0:00Not only did Enthropic release Fable 5.2, they basically released their own version of Microsoft Office with three new artifact types, Sheets, whiteboard, and data. And on top of this, you can now add Claude to video meetings, and it can take actions while you talk to your team. I'm going to break all of this down today. But that's not all that happened this week. Google just released their new model, Gemini 4 Carbon, and I have lots of thoughts about this, some
0:28good, some bad. And we also have three huge new updates to dots, one of which I've been using every single day. So, the first thing that I want to talk about today is Enthropic released three new artifact types. So, here's their announcement from 3 days ago. They said starting to roll out today, cla whiteboards, claude sheets, and claude data. Here's an example of the new whiteboards artifact. So yesterday I went to Claude Code and here was my
0:56prompt for Fable 5.2, their latest model, which we'll get to later. I said, "Hey Claude, I want you to create a new whiteboard artifact. Analyze how I edit videos starting with the computer I use to the final output." I said, "Include analyzing, scraping, and prompting. Put it in an easy to follow whiteboard." And then Fable 5.2 2 actually went through 70 different threads where I've was editing videos through claude code and it came up with this outline and at the
1:25top of the artifact you'll see this chat button. If I click on this I can come back to the chat and say I want you to create this kind of overview and it should be very visual and just kind of like put them into like a hierarchy and just have the steps just to kind of give like a visual overview. This is uh that's less in-depth.
1:47And so Claude can manipulate and edit every single one of the artifact types. And this model is getting much more visual and it's really good at creating things on this new whiteboard artifact. And you'll notice here I'm using Fable 5.2 Extra. This is not necessary at all for this task. So I'm just going to stop it and I'm going to switch to Opus 5.5.
2:10We're going to not use extra. We're just going to use medium. And we'll just say continue. And you can actually see where it's about to begin working. And you can see here it shows puttering. And I'll zoom out a little bit. And there you go. We now have this kind of overview slide.
2:28And then you can immediately edit it directly. And I can add this little section right here. And I can name it overview. And then I can drag this around. Like I can grab this whole thing. And we can move it exactly where we want it. Maybe we want it above this right here. And look at this. Way easier. Now it's so easy to use AI to create these whiteboards. But there's actually a new artifact type as well.
2:53And this one I've actually been using more than the whiteboards one. Check this one out. I want you to create a new sheets artifact with all of these. If I were to make a video on this topic and release a sheet that gave people all of the resources that they needed to start creating videos, that's what I want this to be. [snorts] And you can see when it starts to work because it opens up this new sheets uh side panel and it looks
3:21exactly like Microsoft Excel or Google Sheets. You can very easily full screen it. You have all of these different options down here at the bottom. And you can see Claude kind of working. And look at this. So you can see it started to add these first. And look at this. It's going down. It just added all of these different tools. Oh wow. It's like tagging everything. Just like look at how good this looks. Add some charts on
3:49a new sheet that shows the breakdown of the tools by category. And here you can see this breakdown is working because there's this little like gradient on it. And if I click on it, look at this toolkit breakdown. Here we go. It can create these charts which are fantastic. And I cannot tell you how useful this is to have built directly into Claude Code. And by the way, you can do the same thing in Cloud Duet.
4:20They all have the same artifacts and they're all synced. So, it's super easy to switch from Claude Code to Claude Duet, which is their new agent platform.
4:28So, they created their own version of Excel called Claude Sheets. They created their own whiteboard app, maybe kind of like Miro or Whimsical Diagrams or Excaladraw, and it's called Claude Whiteboards. The one that I didn't show you is Claude data. And this new artifact type can be thought of as a Claude live dashboard. It takes the integrations you have and you're able to create a dashboard. Whether that's users, whether that is how many views you have on YouTube. The difference
4:58between this style of artifact and a dashboard that you would create as a normal artifact, this data stays live and it actually shows you where all that data is coming from. And then yesterday they made a new announcement called Claude Meetings. And they announced this yesterday. They said, "Introducing claude meetings. add Claude to your team's video call and it builds docs, whiteboards, and other artifacts while you're still talking. And I'm just going to watch this video. And just know this
5:26will take some time to roll out. It's only in some team and enterprise plans.
5:42Hey Claude, can you uh start a whiteboard, please?
5:52[music] [music] Heat. Heat.
6:08[music] [music] [music] And so this is by no means a meeting notetaker. This is adding Claude code as an active participant into the meetings.
6:35Hey, can you spin up a whiteboard? And you can say like, "Hey Claude, like as we discuss our ideas, can you keep this organized live in the whiteboard?" So it's almost like adding an active employee into your meetings. So in summary, we have sheets and whiteboards, we have clawed data, and we also have claude meetings. Okay. So now I want to talk about the new models released this week from anthropic and Google. So the first two models I want to discuss are
7:04Fable and Mythos 5.2. Right now, if you use the standard Claude app or if you use Claude code, if you come down to the model selector, you will see the brand new Fable 5.2. However, you won't see Mythos 5.2. As many of you know, a few weeks ago, many companies were posting about how they were under attack. There was a lot of cyber attacks going on in tech. And if you went to Opus 5.5 and
7:33said something like, "Hey, I'm getting cyber attacked. I need you to analyze everything and figure out the problem and solve it. If you went to Opus 5.5 and ran a prompt like this, you would get a response with something like, "Sorry, we can't help with that." And that's because the restrictions on Opus 5.5 were extremely tight around cyber security issues. Now, Claude has released two classes of models, right?
7:59One that anyone can use, right? you can use it right now. And then another one with less restrictions that you can use for different cyber attacks. You can get it to analyze vulnerabilities in your codebase with less restrictions than Fable 5.2. And that model is called Claude Mythos 5.2. And so everyone on the Claude API and in the cloud app gets access to 5.2. With Mythos, only organizations who are vetted through anthropic cyber and life sciences
8:28verification programs. It's much harder to get access to Mythos 5.2. I think the question that most people have is how does Fable 5.2 compare to Opus 5.5, which was already the best publicly accessible model in the world. And I will tell you this, Fable 5.2 is just like Opus 5.5, but it's more refined. It has more attention to detail and it just especially with 3D workflows and video editing workflows. It just does a better
8:57job at understanding what really good looks like and it will spend more time and effort refining the small details and I've found that it just has better outputs across the board. However, it is twice as expensive as Opus 5.5. And so in your Claude app, you have your weekly limits and you also have weekly fable limits. I have been hitting them with normal business tasks on the $200 per month plan. Opus felt unlimited where
9:26now it does not feel unlimited. And I do not think you should be using Fable for everything. And here's kind of a good overview on how I would be using the two models. For most tasks, use Opus 5.5.
9:38For really hard tasks that use a lot of tokens, do your planning with Fable 5.5 and then have Opus 5.5 execute. Fable's really good at coming up with really good plans. But if your task requires a lot of sub agents and maybe it's over an hour or even 2 hours, a lot of Blender work or even creating video games or creating apps, these can take up a lot of tokens. Opus can do a lot of the
10:07work. just have Fable make a very detailed plan and then send Opus 5.5 off for the hours of labor. It might be slightly less high quality, but it will be roughly the same and you will save so much money in tokens. And then for me, Fable is all reliable. If something goes wrong with the app that I'm currently building, or if I need to get an edit done immediately at a very high quality, like we have a 2-hour time crunch, I'll just go to Fable. I'll just say, "Hey, you have this amount of time. Get it
10:36done." And right and this is a lot of the time I don't care how much it costs. I just need something done for my business. I will rely on Fable. So it's like Fable 5.5 is all reliable but he is more expensive. So keep that in mind.
10:49And then beyond Fable and Mythos, we also have Gemini 4 Carbon, Google's latest model that they released a day after Claude Fable 5. So, I've spent around 2 and 1/2 hours testing their new model, Gemini 4 Carbon, and I will say it is their best model that they've ever created. It is their version of Fable. It is thought of as their best model.
11:13And in the 2 and 1/2 hours I tested, I couldn't find a single thing that it's better at than Opus. I don't think this model is better than Opus. I think Opus has way better taste, especially in front end. It's better at 3D. It was better at creating video games, which I think is actually a really good test of its general intelligence. And it's not nearly as good as Opus at editing videos. And on top of that, I don't like using anti-gravity. Anti-gravity is the app that I needed to use to access
11:42Gemini 4 Carbon. And both the Claude app and the Codeex app are way better. And so, I don't really know what app I should be using with Gemini. And I hope this changes in the future. I hope it's clearer what platform I should be using as a business owner who wants to use agents to get things done. I still don't know which app I should be using because I like using the claw desktop app and the codeex app so much. I see no reason to switch right now. One thing I do want to flag is the price. So, if you look at
12:10the price of Gemini 4 Carbon for a limited time, it's actually half the cost of Opus and then it will increase to the same price as Opus through November 30th. So, that gives us 17 days of cheaper Gemini 4 carbon, but even at half the price, I don't see a reason to switch from the Claw desktop app to anti-gravity. And since many of you who follow me use Grobbot, I do want to discuss some small updates they made this week in like 20 seconds. So the
12:39first update they announced this week is you can actually use the Mac apps that are on your computer. So on Mac, Rockbot can now use messages, mail, notes, reminders, calendar, location, and contact. They updated that on November 12th. And overall, they're just making a lot of updates to make it easier to add to your team. And the Slack integration where you basically add Grockbot to your Slack is getting a lot of new updates and I'm actually going to be making a
13:07video on this very soon because it has been very useful for me and my team. And then the next one is they added tail scale. So Grockbot can join a tail scale network and this will allow you to reach private MCP servers and internal tools. For those of you who run a startup, this is incredibly useful. Okay, so now I want to move to some updates from dots.
13:28So, Dots just released a new feature where you can connect your Dot to many different computers. So, if we were to navigate to our Dot right now, you can see here that we actually have our Dot connected to two computers. One is the MacBook Pro that I'm filming on right now. This is my most powerful computer.
13:48I also have a Mac Mini right here that it's also connected to. And I can say something like this. Please, can you list out the latest videos in my downloads folder? Just like list out the files. Do this in two codec sessions. One on my MacBook, one on my Mac Mini.
14:07Point that I'm trying to get across here is with your dot, not only does it have its own little computer that it can use if it doesn't need your computers, but you can also fully control your computers. And because codec computer use is so good, you can just ask your dot to fire up codec sessions and ask it to do whatever you need it to do and it will just use computer use to get it done. And look at this. You can see here latest download videos. So if we were to click on recents, we can see found six
14:35video files directly in downloads. Here are the latest downloads on my Mac Mini.
14:42And here are the latest downloads on my MacBook. And so I can just talk to one agent, my DOT, and we can control an entire fleet of computers. And some people that I've been talking to who are using AI for video editing are buying a new powerful machine just for video editing. And so they're basically like, this computer is only meant for video editing and it's editing videos constantly all the time. It's making clips, rendering it on the computer. And
15:10now you can control that through DOT.
15:12And so you just say, "Hey, the Mac Mini that we use for editing, can you do this task?" And so it's kind of like a computer coordinator in a way. And the latest update to do comes from about 30 minutes ago, and that is the brand new chat GPT wallet. OpenAI just released this. Now your DOT can securely complete purchases on your behalf with your approval. With ChatGpt Wallet, agents pay without access to your actual card details. Every purchase requires a
15:41security check. Chat GBT wallet is rolling out starting today for DOTS pro subscribers in the US. Sorry people in Europe, you're cooked. And this just came out, so I haven't tested this. And since, you know, I'm a business, I actually want to take some time to learn about it before I just set it up right away. But let's take a look at this video.
16:04So, you can say, I love to find a gift for my friend Amelia ahead of the holidays. She likes to cook, make pottery, and loves plants. Budget is about $100. I've been dying for a feature like this. I'll look for a holiday. I'd get this. All right, let's see what it does here. So, it's asking where to ship it. Yes.
16:25And then here, you can see it said you can review the purchase and then it lets you approve. So, it doesn't buy things immediately on your behalf. You get a little notification and then it says purchase complete. and then the order will come. And I think maybe consumers aren't super excited about this, but as a business owner that is ordering things every single like at least every other day, I cannot tell you how excited I am to be able to talk to an agent that I
16:52use that knows my context that's plugged into the tools I already use that can just buy things on my behalf and ship it to my house or business. I genuinely want this so bad. But the problem is is it doesn't always work. And we're actually going to be talking about that a little bit later. And the final update to dots is that they are just getting faster. I was very vocal, I think around a month ago, about how slow DOTs was getting. Now they're significantly faster. They use less jargon, especially on the real-time voice. As you guys
17:22know, the way that I talk to AI, I am generally talking to AI when I'm on my phone because I can spin up cloud code sessions. I'm more creative when I'm walking around and I use real-time voice and it uses less jargon. It has new custom instructions. Then finally, there is GPT 6.1 Terra which is now powering DOTs. To me, this is a little bit of a downside. I would much rather be able to choose which model powers do DOTs. But on the bright side, I can ask my DOT to
17:51spin up a new codec session that uses any model I want. So, that's good. I just wish there was more customizability. But this does mean that it is faster. And so speaking of agents not being able to do payments, I want to move to the final two updates here. Um, one is that United is blocking AI agents. And this is actually a trend that I'm seeing across the board. Uh, Amazon did something similar where they're not allowing agents to purchase things and use their website because
18:20it's bad for their business model. And so, United will restrict AI agents that book flights or manage reservations. And the reason they're doing this, at least I highly suspect the reason they're doing this is depending on who you are, when you go to book a flight, I if you if they think that you're willing to pay more, they will show you higher prices.
18:38And this is a very questionable tactic that they do. So, the reason why these companies might not want you to use agents is because agents are willing to search for hours to look for the best deal. And that actually doesn't benefit them. They want you to go to their site and get ripped off on one of these tickets. And agents prevent that from happening. If people start using agents to buy plane tickets, agents will always search for the best deal. They'll search through all the different competitors because they they'll take their time.
19:08You know, if you ask your agent to do it, um it's not your effort you have to put in where some people are lazy and they'll just buy the more expensive one.
19:16And this hurts United. It's the same with Amazon. Amazon can make a lot of money by getting agents to buy things on Amazon. However, the way they make a lot of their money on Amazon is the ads. And so if people are no longer going to the site and it's just agents, agents will just filter out the ads. The agents will just look at the best option. And so you can't really advertise to an agent. And so that hurts their business model. So, we're in this battle right now between
19:45how convenient it is for us as users to be able to use agents to buy things and the business model of the companies our agents are purchasing from. And I'm going to be covering this in great detail because this is a battle that I think is going to last many years and will be the biggest blocker to actually being able to use agents effectively to run an entire business. And then finally, I'm going to be finishing off with a little recommended reading. And I'm going to start doing this every
20:15single week to give you guys something that you can dig into that kind of dives into the infrastructure layer of the AI space that I found to be really interesting. And so this week there was a really interesting Twitter post from Will Depw and he says he has forensic evidence that DeepSeek Moonshot and Miniax and Ant Group. So these are Chinese companies built their newest models on Nvidia Blackwell chips. And so these chips are actually banned to be used in China. They you cannot ship them
20:44to China. And this post has around 6 million views. and he basically says that he has found conclusive forensic evidence that the top four Chinese AI labs have built their newest models on Nvidia's most advanced Blackwell chips which are barred from export to China.
21:01And so the recommended reading is actually this right here which I'll put in the description um is this link to why he thinks that. And it it was incredibly interesting and I highly recommend reading it if this interests you. Anyway guys, there you have it. There's your weekly agent update and I hope you found this valuable. I do this every week along with other tutorials and I cover other breaking news.
21:25Basically, we want to use AI agents to be more effective in our personal life and at work. That is my goal here. I try to make it easy to understand all these different concepts, stay up to date with AI. If you guys enjoy this video, please hit that subscribe button. It helps me out a lot. and I'll see you here in the next