Claude just became OpenClaw

So, Anthropic just shipped features that make it almost identical to OpenClaw. All the stuff that people were losing their minds over with Moldbots, Clawbots, OpenClaw does all the same thing. By the way, Anthropic just replicated again pretty much all of it in Claude. And not only that, but it's more beginner friendly. It's hosted by Anthropic. There isn't a security nightmare surrounding it to install it and use it. Which begs the question, did Anthropic just kill OpenClaw? Is OpenClaw cooked? Ethan Mollik recently posted that this new thing by Enthropic, it's 90% of everything that he wanted to use OpenClaw for, but it's far less likely to take all his personal information and just post it on the internet somewhere, which is nice. And I think that's probably the best description of what Anthropic cooked up here. First of all, yes, it is 90% of what most of us wanted with OpenClaw. There's a lot of people that are going to switch to Anthropics version of it and never look back. With that said, OpenCaw is not going anywhere. Not by a long shot. All right, let's break down what happened. All right, so first of all, for those that missed the entire OpenClaw saga, what is it? It's an opensource AI agent that runs on your local hardware 24/7. Now, of course, you can put in the cloud, you can put it on somebody else's machine. And yes, if you're using models from OpenAI or Enthropic or wherever, technically, those models aren't on your local machine. You're still making API calls to those AI providers, but a lot more of the data, the files, the memories, all the stuff surrounding it, it does stay on your computer. You control it. This is actually kind of a big deal, which may be somewhat easy to miss if you haven't been working with it a lot. But after running on something like open claw for a while as it builds context about your life, who you are, how you like things done, what your goals are, like all these random little things about you and your business, your work, whatever, then switching to a regular chatbot is tough. You feel like you've lost a good friend that knows you and you're back to this a little bit more kind of faceless and not very personal assistant. The other kind of huge breakthrough with open cloud. I don't know if to call it a breakthrough. It was kind of in retrospect obvious, but we didn't have anything like that at the time. You were able to control it very easily through Telegram, WhatsApp, whatever other messaging service you wanted. So, wherever you were in the world, you just texted your little AI agent and it would get stuff done for you. It managed your files, emails, calendars, browsers. It did research. It wrote code. It built websites. It took various autonomous actions while you slept. It was created by Peter Steinberger. It was the fastest growing open source project of all time. Went absolutely viral actually not that long ago. Early 2026. It's still early 2026. And it caused literal Mac Mini shortages. Like if you went to the store, to the Apple store, whatever, they wouldn't have Mac minis stock there. Nobody knew why. It was literally because people were buying them to run open claw. Apparently, it was also being subsidized for AI development in China. Like this thing is huge. It's kind of hard to overstate how big this thing was. So Fortun Forbes, they all covered it as this cultural phenomenon. Me personally, I was like 24 to 48 hours behind that little viral explosion. So in this game, that means being late to it. But once I got my hands on it, once I set everything up, I got it. This was a game changer. I use it to this day almost every single day. Now during this time, there's a lot of things that Anthropic did that some people disagreed with. some people didn't like. They did send Peter Steinberger a cease and desist. I think in one of the interviews he said that they did reach out. They made sure that everything was nice and friendly. This wasn't like a super aggressive thing, but it did force him to switch to Maltbot. Apparently lobsters when they shed their shells, that's molting. So, it's moltbot. And then various crypto hackers somehow managed to turn that into who knows. There was a crypto scam. So, yet another rebranding was needed to openclaw. Peter Steinberger did go to work for OpenAI. Sometimes people refer to it as OpenCloud being acquired by OpenAI, which is not 100% true. That just seems to be like what everybody's referring to it as, but it's actually just Peter that went over there. Open Claw is still its own thing. Nvidia launched Nemo Claw. There's a billion other various claws and other agents. Like this thing is massive. Andre Karpathy is now referring to everybody having their own claw. So it's almost like a brand new sort of category of AI or software that Peter Steinberger created and now it's called the claw. So, what does that have to do with Anthropic? Well, Anthropic recently launched co-work plus dispatch. So, that's Claude Co-work, which is not super recent, but the newest addition is Dispatch. And actually, surprisingly, there's been a lot of stuff that they shipped. So, this is from the product Compass by Pavle Hurin. He had a pretty cool infographic that I want to share. So, he's saying, "Annthropic ship 74 cloud releases in 52 days. How many of those were you conscious of? " because I'm going to be honest. I don't know exactly how many I caught. There's tons. Cloud and PowerPoint, Sonic 4. 6, Sonic

4. 6 in CC, Figma, MCP, Cowwork for Windows. We had cloud tasks, projects and co-work speed boost. Here's Dispatch Plus Code and Shell in skill. mmd files, the Opus 1 million desktop. I mean, there's tons. This is Boris Chney. So, he is the creator of Claude Code. He is the fountain head, so to speak. So he's saying a little known fact the anthropic labs team the team he joined Anthropic to be on they shipped MCP. So that whole thing that's him and his team they shipped skills. Again each one of these is like these kind of big deal revolutionary things that are added to this AI and agent sort of ecosystem right and then also claude code. So it's just like those three things are just massive. So, he's saying it was just a few of us shipping fast trying to keep pace with what the model was capable of. And now he's saying fast forward to today, which is yesterday and whatever day you're watching in the future, it's March 23rd that this is the day that we're talking about. He's saying he's excited to release the full computer use in co-work and dispatch. Really excited to see what you do with it. Those early computer use days. So computer use is for those that are unfamiliar. That's when we try to get these AI models to navigate the computer, the browser, the desktop just like you and I would with a not a keyboard and mouse necessarily, but by you know moving the mouse around and clicking on things by typing things in and not by just directly kind of sending the commands through the backend interface or whatever. That's an important thing to understand because right there's for example Chrome plugins where the AI can quickly navigate. They can like find the elements in the website. they can interact with them and that's cool and all but that's different from like a human being able to navigate everything and click on it. They have been for a long time now kind of notoriously bad at that like comically bad. In response to Boris's post, Charlie or Charlie posted this. She's saying, "I am never letting this go. Charlie, I stand with you. We will never let this go. We will hold it in our memories forever. " This was in one of the older computer use presentations by Anthropic. They're saying while we recording demonstrations of computer use for today's launch. So they're trying to roll this thing out to, you know, the world. They encountered some amusing errors. I, if this was happening to me, I would describe them as amusing. In one, Claude accidentally clicked to stop a long running screen recording, causing all footage to be lost. If that happened to you, you would not be amused. Just let me tell you. In another, Claude suddenly took a break from our coding demo and began to peruse photos of Yellowstone National Park, which does sound like something that Claude would do. But back to the thing that they're talking about today, the stuff that they're released, well, yesterday, this the stuff that just got released. Number one, you can enable Claude to use your computer to complete tasks. So, it opens apps, navigates your browser, fills in spreadsheets, anything you do sitting at your desk. Do you now view any mdash and any text? Do you now like treat it with suspicion like uh what is that doing there? Did you Did a human write this? So, this is research preview in cloud cowwork and cloud code. It's Mac OS only for the time being. That means I get to dust off my MacBook and see if I can run it there. I've been messing around with it for a little bit. And in a minute, I'll walk you through kind of what that looks like. Right now, it's very limited to who has access to it. So, I think you have to be on the Pro or Max plan. Again, it's Mac OS only. It's desktop only. It's cloud corework and cloud code only. So, it's a very limited, but you know how fast they tend to ship stuff. So, this is probably going to be rolling out to more and more people very quickly. And if you're wondering how interested people are in this thing, well, 66 million or 66. 9 million views within I mean at this point a little bit over 24 hours. So, Cloud uses the connected apps first, Slack, Calendar, other integrations or opens up stuff that you don't have connected so they can use it like a human being would. This is the other thing. side of it. So, this is what they're referring to as dispatch. Basically, it's an app. It's the Claude app on your phone. So, it says, "Assign a task from your phone, turn your attention to something else, and come back to finished work on your computer. Tell Claude once to scan your email every morning or pull a report every Friday. " And it handles it from there. So, this is what Telegram is for OpenClaw. Now you have a specific app cloud app on your phone that you can use to communicate with it. But between you and me, I do think they are trying to enable whatever app you want to use to be able to communicate with uh cloud code, cloud co-work. Why do I believe that? Well, because discovered hidden within the app's code orbits appears to be the internal code name for anthropics upcoming mobile automation framework. So you can kind of see it here in the bottom left. We have artifacts. That's that ability to create certain code and certain visualizations within Claude. We have AI powered artifacts which create apps, prototypes, and interactive documents that use Claude inside the artifact. You have code execution, file creation, web search. These we've kind of seen, we know what they are, but that's what caused the green text to be distorted. I wonder if one of you

commented about this in my previous video. I have to check. Anyways, the point here as you can see this is basically working with different things like maps, composing messages, phone calls, phone use, various browser tasks like browser takeover, request user to take over, etc., etc. Calendar, all of that. By the way, I got to give credit to testing and use catalog. This person usually breaks these stories. One of the first people online on X to break these stories and actually giving credit to somebody else. I love when people give credit to whoever sourced the story. that is uh good thing to do. You get massive karma points for doing it. So my point is they're shipping a lot. They're very quickly adding a lot of functionality that they see out there in the wild and phone use and all that. I think that's coming next. And just like with everything else by the time they get released, you know, with Anthropic and Claude, it's going to be polished. fairly secure. It's going to be ready for your regular consumer use. So the Claude team is shipping fast. You know that expression when people are like, "Yeah, that's like my middle name. " like you know like shipping. Shipping is their middle name. That's literally this guy's last name. Shippar, right? That's thic shippar. I know I'm slightly misreading that, but shipper is his last name. I mean, that's got to count for something. I'm just playing around. Please don't lose it on me. But this, I believe, is Boris Chney's team. So that's the guy behind MCP skills and cloud code. That's the people that are building all the stuff that a lot of us are losing their minds over. And don't tell me it's just me. There's a lot of people that are extremely fascinated and impressed by this. All right, but what does this all mean for OpenClaw? Is OpenClaw cooked? I don't know. I'm I think I'm going to use that in the title and I'll split test it with other titles. So maybe that's even the title of this video that you clicked on because here's the thing now. Cloud code has a lot of the functionality, you know, computer use, using Excel, working through a mobile interfaces from your phone, sending commands, just like we have with OpenCloud, a persistent 247 agent running on your local hardware and messaging to Anthropic. Here we have this persistent desktop agent still doing a lot of the same stuff. You need a internet connectivity in both cases. If you're running open cloud using the cloud and anthropic model, you still have to connect to anthropics servers. In both cases, you still have that sort of quot that Anthropic gives you. By the way, that's a big deal because I'm sure most people are kind of aware of this, but if you're, let's say, paying Anthropic's max plan, you're paying 200 bucks a month, you get a certain amount of quota with that. And that, you know, certain amount of tokens that you can use for whatever work you're doing. And those tokens are kind of heavily subsidized by Enthropic. What I mean by that is if you completely max out your quota in any given month on let's say the Anthropics Max plan, the amount of money that you pay that month is, you know, whatever $200 or whatever you're paying for your monthly plan. If you were to use that same amount of tokens and you paid the real actual API pricing, that would be a lot more. I mean, at least 10 times, maybe 20, 30 times more. It would be a lot more. or I don't know exactly how much but I think easily 10x or more than what you would have paid just paying for the monthly plan. So that means that these openclaw and other apps of this sort they're kind of a little bit under threat for from potentially anthropic kind of pulling the plugin all that stuff and anthropic already said that you're not allowed to use the authorization tokens for example cloud code to run it with other things. They've have banned a few people or at least a few people that I'm aware that were banned. I'm sure there's a lot more. And the other big advantage of doing this cloud code cloud corework approach is it's sandboxed. So cloud is on its own kind of sandboxed virtual machine. So it's not quite the full system access self-hosted thing that you're using with open claw. Like the amount of damage you could do with open claw is I mean like unlimited. I guess you can do a lot of damage if it's not set up correctly. If it gets hijacked, if there's some sort of a prompt injection, I mean you could do serious bad things on your computer. Whereas with cloud code or cloud code since it's a sandbox there's limited amount of stuff that it can do that's truly bad. I mean obviously it can run some commands like if you're using cloud code and you say just you know run all permissions. You don't have any to approve everything that it's doing. Could it get into some trouble? Sure. Put it delete all your code in the codebase and you know maybe you don't have it backed up. Sure. We've seen a number of situations even at companies like Amazon and Meta where stuff really goes south because they use AI agents and it blows up in people's faces because those agents do some damage. But for the average consumer running cloud co-work is much much safer than something like OpenCloud again for the average not highly technical person. Also Tropics version is very beginner friendly from a sense that there's no command line interface there's no self-hosting. There's no config files. You don't even really deal with like API keys. So, it's a lot safer, a lot easier, a lot better. A lot better meaning it's more beginner friendly. And you have various connectors for Slack, Google Workspace

for notion, things like that. Before it even touches your screen, it's already connected. It's built in. It's good to go. So, why is open claw still relevant? And this is I think important to understand, especially for people that are building these sort of startups, but for pretty much everybody in the space to also kind of see where things are going. So, first of all, open claw is free and open source. As you'll see, that's kind of a big deal. Free in the sense that in and of itself, it's free. If you need the models, you still have to pay for model use. Also, it works with any LLM. You're not locked in to anthropic. You can switch them in and out. For me, there's different models that I prefer for different tasks, and that changes monthtomonth. Like, for example, recently Grock 420, you know, fully came out of beta. So, we have the API keys for it. That thing is excellent at search where stuff is happening in real time. Like if you need like up tothe- minute latest news on something that's happening, nothing comes close to Gro 420. There's no other model that's like close. So depending on kind of what your use case, if you're building something with code or you're doing research or you're writing, different models will kind of best for that. Or you might just have your own kind of a subjective preference for a particular model. And Open Claw allows you to just switch them in and out. you run out of your quota here, just switch it out. Where just a high volume of things that have to get done, you can outsource that to a cheap local model. With Open Cloud, you also have the community behind it. All the skills that people are building, all the extensions. Now, you can argue, you know, still with cloud code and skill. m files, you still benefit from that, but it's still different. Something that's closed versus open source, something that a small team Anthropic is working on versus something that is basically global. And of course, it runs on Linux, Mac, and Windows, and various headless servers. You can run it on a Raspberry Pi, like those super cheap computers. Currently, the anthropic thing is Mac desktops only. Even if you have the mobile app, you have to have a mobile desktop thing to send the commands to. Now, that may change, but for the time being, that's as far as I know, the only thing that has it. And a lot of people are looking at this and comparing it to some of the other stuff that happened in the past with these big tech companies where they see a feature that they like in some startup and they just kind of snipe it. They rebuild the fe feature natively and the startup that sort of innovated and created that it dies. We've seen that with Facebook, of course. iOS and we've seen it with Apple and iOS. It sees some cool third party thing. It it pulls it into the iOS. Twitter and Tweet Deck that happened there. Salesforce sees a CRM integration, some thirdparty one that everybody uses and it just like rebuilds it and the startup that created it, well, it's it dies. Instagram and Snapchat stories, etc., etc. Now, here it is a little bit different in a sense that it wasn't just Anthropic that started building something that they saw out there. I think Open Claw was such a massive success that it's like everybody saw it and started building their own version of it. I'm pretty blown away by how many different claw features, how many different companies are adding it right now. We we're seeing just the tip of the iceberg. Believe me, you're going to hear a lot of announcements of every company going, "Here's our blah blah claw feature. " By the way, here is the desktop app. So, I'm kind of doing something weird here because I'm using a remote desktop to log into my MacBook. And basically, you're able to switch between chat, co-work, and code here at the top. In the co-work mode, you have progress for longer tasks, scheduled tasks. It's not cron jobs like you would have with OpenClop. And you have your working folder where everything gets saved to. You have various context files, etc. They give context for what you're about to do. To enable computer use, go to settings, desktop app, computer use. So once you open up settings in the bottom left, go here to general under desktop app. And interestingly, they have for example, they have a keep computer awake. So, this prevents your computer from going to sleep when you have Claude open. So, I was actually curious about how it would be able to run 24/7, but this will definitely help it. So, you don't even have to mess with the options in, you know, the iOS settings. This prevents it from idle sleeping while cloud is open. So, closing the lid still puts it to sleep. The display still can turn off, but you'll be awake and running stuff in the background. And you're able to enable computer use. It wants you to keep in mind that some actions cannot be undone, which is a good kind of a mantra to live your life by just in general. And they do give you control over which apps this will have access to. It even will later ask you if you want to hide some apps so that they can't even click on it. And of course, be aware that claude can your screen. It's taking screenshots, sending it over to the anthropic servers. And they also mention they don't say the words, but prompt injection. So, some sites and documents could contain malicious instructions that could misdirect Claude. So, we're going to turn it on. And notice here, we can also hide certain apps that disappear when Claude is using the system that later reappear when Claude stops using it. All right. So, let's see if it works. I'm going to say open up photo P in an incognito window and create a YouTube thumbnail that says

hello world and a drawn smiley face. All right. So, it's saying that it is going to use the computer use scale to do that. So looks like we could have set those accessibility those system permissions in settings or we can do it now. So you do have to enable that here. So when you click that it takes you here and you just have to enable that and now it's granted. Now we have to do screen recording request and same thing enable it here. All right. Then I had to restart Claude. So let's try this again. So as you can see here it's asking can claude use Google Chrome and we shall allow go for minions. All right. So it is progressing. So it looks like it created a little step-by-step thing. So, open the photop create this particular canvas size colorful background. It looks like you have to be logged in to Chrome that has the Claude Chrome extension. You have to lo be logged into it with the same account that you're logged into Claude code. So, that's one thing to keep in mind. And now it's asking if we're allowing it to use photop. com. We're going to say yes. We're just going to say all for this website. So, you can actually kind of see it working here in photop. com for people who are not familiar with photop. Some madlad decided to create Photoshop, the entirety of Photoshop at photop. com. So, thank you to whomever did that. I'm sure they're very well compensated because they have a lot of ads on there and I'm sure there's millions of people that use this. But it looks like Cloud Code or Cloud Co-work is actually doing a great job. So, it opened the correct sort of dimensions 1280 by 720 and so far it's doing a great job. So, let's see if we can create a cool looking thumbnail for us. you know, it's going to attempt to add a colorful background. I'm very curious to see what happens to the ad industry. It has to change somehow with these AI agents because I mean, right now, these ads aren't being seen by anybody except this AI agent. Well, maybe not these ads. These ads are getting seen by a lot of people since I'm putting this on this video, but you get what I'm saying. Actually, why am I showing you these ads? Let me cut them out. There's no ads. But so the cloud is thinking that he needs to find the gradient tool. And now it wrote some sort of a script that says hello photo or that's the default text. So it's going to try to actually write a script of some sort. That's interesting. And now it looks like it just took over the entire computer, it is completely hijacked what I was doing, interestingly enough. So instead of focusing on that particular tab, it actually took over my entire computer. So there's this golden outline around my entire screen. Let me show you. So you can see in the right it pulled in itself I guess the cloud coded desktop it pulled it into the same view as it's working in the tab making the thumbnail. I got to say it's working well now instead of actually clicking and building the thumbnail itself. It did use some sort of a script but that's fine. It is figuring out stuff on the fly. And it's interesting they found that kind of what shortcut or whatever you want to call it some easier way for it to do what it was trying to do. Anyways, I'll do my gauntlet of testing and try to figure out how good it is, how similar is it to open claw. The thing that I doubt that it's going to be able to compete with OpenClaw right now head-to-head in is this idea of memory and collecting various memory and stuff about you and just getting smarter and better and more tailored to you over time. I think it will struggle with that versus something like OpenCloud. It just basically saves everything. One of my open cloud agents helps me with health and fitness and all this stuff. It has a lot of my medical data. I for example take my blood work that I get every year. I upload it there. So now if I have a question, I can type in a question and it just spits back answers because I've uploaded those PDFs to it. And if you think about it, it's pretty simple. But at the same time, you know, here in the US, we have tons of laws about how that data can be stored or used or whatever. And of course, those laws apply. It makes sense why, right? So the doctors, the hospitals, they have to be careful about how they store all of their patients files. But of course, if I downloaded that file, so like my data, my personal stuff and I put it in some folder, those laws don't apply to me cuz that's my stuff. I can put it in whatever folder I want to. But you introduce AI to this whole situation. Like what if Claude has to now take those things and put it in a certain folder or more importantly read through those files and make certain decisions about them. All of a sudden, it gets a lot more hairy. are we now violating some I don't even know was it HIPPA laws whatever those laws are called surrounding how you're able to store health information all of a sudden things get a lot trickier so having an open source thing where each individual kind of takes ownership of it and just tells cloud what to do you know you're able to do things easier that way with these kind of gray areas anthropic can't have that liability so they're not going to want to have something like this be easily able to do that although I am curious to see if that's indeed the case if they have any sort of guardrails against it because again if millions of people begin doing that then that might attract some government attention some regulators that want to that are just like itching to pass some new laws

to make things more difficult. I understand that there's benefits to that, but for me personally, this has been a game changer. Just the ability to access all of my data, just throw it all in a folder and then have these like deep research and things that it's researching that's specific to my personal data and you know, how to improve things like for me personally, so to speak, not for the average population, but personalized and tailored to my conditions, my DNA, my blood work, etc. Let me know what you think about this. Does this replace OpenCloud for you? Or if you haven't used OpenCloud because you maybe you were intimidated because of the security features or it was just too difficult to install. This are you happy that now we have something like this that allows you to get most of the features and power and capability of OpenClaw like is this a good enough alternative? Because again this will be able to do a lot most of it. You know Ethan Moll said 90% of his use cases for it. So, I'm definitely very happy about this, very excited about this, but is it a replacement of OpenClaw for you? If you're wondering how well it's doing at creating a thumbnail, it's still struggling. Not quite ready to replace my thumbnail designer. That's me. I'm think I think you know that I'm my own thumbnail designer. But here's the thing. I think it did like five different versions at this point. I think Claude instead of perusing various what was it? Sequoia National Park photos or whatever got distracted by it. I think now is just it's all about making these happy face thumbnails. This is like version four or five. I'm not even kidding. And now it's beginning to export it. So, it's happy with the thing that is created. So, I'm very ecstatic about that. But it was able to save it as a I think it clicked on PNG. Oh, interestingly, just it waited for me to answer when I didn't answer. It went ahead and saved it as a PNG. Technically, we need a JPEG, but that's okay because if I wanted to do this in the future, I could create a skill for it. And you know what? With the skill, it's doing everything by itself. I was I'm not even sure what it was doing there, expanding itself to fill up my entire screen. In the context, it does have Claude in Chrome. So, it added that connector, so it knows it can use Chrome. In the working folder, our hello world thumbnail is added to it as well. So, obviously I need to do a lot more testing and obviously it's not perfect yet, but this is one of the hardest things that I've been asking the computer use agent to do throughout the years and I think this is the first time ever that it's been able to successfully complete the task. It's beautiful. I think I'm going to use it for this video. If you made it this far, thank you so much for watching. My name is Wes Roth. Please subscribe. Please hit thumbs up. and I will see you in the next