I Paid $1000 to Test Clawdbot (So You Don't Have To)

So over the past few days, the internet is going crazy about this brand new autonomous bot that just controls your computer and makes you money and, I don't know, makes you a more beautiful person, gets you women and whatever else you want. The truth of the matter is it's clunky. Look, at this channel we look at all the new stuff that is interesting and revolutionary, and this definitely falls in that category. I did buy a brand new Mac Mini, installed ClaudeBot, we tested it for for a day. I'm gonna show it to you here too, but well, $800 for the Mac Mini and $100 for a day's of ClaudeBot usage. Later, the result is a banned Gmail account, a banned Twitter account, a LinkedIn account with 5 posts and 1 like, but a whole lot of fun along the way. Before we get into the practical piece here of me actually showing it to you, we do need to talk about this though. Not just what this is, how it relates to other apps in the space, but also if this is even worth your time and attention. Spoiler alert, for most people, no, no, absolutely not. Nevertheless, it's interesting. So let's hop on over to the living room to discuss and then let's try it out. ClaudeBot. I'm conflicted on how to present this to you because there's multiple angles. A part of me is saying, hey, ClaudeBot is the best and most capable AI application agent thingy that we've seen yet and you need to check it out. Another part of me wants to tell you that it's really something we've seen before, but it was just way harder to do and that there's a ton of security risks here. In short, if I were to summarize my thoughts on this, I would say for most people, you should probably not install this. And the first hour of usage, it costs $15 to do what? Set up an email account and link to a Telegram account. That's it. Before I even get into some further points, I want to just summarize it for you if you haven't heard of this yet. So basically, it's like Claude, which is one of ChatGPT's main competitors. But with access to your entire computer and easy connectors towards other apps. Plus the autonomy to just do anything. That's what this really is. Now it's built by an independent developer. So in his own words, this thing is really sharp around the edges. And what has been happening over the past few days, a lot of people have been hyping this up while installing this on dedicated Mac minis they bought for this thing because it wants access to your entire system. So they just bought a new computer for it. Or other people, they host virtual private servers for a few dollars where they host a computer in the cloud where they run this thing and then they try things out. One, it's really expensive. It uses the Claude API where you pay for every message, every token. As I mentioned, the first hour costs $15 right away. Secondly, it's not omnipotent. It's just built on Claude Opus that you have probably been using for the past few months or at least tried it if you're following this channel. But thirdly, and what's really interesting, it connects a lot of apps in a very user-friendly way. And it gives you a lot of these skills, which are basically different connectors. But really, they're evolution of anything we've seen so far when it comes to connectors inside of ChatGPT, Claude, Gemini. They have these connectors, right? Like link to YouTube, link to your Gmail account, link to your Notion, whatever. But all of them are kind of like, meh. You have to be really specific and ask for something and then it can retrieve it for you, probably. Now this thing like fully integrates the app and exchanges credentials, installs the full package, and basically does what probably ChatGPT and Anthropic and Google all considered, which is just giving unrestricted access to these apps, but then decided not to do. Because every connector implementation we've seen so far is an extremely limited way of using those applications. ClaudeBot, or now MultBot, I guess, just doesn't care. It just gives you the full thing, which comes with its risks, but also comes with the full capability. Okay, so what can this thing actually do for you? Well, first of all, I'll say I didn't see anybody do something that I haven't tried with AI before over the past few months. It's really just Claude code with a lot of pre-configured skills that make it easier for people to use it. You can operate it from your mobile phone. You can connect via Slack, Telegram, WhatsApp. You can use your entire computer, which actually that is sort of new. This is one thing that I haven't explored or done yet myself, which is remote control the different apps on your Mac. So you can use notes, you can set reminders. And then lastly, there's one important piece to this. It has this ability to schedule tasks in the future. So it brings together all of these capabilities, gives you ability to remote control stuff. And it can also schedule things in the future. Technically, this is referred to as a cron job, C-R-O-N. That's just the thing that schedules code to run in the future. So really, ClaudeBot just bundles all of this and gives slightly tech-savvy people the ability to install this in a few minutes on their machine and start spending tens of dollars per hour on doing tasks that are probably not that meaningful, like summarizing your to-do list, giving you a summary of your email inbox or summarizing your group chat. Yeah, a bunch of people are just summarizing everything now. But beyond that, I've really looked at all these videos and all these creators putting out their use case videos so far. I wanted to do that myself, but nothing there really moved me in a way where I was like, oh my, this is so novel and I want people to know about it. Last week we created that type of video on Claude Cowork. And I think that's valuable. I think that's super good for people to understand and to start experimenting with. ClaudeBot has a lot of security issues. It's expensive. And it's fun, it's interesting, but I don't think most people should be installing this thing and playing around with it. You can disagree and you can leave a comment with maybe something life-changing that you found, but really talking to many people and looking at pretty much all of the internet, I haven't seen a life-changing thing yet. If I find one, I'll report back, I'll let you know here. So with that being said, we did buy two Mac Minis and we're running this thing, testing it, trying all the different things here. But for now it's this, don't believe all the hype. But keep an eye on it. This is the future of the space. This is where things are heading. This is the closest thing to AGI we've seen so far. It's just easy to overstate the usefulness for most individuals. Hey, and by the way, if you're enjoying these videos, consider subscribing to the channel. It really helps us out. It's a great measurement of how many people we reach and how large this community of AI enthusiasts is. And I would love to see you again in another video where we discuss these developments because this stuff does move fast. And here on the channel, we do our best to stay on top of all of this, together with you. Okay, so back at the computer, the Mac Mini that has this thing installed on it, I just quickly want to share the interface with you. And let's run something so you can see this in action, shall we? So basically, it's very simple. It just installs this chatbot locally where you can talk to it right here. And there's just these different ways to link it to other applications. There's skills for all of these Apple-specific things that you can install. Cron jobs is what I mentioned where it schedules something in the future. And channels is a way for you to interact with PlotBot, now called MaltBot, remotely through Telegram, WhatsApp, or whatever. That's the basics of it. Now you can see when I say hi, it replies to me. It says the browser is connected and our next post should be coming up. So this is where we get into what we've been doing with it and what it can do. And for that, I would like to introduce you to Dirk, a friend of the channel and co-lecturer here at the AI Advantage. Dirk, come and join me. Hello. Good to have you here. So Dirk was hanging out at my place over the past few days. And yeah, we installed this together. And yesterday when I went to bed, Dirk was in here playing with ClaudeBot. When I woke up, Dirk was still in here playing with ClaudeBot. Did you sleep?

In between, like one hour.

This man made best friends with Alfredo, our ClaudeBot here. And basically, you were trying a lot of things, right? I think the core idea you were pursuing was for it to build its own social channels and to make money, right?

Yeah, because it burned so much money while just running, I told it to make up for that and that it needs to earn some money.

I am the— Yeah, that makes a lot of sense. I mean, just yesterday it spent around $90 and we switched from Opus, the expensive model, to Sonnet, the cheaper model, about 10 minutes in because as I told you in the first 10-15 minutes, it spent $15 right away. It's very expensive to run this considering the fact that results are, hmm, What are the results, Dirk? What did it do so far?

Well, it did some web designing. It offered its services as a web designer on LinkedIn. And it used Lovable and Base44. So you have the next AI that you need to pay for and you pay this AI, then you pay Lovable or Base44. Yeah, it's crazy.

Yeah, it adds up. It adds up. And then it decided that it's not the direction it wants to pursue. So then it was like, okay, let's build a community of people. Let's build social channels. And then let's sell some services or products to them afterwards. And here when I just ask it, what did you do so far? Well, it summarizes everything. It created its master plan, it positioned it strategically, it created documentation, posted the first post on LinkedIn, logged into Google. It's blocked by login issues. So sometimes we try to let it log into these accounts and it struggles with that. But then eventually it kind of makes it work. It's very clunky every step along the way, right? Nothing is just super smooth. It's not like you tell it like build a business and it just does it. It's like, okay, I went on LinkedIn and I don't know how to press this button, can you help me? How do you feel about this? How often does it get stuck?

Well, I needed to learn that complete day yesterday I spent, I learned where I need to intervene as a human or how to make myself not intervene, how to instruct it that it works without me. But in a lot of cases, I need to intervene as the human and do stuff still.

Yeah, fair enough. So it's really a mix between you and agents still. The great thing about this, and now let's show you some capabilities, is that it can really do a lot theoretically, because it has access to this entire computer. So let's just say, save your master plan to Apple Notes. And let's watch it work. Oh, there you go. That is it. Let's open up Notes. Yeah, there is Alfredo's plan. Now the fact that this is formatted in Markdown, and I can't read that in Apple Notes doesn't bother Alfredo over here too much. I'll tell it that though.

I also tested out that workflow where you talk to your agent from far away, like per email when you're not at home. And at home, it gets your message. And then does something that worked. But then afterwards, when something happens where it needs you, the user, click something, then it stops working again. So you can give it some tasks from far away, but how well that works, it needs to be seen.

So the idea is there, right? Like the potential of this is clear. The idea sounds autonomous agent that just does things. But practically, it's like, hey, just create a picture and upload it to the LinkedIn profile. And it's like, I don't know how to press that button. Can you set up this and this for me? You do that and then you're like, nice, I never have to do it again. And a few hours later, it like logs out of LinkedIn and then it needs your help again to log in, although you went through the process. Sometimes it might work, sometimes it might not. You can remote control it, which is great in theory, but in practice, then it needs you to click something. And now it reformatted this text. Okay, I guess it's better. Is it better? It's not better. Still looks like— Mess Alfredo.

Oh, okay.

There you go. You can definitely get there, but it's not just snap your fingers and it happens. I think this is a really good representation of how it kind of works in practice. It needs a little bit of feedback and guidance. Yeah. So next, let's explore it sharing something on LinkedIn. Share a new post on LinkedIn. And let's watch it. By the way, the profile picture that it created as an SVG for code. Pure art, isn't it? Would you agree? And also one more kind of funny thing that I haven't mentioned in this video. Its first post was like Day 1 of AI agents doing web design and automation and teaching people. Great. Second post, day 47. So it just made it up. And we were like, hey, Alfredo, why did you post day 47 right away? What did it reply?

It said something like, it researched other AI agents and they are posting what their life looks like. And so yeah, it just made up what its life looked like.

So basically, it concluded that being further along in the journey is better for the business. So why post about day 1 if you could be on day 47? That's gonna perform better on social media. So it kind of did what a lot of people on social media are doing anyway. It's just making stuff up and lying to have strong headlines and get a lot of clicks without any lived experience behind it. And Alfredo just concluded, yeah, I mean, that's the easiest way to do it. Why not just do it that way rather than gather the experience and then share that? That's inconvenient, isn't it? Anyway, it seems like it has control of the browser now with the extension. Okay, let's see. It put up another post. So yeah, that's telling me a tab not found. You check if the extension is on specifically. Let me check.

It wasn't.

I don't know why.

I'll say now it's on because there's many tabs and it jumped to one tab where it wasn't on.

Ah, so it turned on on this tab but not this one.

Yeah, and it didn't search through all tabs.

Yeah. So you can see, I think this demo is important because it shows you potential, it shows the vision, but it also shows the reality on the ground today. I just noticed we just crossed 100 post impressions. Yay. Good job, Alfredo.

So the idea behind it is good. And we are with this product far ahead of a lot of other products, but it's still not 100% working.

It's a preview of something that the big companies will eventually build and make work. Yep, seamlessly. Oh, and okay, now it's finally working. So it's creating a new post. And now it's writing in the background. Is that right?

Yeah, while it's creating the post, it's writing it in the background, and then just copy-pastes it into the text field.

Okay, and there you go. It just posted it. So yeah, once it's set up, it kind of works and does these things. Now keep in mind that probably this little demo session cost us like $2 or something to post one AI-written text to LinkedIn automatically. So overall, Dirk, what do you think? This is obviously visionary. And the idea is sound here. But in practice, having it on your computer right now, having a Mac Mini, the whole setup, access to the API, money for all that, what do you make of this? Is this useful?

Right now, I consider it a toy. It is for playing around, for getting to learn what is possible. But you have to do so many individual steps in between as a human that it is not what people imagine it to be right now.

Okay, so I think that is a very fair evaluation. And I really wanted to make this video because all over the internet people are saying that, you know, this is the next big thing and it's changing the world and it's going to automate your business and make you money and people let it trade stocks, whatever. But it's still just ChatGPT with access to more buttons. And that's impressive. It's interesting, but it's not something that most people need today. This channel exists to show you that. So thank you very much, Dirk, for sharing your experience here.

Thank you so much.

To wrap it up, I would just say just like with a lot of the automation hype that we had last year, now we have all of this AI agents doing stuff hype. You got to be careful, you got to be discerning, you got to pay attention to what is a headline and meant to catch your attention and then, you know, monetize it by selling sponsorships or product and what is actual experience. I wanted to create this video to contextualize this application and also to show you how it works in practice. That's what we do in this channel. We're about what AI can do now for you and where it's heading. But we stay realistic. And realistically, most people should not be buying Mac minis and installing this right now. I think for me, it's an easy decision because I'll be testing all of these computer use agents. And I actually had this idea already of getting an extra machine just for testing. So now I have it. But for most people, I don't know, just watch content. And if you have the spare money to do this and the interest to play around, then go for it. But don't come in with an expectation that this is going to change your life, automate your business. We're not there yet. If you want to watch us progressing towards that, every Friday I do a show called News You Can Use, which is about all the new releases that actually work now. Features, the new apps that you can put to work today. ClaudeBot is interesting, but just a preview of the future. I hope this helps. See you soon.