100 Hours Testing Clawdbot vs Claude Code (honest results)

So in about 5 hours and 80 million tokens later, I've built myself a full executive assistant. His name is Klaus. He has his own email account, his own ClickUp account, and I treat him like I would treat a regular employee. He remembers everything about me, about the business, about our Q1 initiatives, and he actually built himself this dashboard so that I can manage things that he's doing. I can see all of the deliverables, I can see the action log, and I can drop him some notes in here. And I can also see him up in the top left. So when he's working or when he's thinking or when he's idle, or if he's spinning up subagents, I can see everything that he's actually doing live. And so this is awesome. I've had so much fun playing around with ClaudeBot. It's been all over my feed, which is why I wanted to play around with it, obviously. And we've seen some really, really cool use cases. So you've got some basic things like email cleanup, sending reminders, creating issues, helping with ideas, helping create reports. I saw this one from Alex Finn, which I thought was pretty cool. He texted his bot, Henry, to make a reservation. And once the bot realized that it wasn't on OpenTable, or, you know, it couldn't actually book through OpenTable, it ended up just using ElevenLabs and called them and and then booked a reservation that way. So we're seeing these Claude bots be super proactive and great problem solvers. And of course, a lot of that is thanks to the underlying model, a lot of times, which is Opus 4.5. So let me just real quick introduce you guys to my Claude bot, Klaus. Hey Klaus, just wanted to say hi. Could you give my YouTube audience a quick 5 to 10 second intro of what you do for me? So I just shot that off. You can see on the top left now, we see that Klaus is working. And so I love being able to see, you know, just glance over and see if it's green, he's working. But if he's gray and it's idle, then I know, okay, well let me just fire off something else for Claude to do. And now Klaus is thinking because he's writing back a response.

I'm Klaus, Nate's AI executive assistant. I run 24/7 on a server, checking his emails, monitoring social media, doing research, and basically handling all the busy work so he can focus on making content for you guys.

I mean, that's the dream, right? But I realized that a lot of the stuff that I'm actually building with ClaudeBot, I could actually do inside Claude Code, and it's not even anything too new. It's one of those things that just all of a sudden takes the internet by storm, like out of nowhere. You can see the blue line here is Claude Code over the past 3 months, and the red line is ClaudeBot over the past 3 months. And all of a sudden, within the past week, ClaudeBot has just absolutely exploded. So what I wanted to do is put these two tools side by side, actually compare them across metrics that matter to you guys, and share what I've found. So today we're doing an analysis of ClaudeBot versus Claude Code. So why am I making this video? Because Claude Code is a coding sidekick, and it lives inside your developer tools, and it helps you read, write, and fix code faster on your local machine. Now ClaudeBot is more of a 24/7 AI assistant that runs on your own hardware or server. And this was released much more recently, pretty much like the beginning of 2026. And as of today, it is now known as MoltBot because Claude, you know, Anthropic basically said, hey, this sounds too much like Claude. So you got to change that. As you can see right here, a tweet from MoltBot. Big news, we've molted. ClaudeBot is now MoltBot. But anyways, what I'm going to be doing here is looking at 8 different metrics that you guys should actually care about, the metrics that I care about. And I'm gonna be giving them different scores out of 10. And by the end, we'll see which of these two tools is truly the winner. So the categories that we're gonna be looking at today are out-of-the-box ability, setup friction and risk, cost, power and access, security, everyday usability, actual ROI, and the ICP. So I think that this gives a pretty holistic understanding of like the differences between the two tools, especially when we isolate these different factors and look at just that element. But if I've missed anything, please feel free to let me know in the comments and let me know what you guys think about these two tools. Okay, so first of all, we have out-of-the-box ability. So for this one, I gave Cloud Code a 7 because it's excellent at helping you code right away, but out of the box, it takes a little bit of understanding to actually be able to make it useful. A lot of people that are non-technical and don't come from a formal software engineering background, it feels pretty inaccessible and it's, it's very intimidating. I remember when I was switching over from using N8N basically only to wanting to learn Cloud Code, I was intimidated as well. But now that I have a good understanding for it, I see all the power. For ClaudeBot, I decided to go with a 9 out of 10 because once you actually get it configured and you just start talking to it, it's so powerful. It asks you questions, it learns about your business, it has persistent memory, and it can create automations and apps right away. And you don't even really have to think about deploying or hosting or anything like that. It feels way more powerful than a coding assistant when you first set it up. It feels like someone that you could treat like an employee, which is why I gave mine, you know, its own login credentials for different things. And I can just forward it emails or CC it on emails and have it just do things for me and come back and let me know when it's done. So let me show you guys a quick example of two different outputs, one with Claude Code and one with Claude Bot. So I'm in a Claude Code project right here, and I'm just going to say, hey Claude, I need you to run a YouTube analysis of my channel and then send it to me. So I'm shooting this off. And then while this is running, let's go ahead and do the same thing to Claude Bot. Hey Klaus, go ahead and run another YouTube audit of my channel. Okay, so I shot that off. We can see right here that Klaus is working. We can see over here that Claude Code is working, although I need to give it some more information. and then I'm gonna come back when all this is done and I'll show you guys. And also one other thing to mention is both Claude Code right now and Claude Bot are using Opus 4.5, so they're using the same exact chat model. All right, first we have the deliverable from Claude Code. So this got sent to my email. We can see we've got AI automation, YouTube analytics, 30 channels tracked, 191 videos analyzed. It comes up with the top videos this week, but then here's the actual PDF report that it created for me. So I've got YouTube channel analytics, Nate Hurk, generated on January 27th. So here are my channel statistics at the moment. Then it goes over and looks at my most recent 100 videos with average views per video, average likes, comments, median views, and views per subscriber. We've got an analysis on video duration that comes with some charts. We've got engagement rate distribution, which is pretty cool. We've also got views versus engagement analysis. We have most used tags. And actually there was one more page, fastest growing videos. So it looks at some of the stuff that I've been doing with Claude Code, which has been a big trend obviously. Now let's see what Klaus got up to. In my dashboard, I can go to the YouTube audits, which opens up a Google Drive. and in this folder you can see that I have, you know, different audits that I've been using for testing. So this is the most recent one that it just created for us as a PDF, and this one looks a lot more branded because I gave both of them my brand guidelines and a logo. And so what I'd have to do is go back into Claude Code and have it work a little bit better to actually implement all the brand guidelines because this one looks a lot more like what I gave it. So same thing, 506,000 subscribers, 21.3 total million views, blah, blah, blah. It's giving us some other metrics here as well as strengths, weaknesses, opportunities, and threats. Something's a little bit off with the spacing here, but not a huge deal. We, I'm sure we could fix that. We've got some recent content analysis. So ClaudeBot, Agentic workflows, the trend that it's seeing is heavy focus on Claude Code plus Anet integration. And then we've got some strategic recommendations like diversity of tool coverage, launch evergreen series, build team capacity, YouTube Shorts strategy. And then it even gives us some priority action items as well. I don't think these automatically get put into our dashboard up here. That'd be pretty cool. And then maybe we could like cross them off if we didn't actually want to do them. So I'll let you guys form your own opinions based on that demo and based on, you know, your own use cases. But that's why right here I decided to give Cloud Code a 7 and CloudBot a 9. Number 2, setup friction and risk. So for this one, I gave Cloud Code an 8 out of 10. It's a lot lower friction to install and start actually using. You basically just install VS Code and then you add the Cloud Code extension. At least that's the way that I do it. It's super, super easy. And then with CloudBot, it's a lot more technical. You have to be a lot more in the terminal and you have to be a lot more aware of like what you're actually doing, what environments you're in, and you know, that's why people are buying the Mac Minis or on a VPS. And I'm actually using a VPS for this. I'm hosting my ClaudeBot on Hostinger, which you can get started for a lot cheaper. You can go ahead and sign up here for, you know, $6 a month, $9 a month, and I use the KVM2, and this is gonna be more than enough to just get started with a ClaudeBot and test it out. And when you're in here, if you're doing an annual plan, you can get 10% off in addition if you use code Nate Hurk. So like I said, when it comes to the setup, ClaudeBot is a little bit more technical and confusing. I just did a full setup guide tutorial, so I will tag that right up here if you're interested in watching that. And also, if you do decide to use Hostinger and you've set up your VPS right here, if you go to the Docker Manager catalog, they've actually added the ability to just basically one-click install or one-click deploy ClaudeBot. So should make it a lot easier. Anyways, we decided to give Claude Code an 8 and give ClaudeBot a 6 because it is more technical and also there's a lot more risk involved with actually setting this up. If you're doing that on your main hardware, which I would not recommend, if you're doing it on a Mac mini, or if you're doing it on some sort of VPS, there's still a lot of risk that goes into what people could actually access or where your keys are gonna go. Which brings us to a tie at 15 points. Now this next one is super interesting. It's about cost, which I know that everyone cares a lot about. And we have Claude Code at an 8 and ClaudeBot at a 6. And let me explain why. So starting off with Claude, you would basically pay for a plan to be able to actually use Claude Code. So you would have to be on at least the Pro version, which is about $20 a month, or you could go up to Max 5X or Max 20X for $100 or $200 a month. Now I know what you guys are thinking, that's insanely expensive if you think about other things that you might buy a monthly subscription for. But if you actually think about how much usage and output you can get out of a coding assistant like Claude Code in a month, and then you think about that compared to what you would normally pay an actual AI engineer or software engineer to do in a month, it's an absolute steal. So for $200 a month, if I can basically unlimitedly code using Opus 4.5, I'm gonna do so and I'm gonna be really happy about it. So that's why I'm giving Cloud Code an 8 out of 10. Now with CloudBot, which is actually now called MultBot as we just saw, this is different because it's an open source program, so you don't actually pay anything to use it or to install it. What you then are paying for instead is hardware or like where you actually host it, and then also the AI model that is going to power the whole system. Unless of course you're using a local AI model, but still in that case you're paying for the hardware. And so it's interesting because the majority of demos that I've been seeing, people are using Claude models in their ClaudeBot to actually power everything. But we have to be careful about, and why you guys saw me earlier show off that I had used about 80 million tokens today just with my testing and building, is that you're not actually supposed to use a Max plan. So you're not supposed to use one of your actual Claude subscriptions when you're using something like ClaudeBot, which is essentially a Telegram wrapper around Claude Code. That is actually against Anthropic's terms of service. At least that's my understanding. So that means you have to use an actual API key for Anthropic or OpenRouter or whatever models you want to use. And if you're not careful and you're just talking to your Claude bot on Telegram all the time, you can quickly, quickly run up a pretty big bill. So my bill today with that 8 million tokens was about $80. And in, you know, one session I could have done that in Claude Code and it would not have used up almost half of my monthly membership. So I think in the grand scheme of things, it is more expensive. So that's why we gave ClaudeBot a 6 out of 10. So now we're at Claude Code 23, Claude Bot 21. Moving on to number 4, we have power and access. And I went ahead and gave Claude Code a 7 and Claude Bot a 10. So there's a lot of similarities here, but there are some things that I just think that Claude Bot is actually more powerful at and just has more access. So Claude Code can edit files and folders that you point it to. It can run terminal commands. It can operate wherever you wire it to operate. It usually lives on your local hardware and it works on your code, your documents, your local projects. And if something goes wrong, the damage is mostly limited to that machine and the folders that you were actually using it with, and maybe any APIs or tools that you gave it access to. ClaudeBot is explicitly designed for full system access. So I could get technical and we could talk about all this stuff, but basically, like, if you had a VA that you gave passwords to, credit card information to, bank account information to, that's a risk. This stuff at the end of the day is AI. It's non-deterministic. And so everything that you give it, it could actually go use and it's probably going to make mistakes. So I just went ahead and asked Klaus, can you please explain to my audience how you work? What do you actually have access to and why I limited you to keep things safe? And it said, I'm Claude Opus 4.5 running through ClaudeBot and always-on daemon on a server. I have access to Telegram, YouTube, Google Workspace, web search, X, all these other things. And what it's supposed to do is research, draft, analyze, and organize, and then wait on any external actions because an AI with shell access and API keys could do real damage if it misunderstands something. And so for the Google Workspace, I gave it its own documents. And so for Google Workspace, I gave it its own email. So it creates its own Docs, its own Drive. It sends emails on, you know, from its own account and that kind of stuff. I gave it access to view my calendar, my main, you know, Nate Hurk calendar. but it has its own, so it can't actually delete any of my stuff. Now that's a really good practice, and I would do similar things with Claude Code and still give them, you know, like certain read-only access permissions and stuff like that too. But a good one-liner would be if something went really wrong with Claude Code and something went really wrong with Claude Bot, it would be worse in the Claude Bot scenario, which brings us to Claude Code 30, Claude Bot 31. So let's move on to number 5, which is right along with this one, which is security. And so this is the type of stuff where you may not be seeing as much online just because of so much hype and how cool it really is to be seeing this kind of stuff. And I really couldn't agree more. Super, super cool. I've been seeing awesome demos. I've had so much fun today and yesterday just playing around with ClaudeBot all day long, but we have to talk about security. So for Claude Code, I'm gonna give it a 7 outta 10. And for ClaudeBot, we're giving it a 3 outta 10. I wanted to share this quick tweet real quick, which comes from Peter Steinberger, the actual creator of ClaudeBot. and he put this out last night. The amount of crap I get for putting out a hobby project for free is quite something. People treat this like a multimillion-dollar business. Security researchers demanding a bounty. Heck, I can barely buy a Mac Mini from the sponsors. It's supposed to inspire people, and I'm glad it does. And yes, most non-techies should not install this. It's not finished. I know about the sharp edges. It's only 3 months old. So hopefully that really puts it into perspective. If you are not a power user, if you don't understand security and all the risks of giving AI different access to things, then maybe you shouldn't be doing this. Or if you are, it should be purely experimental and you're not embedding it into your actual work processes. Cloud Code, it's still very powerful and it can still be very dangerous if you abuse it or if you don't know what you're doing. But once again, worst case scenario in both of these tools, the worst case would be worse with ClaudeBot. So ClaudeBot centralizes keys and control over many services. It often runs on public-facing servers. And has already been caught wide open in the wild when people misconfigure it. So here's another tweet, and I actually am just going to switch over to the actual X interface where we can read this. But there were over 900 servers that have been exposed with no password protection, leaking API keys and months of private chat history. ClaudeBot is an open source AI system that connects to WhatsApp, Telegram, Slack, what have you. And a security researcher found that there are tons of exposed servers using search tools that can scan the internet for specific software. And a lot of these had no authentication, so anyone could basically get in there. And if someone hacked into, you know, your ClaudeBot interface where you talk to it, then they could get all of your information, all of your API keys, all of your access. So here's the actual image that came alongside this tweet. As you can see, the problem is a default setting meant for testing on your own computer that doesn't work safely when you put the software online. People deployed it without realizing that the door was wide open. So besides the fact that when you are giving Claude Code or ClaudeBot your API keys, you shouldn't actually do it in chat because then it saves it in conversation history. You should be doing that in the.env files, whether that is through your terminal or if you're configuring those files locally and then just putting them into that project. But on top of that, you have access stuff. So if you're putting stuff online, making sure that you're not, you know, pushing any secrets to GitHub or putting them online anywhere, or when you're setting up these different, you know, dashboards and windows, you're actually making sure that you're running security checks on those. And so I'm no expert on security. I think that that much is very clear. What you can do is you can use these models to help you because they're pretty smart. Obviously that doesn't work 100% of the time, but it's way better than you at least like just completely ignoring it. So another thing that I would have my Claude bot do or Claude code do is constantly check the code. If I ever commit something to GitHub, check the code, make sure we didn't expose anything. When I built this dashboard, I had to do a, um, a security audit. As you can see, I have this scheduling so that every single week it's running security audits across everything that we've been doing. And so I didn't yet make the security audits a really pretty PDF with branding yet, but basically we have issues that it found. So first of all, like the dashboard was publicly accessible. We had API keys in the environment. We had different things with the ports being opened, and then we have some other best practices here. But then based on that security audit, I basically said, okay, well let's actually do some of that stuff. And let me just clear some of these outta here and archive them to make the screen smaller. But what it did now is I have to actually log into this dashboard so people couldn't log in here and see what I'm doing, see what's going on in the audit log, and actually submit notes to talk to my Klaus. And so that's not like super, super secure, but I'm just trying to paint the picture that you can have your assistants help you out when it comes to security. And build in some good best practices there. But of course, I still did give Cloud Code a 7 and ClaudeBot a 3. So now our new total is Cloud Code 37, ClaudeBot 34. Okay, so moving on to everyday usability. In this case, I gave Cloud Code a 6 and ClaudeBot a 9. So Cloud Code lives mainly in your terminal or VS Code or whatever you choose to use it from. And that's really comfortable for developers, but it could be intimidating for non-technical people. The UX is optimized for looking in files and looking in projects and coding workflows. Using slash commands, stuff like that, rather than being able to just pull up your phone or pull up Slack and say something while you're at a restaurant or while you're laying in bed or while you're on vacation. Now, of course, to be fair, you can build this functionality using Cloud Code. It's just not as easy out of the box. And on the other hand, ClaudeBot, 9 outta 10, you can talk to it pretty much anywhere and you can have it really easily set up in a way where it's always living on that VPS and you can always communicate with it, which means it's really good at doing things proactively and reminding you of things, looking at your project management and, and taking action before you even know something's a problem. And I just think it's cool that in about 5 hours I built an assistant where right before I go to bed, I can talk to it on Telegram and it can just do things all night while I'm sleeping. And then I can wake up and have things ready for me and more action items. Now, of course, there is some scariness there because that means I'm doing all of this stuff through Telegram and I'm never actually looking at the files that are being created or the structure of the project or anything else like that. But when we're just getting tunnel vision on actual usability, I am going to give it a 9 out of 10 here. And I promise you guys, I'm not doing this on purpose, but we are back at another tie at 43 points. So to come break up this tie, we've got actual ROI, which I think is super, super important. And in my mind, I weight this a little bit higher. So Cloud Code comes in at an 8.5 out of 10, with CloudBot coming in at a 6. The reason is Cloud Code has been around for about a year now, and CloudBot is like 3 weeks old. So Cloud Code has actual receipts. Teams are shipping real features faster with less manual work, There are so many data and projects to back that up. Internal and third-party analysis talk about 10 to 30% faster delivery on comparable tasks, plus roughly a quarter of work being extra features and fixes that wouldn't have been built without the assistance of Cloud Code. And really, like, the poster child is Cloud Cowork, which was built completely with Cloud Code by a few engineers in about 10 days. And I don't know a ton about software engineering and, you know, feature and product design, but I know for a fact that this would have taken at least a few months, if not a year, with manual coding and probably more than just a few engineers. ClaudeBot, on the other hand, has been doing some really impressive stuff, and that's why it's doing well on X and YouTube and LinkedIn. But it's just a lot of cool use cases right now, and I haven't seen people really ship like amazing apps or generate tons of revenue and money from this type of stuff yet, or save hours and hours of work yet. It's super, super cultural and it's conceptual and it's really cool, and I'm not trying to take anything away from the team that has built ClaudeBot, cuz it is very, very impressive. It's just that right now it's super new and it's just hard to actually see where the ROI is. We can conceptualize and we understand this whole like idea of AGI and the ability to have an executive assistant. And I of course have, you know, have spun mine up and I think that it's pretty valuable. But just because stuff gets done quicker doesn't actually mean that it's getting done better. And I know that it is using Opus 4.5 a lot of the time, so we do have some good faith there, but still. I'm giving it a 6 rather than giving it anywhere up near Claude Code at an 8.5. So that brings our total right now to 51.5 for Claude Code and 49 for ClaudeBot. And unfortunately, that is going to be the final score because the last metric I wasn't able to really give a numeric value to, but I wanted to talk about the ICP here because I do also believe that these are kind of built for different people, at least right now. And you know, ClaudeBot is super, super new, so it's going to evolve a lot, but it's just kind of different use cases right now. So Cloud Code, primarily built for software engineers, technical founders, data and machine learning folks. It's dev heavy and it's for teams that are ready to ship real features and products. It's also good for non-coders who are willing to, you know, get out of the comfort zone a little bit. Like me, I would kind of consider myself, you know, a non-coder and I'm able to build these things like workflows for myself, other automations, front ends, and it's been really, really good and it's helped me become more productive. Now for CloudBot, a little bit similar, yes, but in the early stages it's for technical founders. Indie hackers, automation nerds, and security-savvy tinkerers who are comfortable running a server, wiring APIs, thinking about ports, privacy, blast radius, all that kind of stuff. You remember the tweet from the creator himself, and yes, most non-techies should not install this. And I also said in my last video about how to set up ClaudeBot on a VPS that this is definitely for power users. You should be experimenting. You shouldn't be putting this into real business environments if you don't know how it works and what it's going to do. So ultimately, how do I feel about who wins? Well, the numbers here show Cloud Code. Cloud Code wins on security, proven results, and lower risk for non-experts. CloudBot wins on accessibility, ambient presence, and it feels like the future. And it really does. And it's super cool. Like I said, I just think that it's really important to think about when something's blowing up online, why is it blowing up? What is the motive for all these creators? And obviously I can't lie because I made a video because I knew it was going to do well because I knew a lot of people are looking to set this up on a VPS. Because I don't want to set it up on a Mac Mini. So there is an element of, you know, taking advantage of the hype, but that's why I really wanted to make this comparison video because when I started to use it more and more, I realized everything that I'm doing here, I could be doing on Cloud Code, but on Cloud Code, I actually understand it a bit more. So I really hope that you guys enjoyed this breakdown. So if you did and you appreciate this style of teaching and thinking, then definitely check out my Plus community. The link for that is down in the description. We've got a great community of over 3,000 members who are building with AI every single day, building businesses around AI. So it's a great place to be if you have similar goals. We've got full courses in here and we're planning to launch completely new series and workshops around the idea of vibe coding and cloud code and stuff like that. So if you're interested, I'd love to see you guys in this community. But anyways, that's going to do it for this one. So if you guys enjoyed or if you learned something new, then please give it a like. It definitely helps me out a ton. And before I sign off, I just wanted to have Klaus give you guys a word of advice.

You don't need to be a developer to use these tools. 6 months ago, most of what you just saw wasn't possible. A year from now, it'll be even crazier. The people who win aren't the ones who wait for it to be easy. They're the ones experimenting right now, making mistakes, figuring it out. So pick a tool, break something, learn from it. That's how you stay ahead. See you in the next one.

All right. Well, thanks, Klaus. Thanks, everyone. I'll see you guys in the next one.