. Once the email and Slack deliveries are confirmed, I’ll kick off the landing page task on your Mac. >> And look at this. You can see that it is spinning up a new session right here for a redesign agent native landing page. And so if I click on this, it takes me to a new session in codeex. And this is sent by bluey. And we are creating >> both messages reached and the new local codec session is running on your Mac

08:16 creating three landing page options with a toggle and a sites preview to compare them. >> Louie, thank you so much. >> Anytime Riley, glad to help. Anyway, that is Dot. And so this is their personal agent platform. Like I said, it’s like Grockbot and Muse, but again the OpenAI’s models. Now, it’s built directly into the chatbt app. So, the app is starting to feel a little bloated, but I would say especially with the voice feature, it is the most fun experience I’ve had using agents because it can spin up codec sessions for me.

08:48 And so basically I’m using a really fast model that doesn’t cost any credits towards my plan which is one thing that you need to understand until it delegates to codeex or GBT work. But anyway, see this is this really smart agent that you can use with the latest models inside chat GBT. But they also released another cool feature which is called GBT space which is related to the new dots feature. So let me explain this one. So in the sidebar we obviously have

09:19 home and I can close the sidebar. We also have space. So same way you can think Google Drive GPT space. Similar idea. If I click on this right here I can now see all of my documents. And so I can actually create a new document and I’m going to click new page. So within page you can create I can say research on dots and how it works. Open AI it’s

09:49 called a highlight on here. Open AI released uh dot and now I can very easily add an emoji and so this reminds me a lot of notion and I can come down here every page I can very easily just use AI to edit the page but I can also do with Bluey. So I can talk to Bluey. I can be like I can say yo and then remember we called that page research dots and how it works. I’m working on this doc. Can you add info there on how

10:21 it works? Use inline diagrams and text mostly diagrams. I’m a visual learner. And so this space is basically a place for all of your documents that not only you can work on those documents there, but your agents can as well. And so I’ve been testing it. How do I work on this with Bluey? And it’s really fun when you’re on a call with Bluey and you’re maybe out on a walk, right? Every morning I wake up for 45 minutes and now I’m just talking to AI in the morning.

10:52 What do I need to do? And if I need to write something down, I have a little to-do list here that’s actually in my pages that Bluey will add things to, which is really fun. Okay, look at this. It just added and you could actually see that Dot added them. It had this animation which showed Dot adding this to the board. And look at how cool this is, right? I can full screen this and take a closer look. But here it dot wrote this on the board and all I had to do was ask dot to write this in a page

11:24 and I can show pages here and just like Google Docs I can create a subpage. So this is research on dot I could create a subpage dot use cases and then I could call >> good evening Riley good to hear from you. >> Hey pal I’m going to need you to add dot use cases. It’s a subpage under the document you just edited right. >> Mhm. Let me take care of that. Yeah, buddy. >> Okay, adding that. No, >> I don’t see the dot use cases yet. >> Yeah, it’s still underway. Let me get you a status check. All right. I don’t

11:54 have the save link yet for dot use cases. The main research page with the six diagrams is saved, but the subpage is still being created. So, no link yet. >> I already created it. >> Oh, right. Right. Let me check. Okay, got it. I’ll use the one you made and stop the duplicate. I’m going to add the use case content to your subpage. Okay, I found your use cases subpage and I’m filling that one. No duplicate was created. >> Nice. You added them. >> Sweet. Let me confirm the save real

12:26 quick. Glad that they’re showing up. I’m confirming the full save now. >> Looks like it’s working. >> Nice. That’s what I like to hear. >> Anyway, your Bluey can just add things to your pages. And so currently right now the things that are included in the GBT space are pages, sites. Remember we created the sites earlier and we can also so we can click on sites and here are all of our sites that we’re creating and then spaces have your images and

12:57 then Google Drive as well. And so that’s the second main update. Right. Now, one thing I could have included was their new model, which is GPT 6.1 Soul, which is five times cheaper than 5x cheaper than Astra. And it is pretty damn good based on my testing. Like I said, I don’t like to dive too much into the model cards. I just play it by vibe. If I love using the model after hours, then I continue using it. And it is a really good model. and I’ve have not

13:27 really been hitting my limits like I was with Astra. But the final thing I want to talk about with OpenAI is just really quick overview on their like codeex revamp. They have this new sidebar and the main reason I think they have this sidebar is they have this new thing called plug-in extensions. You could always add plugins in the chat GBT desktop app, right? You could add the Figma plugin which allows you to very easily use Figma. But now they have plug-in extensions. And plug-in extensions look like this. I’m going to

13:58 show magic path as an example, but plug-in extensions are apps in the chat GPT app. And it works inside the web app as well. And so here is a design that I was creating in Magic Path the other day. And I can very easily open this in the chat GBT app. And because I’ve signed in with chat GBT into this app, I can very easily use the chat GBT app and it will control the app that I’m using. So they are truly like an agentnative

14:30 app that I can use. It’s truly insane. I can say please make a light mode version. And look at this. Chat GBT is now working inside this app. And it is going to create a light mode version. And so I’m using my tokens from my chat GPT account or app and I’m using it inside another company’s apps which is magic path which is like this AI designer. Now they have a built-in AI that can create designs but when you use

15:01 the extension it’s really easy to just use your chat GBT account. So that is a huge update and there you go it’s done. Okay now it’s time to move on and talk about a company I haven’t talked about in forever. And so the third update of this week is Gemini Argon, their new model, Gemini 4, and they’re naming it Argon. And so I told you I don’t like to look at these model cards, especially with Gemini. In the past, they’ve released models that have looked really, really good on paper. I think this was

15:33 like Gemini 3 and Gemini 3.1. They said that they were really good and the demos that they were showing made it look like it was really good, but the models just don’t end up doing the things you want them to. And so my main caveat here and my main question is, is the model really as good as Google is saying? And because the main reasons we should be skeptical is one, no one in the public can use it yet. They’re only releasing it to people called cyber defenders. And so this is

16:04 the announcement from yesterday, introducing Gemini 4 Argon, our new Frontier model, rolling it out to cyber defenders starting today and more widely as soon as possible. I’m really excited about the progress we have made here. Argon is priced at $2 in and $10 out. And notice here, that’s actually the same pricing as Claude Sonnet 5. So it seems reasonably affordable if it’s truly a Frontier model. And then I had to make a joke. I said, “Yo, can I get access Logan?” And I changed my Twitter

16:35 bio to Cyber Defender. And then guess what? It worked. He said, “We’ll try to get you in the early access test.” Which is great. Let’s go. Cyber Defender. And so I was thinking to myself, when was the last time that Gemini actually had a good model? Right? So we’re in month 10. Google has not had a good model since Jan. So at the very beginning of this year, Gemini 3 was really good and it was really good at frontends. But as a as we

17:08 used it more, especially for coding like tasks, Gemini 3 would kind of just refuse to use tools which is so important for creating agents. And because of this, their platforms never really took off. Like anti-gravity is their AI coding tool never really took off. Gemini was never that good of a model. And I really just haven’t met anyone who has used this for knowledge work. Like no one I know who’s like a founder or like a power worker. Like they do really good work and they love

17:38 to use AI. No one’s like, “Oh, I’ve been loving using Gemini.” Like the models just haven’t been that good and they haven’t been that cheap. You know, the only redeeming quality about Gemini’s models in my opinion is the fact that they have direct video input. Like you can literally upload a video to the prompt and it will like ingest like it will basically like watch the video in 10 seconds and have a deep high quality understanding of the video. No other model does that. But other than that, the Gemini models don’t really do

18:09 anything that interesting. And so if I had to describe my opinion on this is I’m skeptically excited. I hope this model’s amazing. I think competition is great. You know, we have all of these platforms like we have chatgbt desktop, we have claw desktop, obviously we have grobbot, cursor, all of these really high quality platforms. I would love to see one come out of Google. The unfortunate thing is is like even if this model’s good and we want to use Gemini, what app do we even use,

18:40 right? What app should we use? Is it anti-gravity? Is it Gemini desktop? Is it their AI studio? or should we use their notebook LM? Right? There’s so many different platforms. It’s hard to tell what platform we should be using. Like ChatGBT went all in on the Chat GBT desktop app. We knew that if we want to use OpenAI’s new models, we should use the ChatgBT desktop app. If you want to use Claude, you should use the Claude desktop app. We have no idea which app to use. And so, I would love to see them

19:11 commit to one of the platforms and make it their super app, but we’ll see what they do. And I’ll keep you guys updated in future videos. And so the fourth major update of the week comes from SpaceX and we have some updates to Grockbot that I think have slid under the radar. Okay, so this one is actually from an hour ago. Grockbot can now suggest ways to help without you needing to ask. They definitely got this from Muse. Muse is the first platform that I think got proactivity right. If you’ve downloaded Muse and Muse is free. It’s

19:42 basically their uh meta’s free version of Grockbot. It sends you these proactive updates and gives you ideas of things that it should do. It keeps track of all of your goals, which is really interesting and that’s why it became number one on the app store. And I think that’s one of the things that Grockbot or SpaceX is going to try and copy with Grockbot is making it more proactive and making that experience more seamless. And so beyond this, the next update to Grockbot are inline drafts for emails in

20:13 Slack. So, if you come over here to Grockbot, I just did this. I said draft and Emily saying Grockbot is in Slack. You’ll see here that we have this inline draft which is incredibly useful, right? It allows me to very quickly like make changes to the email and I can just send it off and this will actually send it using my email integration. And here it has send and then that information is sent back to the agent. You can also create team bots inside Grockbot then add them to Slack. Let me show you how

20:44 this works. So in Grockpot, you can see here that you have team bots and a general bot. And in a team bot here, you can see Slack is connected. Let me show you. So we can open up Slack. And if I come down here to my bots, we can see that the agent native bot is here. And I can say, hey, you are Grockbot, right? And Grockbot will immediately view the message. And you can see here that this Grockbot’s name is agent native and it says, “Nope, I’m agent native. The team

21:15 bot for agent native. Grockbot has a separate assistant of yours.” Okay, that’s very interesting. But I created it on Grockbot. But anyway, I digress. I can say like what skills you got. Agent native. And there you go. We have all of the different skills that I’ve created inside Grockbot and have connected in my account right here. So, if I were to go to connect apps, all of my plugins and skills are here. And so, this teambot has access to this, which is very cool.

21:45 And then finally, one thing that they added last week, which you can find in settings and computer, is you can actually run traffic through this computer. Let me explain what that means. So, when you use Grockbot, Grockbot obviously has this little virtual computer in the cloud right here. You can see the computer right here. And so it can actually surf the web, click around on things. However, all of these are uh running in a data center and there are certain clues. Let’s say it goes to a website like Shopify or it goes to a website like

22:17 Target and Target may want to block bots from using their site. If you’re using the IP address from this computer that’s running in the cloud, it’ll automatically flag it and not let your agent get access. when you go to settings computer route traffic through this computer it basically will use your information from your computer so if your computer is on it’s able to use your IP address and it will appear as though it is human it thus bypassing a lot of the restrictions that you may get

22:48 by being a bot roaming the internet on a virtual computer so it basically allows your bot to have more access to more stuff and so that is the final update for Grockbot so that covers all of The main updates from this week, a lot from the huge players, Anthropic, OpenAI, Gemini, and Grockbot. I hope you got a lot from this video. As you may have noticed, we do have a bonus update in the world of AI, which is my new YouTube channel. So, I’m creating a vlog. And you know, a lot of my videos are scripted and I know that. Um, and I try

23:20 to make all of the concepts in AI, all the things happening, as approachable as possible. And that’s what I do on this channel. My goal with this channel is to educate. I’m creating a new channel where I just speak off the cuff. It’s more like vlog style. As I learn things, I’ll just make a really quick video and I have an AI workflow which posts it like it edits the video, cuts the video, adds B-roll over the video, and then immediately posts it. So, it’s very raw style, not a lot of edits. I’ll be 3D printing a ton of stuff. So, on this channel, this is where I’m going to show

23:50 off all my experiments, what I’m doing behind the scenes, my internal software that I’m building for those of you who want to build internal business software. We’re building hardware devices very soon. I just got this 3D printer. I just wanted Opus and Astra to be able to like create physical objects and it’s awesome. We’re going to do a ton of fun cool things on this channel. So, if you want to follow that, the link will be in the description. You can look it up. It’s just Riley Brown, too. Posted my first video a few days ago. And I really enjoy making these videos.

24:20 It’s a completely different format than you’re used to. And uh yeah, it’s more me, I would say. So, make sure to follow that if that interests you.