Grok Bot: Erstelle deine eigenen KI-Mitarbeiter!

0:01The next level of AI is here; we no longer just have simple LLMs, chatbots that generate text for us, and coding agents that develop software, but now everyone can have their own personal agents that are easy to use and available 24/7. I am talking about Crockbot, but there are also other personal agents, such as Dots from OpenAI, Duet from Claude, or even Muse from Meta. It is becoming clear what we can do with these AI models, because we started by simply chatting with AI or letting AI build things, such as writing code. And now we have our own personal agents that have access to a variety of apps. They can work for us

0:4624/7 on our own computer and, via computer use, even fill out forms. They also get better over time, meaning they remember our preferences for tasks and work routines, and we can assemble an entire team of bots. For example, with Crockbot, I have also created a team of digital employees for these videos I make, specifically for video editing, but also for thumbnails, content, research, and data analysis to better understand the analytics of these tutorials. And by the way, this video you are watching was created entirely by Crockbot by my digital employee Moritz, who is responsible for the motion graphics department. Such motion graphics can also be generated very

1:36easily that way. So let's dive right in and start with the simple question of what Crockbot actually is, namely a so-called personal agent. It is not a new idea, as personal agents have been around for a while. At the beginning of this year, Open Interpreter was introduced, and I believe that is where it all started about nine months ago when Open Interpreter went viral—and currently, not many people are talking about Open Interpreter anymore. Sure, there is Open Interpreter 2.0, but you can simply look at the development of the GitHub stars for Open Interpreter again. The GitHub repository from Peter Steinberger really exploded. Open Interpreter is spoken of as the AI that

2:25can really get things done for us. Meaning our personal agent that works on our own machine. We can interact with it, with this personal agent, at any time. There are various interfaces for WhatsApp, Telegram, Discord, Slack, and so on, with a persistent memory system. Um, it can work on our local file system, we can add tools, skills can be deployed, plugins, and I found the concept of the heartbeat particularly exciting here. and cron jobs. The whole thing also exists with the so-called Hermes Agent, and I've naturally already published tutorials on both, and I find this idea so exciting simply because we had the first AI systems here that can truly

3:12work for us even in our absence. Yes, and with the Hermes Agent, I also find it interesting that they talk about— that is, News Research, the AAB behind it—that this is an agent that really grows with us over time, specifically because the Hermes Agent revises its skills after every task and writes this knowledge into markdown files. Then there is also the Muse Agent from Meta,

3:40which was introduced in September 2026. Grokbot already existed a few weeks earlier, in August 2026, and so you can see that everyone is currently entering the field of personal agents, such as OpenAI with the Dots. Funny, you just have to visit dot.com, because if you type in dot.com, you don't end up on the Dot page from OpenAI, but on the page for Grokbot again. So Elon Musk definitely pranked Sam Altman or OpenAI . And last but not least, there is also Claude Duet. That is also another personal agent, this one from Anthropic . And so we naturally have to ask ourselves the question: what is actually the difference between

4:27chatbots, coding agents for vibe coding like Claude Code, Codex, or the Grok CLI, and personal agents like Grokbot, Muse, Claude Duet, or even Dots? And here we can see directly how a large language model works, which we use purely for text generation. We can enter a prompt here, for example a question or a request, and then the whole thing is processed and answered based on its training knowledge. Of course, tools can also be used via function calling, or web browsing can be utilized, perhaps even together with reasoning—that is, the thinking ability of language models—so that text is simply generated again. That is exactly what classic chatbots are about

5:11. We write something, the model responds with text. However, it can also be generated images or code snippets. Coding agents go a step further, and most people use them in the area of vibe coding to actually build solutions for themselves or for software development. This means that, in particular, code is being written here. You can also use the whole thing in the CLI or in an IDE like, for example, Visual Studio Code. But there are, of course, also very powerful desktop apps like Claude Desktop or the ChatGPT desktop app. And these personal agents go a step further because they are proactive agents that actually do work for us. And what’s interesting

5:56about Crockbot is that it really feels like you’re putting together your own team of employees here. And the same naturally applies to Dots, Duet, and Mus. And here we see the whole thing in an overview again. Chatbots are for pure text generation for discussions, i.e., for conversation; coding agents are for software development, running directly as agents in the terminal, but also in IDEs, coding clouds, or even desktop apps; and personal agents are designed so that we can assemble a team of digital employees. Okay, let’s get to a few concrete numbers, data, and facts about Crockbot. And as I said, Crockbot was released in August 2026, initially in an early beta version. I

6:41think this is the right approach to first optimize the whole thing and then actually roll it out. Back then, the whole thing was still using Croc 4.6. One must, of course, also say that the Croc models themselves and also the Croc CLI for the field of agentic coding—these models were never the best frontier models, especially if you were to compare them currently with GPT -6 Astra or Claude Opus 5. These models are, of course, significantly better after all. For instance, Chemie Kart 3.1 just came out, but I’ll also get to an interesting approach from SpaceX SI, or AI as they are called now, namely the possibility of using other

7:27AI models within Crockbot. Perhaps first the question, which isn't so easy to answer, which plan do you actually need for Crockbot? It was announced that everything is being moved into one subscription. Basically, you need Supercoock or Cursor Pro to get access to Crockbot. And so you pay about $ 30 for Crockbot. That’s where it all starts. At launch, it was such that you could only work with Crockbot using the $ 200 plan. Now there is also the small $ 30 plan. You get, I believe, 3.5 times more usage with the $ 100 plan, and then there is also the large plan for $ 200. My recommendation would be

8:15to simply start with the small plan for $ 30 a month. We are currently talking about more than 800,000, um, monthly users, and that, of course, makes it a very rapidly growing software application. Elon Musk even speaks of it as the fastest-growing, well, product he has ever seen. But one must also fairly say that, uh, ChatGPT is likely the app with the strongest growth in history, period. I still like to look back at November 30, 2022, when ChatGPT was introduced with the GPT 3.5 model. Here we see the tweet again; it’s been quite a while, almost four years old this tweet. And it gained a

9:02million users within just one week. Back then, the whole topic of, um, Large Language Models was completely new, and the world realized we can all use AI now and, well, use it for our own fields. Yeah, also quite funny what Elon Musk wrote back then. And by the way, ChatGPT reached 100 million users within two months. Undoubtedly, however , Grok is a very fast-growing application, and we should definitely take a look at Grok. And that naturally brings us to the question of how we should use Grok or how we even start with it. And this is also a very good recommendation from Elon, namely, um, that we just try to create our own

9:49employee with Grok. Maybe we can just think about our own areas of application and then try to implement them with Grok. Things get particularly exciting with this announcement. As I mentioned before, the Grok models, meaning the pure LLMs or coding models, are actually always in the middle of the field. Um, they are never really at the top playing in the league of

10:13frontier models from Claude or ChatGPT. And on October 7. 2026, there was this announcement that we can flexibly use different AI models under the hood of Grok. Specifically, the most powerful frontier model currently available, Claude Opus 5. We can work with Midjourney, so also in the field of creative AI, or even Suno, the AI for music generation. And that naturally enables extremely interesting workflows : for example, you can have your digital employees write custom software with Grok, generate suitable music for it via the Suno API, or. For instance, have images, videos, and so on created directly within Grok. Yeah, now there is also this X subscription. A bit

11:03confusing for me, I have to say. However, it is the case that they are trying to pack everything into one solution. So, Croock Curser was just acquired for 60 billion dollars by SpaceX SI, which was formerly SpaceX AI , and also X Premium. I don't really know everything you get with X Premium. I don't need it, but I have to say I really like working with the X platform right now, especially for the AI sector . Sure, in other areas it might be a bit annoying. Um, there is a lot of rage-bait on the platform, but I tune all of that out and really only use it

11:40for the AI sector and, of course, Crockboard. I have my own Crockbot employee that scans for me and gives me the most important AI news every morning. So, how can we get started now ? Well, I think this tweet is pretty good, and by the way, I find all these posts through my, um, bot that I set up with Crockbot. I simply ask questions about topics, and then I get these, um, tweets directly, for example, this X post here. Specifically: If you are working with Crockbot for the first time, my recommendation would be to first connect a few apps that you use regularly anyway, for example for your

12:21job, for your own business, or for tasks you have on a regular basis. For example, this could be a calendar, Slack, issue trackers, CRM systems, Google Drive, and so on. And then you can just start interacting with the first bot you create. You can just ask, what can you do with these connectors? You can, of course, also carry out an onboarding, and that would be my recommendation as well. Before you create the first bot and say, you do this and that for me, you can first connect these tools and perform an onboarding just like you would with a real employee. And that can also mean that I introduce myself first. That is

13:03exactly how I did it. I said: "Hey, I'm Fabio." By the way, I also used my microphone, simply because I'm much faster that way. And that is also my recommendation: to work much more with the microphone instead of just typing all the time. That way you can provide much more context and describe more precisely what you really want. And by the way, I hardly use Whisperflow anymore because I work a lot with Cloud Code, with Codex, and also with Crockbot. And with all solutions, the native functionality for dictation, the audio function, the voice mode, and so on, is already really good, even in German, so that I don't really need

13:40applications like Superwhisper, Orca, Whisperflow, and so on anymore. That's how I did it. Meaning, I started with an onboarding and introduced myself. I'm Fabio, I like creating content about AI, I'm someone who learns a lot, but I also enjoy passing on that knowledge. I create YouTube videos, courses, and then of course I need course images, I need thumbnails for

14:01them. The videos need to be prepared. Information has to be gathered from the web. The content has to be created first. I also need designs, sometimes even motion graphics. The content has to be edited. And so I first did some brainstorming as part of the onboarding with Crockbot, and we figured out together exactly where Crockbot can support me. And that would be my recommendation. Plus, you shouldn't rush to create multiple bots. You can start with one bot, and since Crockbot uses the strongest frontier models under the hood, it's actually strong enough on its own to handle many, many tasks for us. Most people start with way too many bots right away, and I

14:47wouldn't necessarily recommend that. You should keep things simple at first. That's how I did it, and then you can make things more complex step-by-step and add further teammates. Yes, more and more connectors are being added too . So my recommendation would also be to follow this profile, the official Crockbot account. Here you can see at any time which connectors are coming next, and you can of course also see all that under Connectors within the interface. Crockbot can, for example, create dashboards, reports, memos, and even motion graphics very well through the connection to cloud models, and also especially exciting, you can give

15:30your Crockbot its own email address. That makes the work especially exciting , that every Crockbot, for example, has its own email, its own domain mail.crockbot.com. And by the way, it's also the case that if we have a Supercrock subscription, meaning we pay $ 30 a month, then we not only have the possibility to interact with these AI models on our computer, but the whole thing actually runs on a virtual machine. That means you have a computer somewhere in the SpaceX data center— which is hard to pronounce—and that makes the whole thing super exciting for scheduled tasks or work routines, so that tasks can be executed even in

16:15our absence. And you can easily link the whole thing to your smartphone, so I can flexibly speak into my phone on the go or monitor my workflows with Crockbot. Here is also a recommendation from Elon Musk, namely that you can use the voice mode or the call function.

16:36That makes the whole thing super exciting, because you can call your Crockbot at any time. When I did that for the first time, the bot didn't even pick up. So I have my own digital AI employee who doesn't pick up my phone calls. Yes, there is also a dictation function, and it actually works really well on the smartphone. My recommendation would be to start with the desktop first, but you also have the option to operate Crockbot from your phone. For example, there are also options to link your calendar, reminders, contacts, and so on. It works on both iOS and Android. And there are always many, many use cases

17:19shared here on how you can really use Crockbot. I think it's really good for just getting a few impressions, a few impulses, suggestions, and so on. And there are some super interesting things there, for example, that you can tell Crockbot to join a meeting, for instance in Google Meet or Zoom, and it actually worked well. For example, to simply create a summary of what was discussed in the meeting. I also find the One Password Account function exciting, where you only set up one password and can connect all your tools via this interface. And this, of course , makes it much easier to connect all these tools. Everything can be

18:11controlled via just one connection with a single password. Likewise, there is also a way here to save additional things as well. We can see that here too. Mirx, Chase, and so on, Gitab. You have all of that in one central view. Now I would say let's get started, and I have already downloaded Crockbot; I'll show you how I did it now. To get started with Crockbot, we just visit our browser and we type in—a little joke on the side. That brings us, as mentioned, to the XAI website/bot, and there we can see right away that we can click on "Download for Windows" to download the whole thing. By the way,

18:51here we can already see the entire interface and what it looks like. And I have to say, this interface is structured relatively simply when you compare it to all the other coding agents like Codex, for example, or Claude Code, Antigravity, and so on. There are far more setting options and technical components there, like choosing models, reasoning effort, plugins, skills, tools, memory, and so on. Actually, this is built more for simpler users, I would say, and that naturally has both a pro and a con. You can find your way around it quickly, and the whole thing seems to be very simple. Personally, I have to say that

19:34it bothered me a bit, simply because I've worked so much with Codex and Claude Code, and a lot of things weren't really clear to me. Like, are skills being used here, plugins, which AI model is actually running right now? Because there is also talk that, for example, Claude Opus 3.5 is used. So, many things aren't really transparent to me. I see that more as a disadvantage. So, the interface is very simply structured. Here we see numerous Crockbots in the overview, and there is also, for example, this Chief. But before we talk about a manager or an overall agent that orchestrates other sub-agents or teammates, let's first

20:12look at these created Crockbots for different functional areas, e.g., Sales Outbound. And for this exact area, this Crockbot works, for example, with Salesforce, or with Hubspot, LinkedIn, sequences, and so on. So, flexible interfaces can, of course, also be set up here. Then there is, for example, the Inbox Manager, which could, for instance, manage our emails, bring some oversight here, and automatically assign emails. For example, they can be labeled, moved into folders, archived, showing which ones have been answered and which ones are still open. There is an Account Manager, Talent Scout, Expense Manager, perhaps also for the finance department, and so on. And here is the interesting part: you could set

21:01up a Chief Crockbot that, for example, then orchestrates the other bots. That means it makes the work even easier, because you only have to talk to this " Chief," and they then have the task of orchestrating the tasks and forwarding them to the responsible Crockbots. And there is also the possibility to even work with group chats, and overall the whole thing feels much, much more natural now when you compare it to pure LLMs or coding agents—that is really what it is made for. So, um, so that we have personal agents here. Perhaps also quite interesting when you compare it to Muse from Meta or also with Cloud

21:43Duet. It’s not as if we have one entire thread here. That would be even simpler, one could say, but rather you can organize everything nicely in individual Crockbots, so in individual, um, chats. So the whole thing should really feel as if we have employees here. And I, of course, tried it out directly for my own area for video creation, for content planning, and so on. And I will show that right now in the Crockboard interface. And the exciting part is that it works 24/7 for me. Um, especially such scheduled tasks are particularly valuable, so that I come to my PC in the morning and already have the most important AI news

22:24with really well-researched sources based on X. We just click on Download for Windows and then the whole thing can get started right away. That’s what the interface looks like for me. Of course, there is also the dark mode and light mode here again. Down here, for example, we can go into the settings, and the whole thing is set up relatively simply. So there isn't really that much to adjust. That would be the light mode here in the interface . However, I prefer to work with dark mode. If you scroll a bit further down,

22:56set the whole thing back to dark mode. Here you can, of course, adjust a few more things. For example, you should also activate the microphone. That would be a recommendation, to speak in quite a lot, simply because you achieve even better results there, because you automatically shape the conversation more naturally, like with an employee, when you speak to someone and simply provide more context. You should, of course, also set the language to German , and yes, regarding usage and billing, as I said, I use it via the Supercoock plan, simply pay $ 30 a month, and so you can get started right away, and I, for example, haven't worked that much

23:36this week yet. Of course, if you integrate more and more tools, um, connectors and make the work more complex, then, of course, more of your quota is used up. But for example, with these bots that I have set up, I haven't used up that much quota yet. This is actually my second month with Crockbot, and naturally, the whole thing gets reset every so often. And here we see the first bot. I would recommend that everyone set up a bot like this, because it really affects all of us in some way. Especially those of you following this channel who don't want to miss any AI updates. So feel

24:12free to subscribe to the channel, specifically for the topic of AI news. And I have to say, I really underestimated X when it comes to the powerful information you can find there in the AI field. I’m not the biggest fan of the X platform, formerly known as Twitter, myself. But especially for the AI sector, it is really strong, and

24:33as you can see, every morning at 8 a.m. , Xenia delivers the latest AI news to me. And yes, every now and then I ask. For example: "Yeah, find me something specific on X about a certain topic." And otherwise, it’s always just, um, new news from AI labs or other developers. You can simply click on " view post" here, and it will open directly in the browser, for example, here’s a tweet from Ethan Mollick about ChatGPT, Codex, and so on. How do

25:05you set this all up? It’s super easy. You just click the plus sign here and you could create a new bot right now. You just give it a name. I just named this bot Xenia and simply stated that it was my X bot. And here you can also select a fun icon. There are many, and there are also different colors you can choose from here. And my recommendation , to make it feel a bit more natural, like working with a real employee, is to use real names. That way, you might remember the agents better, or the whole process just feels a bit more

25:41natural. And yes, that is how I did it. Xenia, you can see here in my bar that I have created numerous other AI employees. Besides Xenia, there is also Moritz, Thea, Vivian, Eddie, Daniel, and Connie. And in parentheses, I just wrote down for myself which areas these agents are responsible for. What else can we do? We can, of course, dictate tasks at any time. For example, you can just record your voice. By the way, you can also do this with Control + D: Also search for new updates on Crockbot from the last week. What new features were added? You can pause it here, and then the whole thing is transcribed into the

26:25prompt area. It even spells Crockbot wrong here. Or you also have the option to call your bot, that is, to start Voice Mode. I could do that right here as well. Hello Xenia, what were the three most important news items about Crocbot from the last few weeks? The three most important ones from our conversations. First, Elon Musk announced on October 7th. on X that Grockbot will use the best model in the background for every task in the future

26:59. Explicitly including Claude Opus 3.5. Second, since October 13th. there has officially been the X Subscription, which bundles Grok, Grockbot, Cursor, and X Premium into one subscription with five tiers from Free to Ultra and a shared usage quota. And third, there is no official list of the models behind Grockbot from either xAI or SpaceX. What is running there is currently based only on user observations. Should I go into more

27:28detail on any of the three? No, that’s enough for me. Thank you very much for that. You're welcome, Fabio. See you later. See you later, Xenia. And of course, one has to say that at xAI, it's the case that you don't just have Grok as a model, but rather, as we just heard, you can flexibly use different AI models under one roof with Grockbot, but above all, you also have Grok Imagine, the image model, and then also the voice model; and yes, that works exceptionally well in the German language. Of course, there is also a bit of the American slang that is used there when you call your bot. Otherwise

28:07, the interface is kept relatively simple, one could say. So, when you click on plus here, that bothered me a little bit. You can't even see, for example, which model is actually being used right now. Is it a Grok model, Grok 2? Is it perhaps a Claude model or something, a GPT model? Yes, you can simply attach files and you can, for example, teach it a task. You can use file attachments to provide context, to upload images, PDFs, and so on. I personally prefer doing it like a coding agent, where I simply give access to a specific folder on my desktop, for example on the drive or

28:44something, so that it works directly on my operative file system, and that " teach a task" function is also super interesting. You can simply record your desktop here, and that way I could just show what a work routine should look like. It then looks like this. And there you can show exactly, please format my Excel spreadsheet for example , then I just show the whole thing, click around, and it's all learned that way, and that's how you can teach tasks . The second piece of news that Xenia told us, one of the most important updates, is this X subscription, that the whole thing is being moved into one

29:29subscription. And the most important news is still that we can flexibly deploy various AI models. I find that particularly interesting, specifically to use AI models for the field of motion graphics. The video we saw in the intro was also created by my Crockbot. I simply said, create a motion graphics video. Motion just means movement. Graphics are just graphics, could be images, texts, various icons, and so on; animating such graphic elements can of course also be combined with business presentations, creating such beautiful dynamic animated presentations instead of PowerPoints, or even marketing clips . Let's take another look at the whole thing, and here we see that I simply

30:22entered a story, specifically that we are on the next level of AI—that we don't just have AI models that generate text or write code for us, but that we can really put together employees who then work for us in various functions. Then I also created further versions of it. For example, this was a previous version, the next level of AI: the first level is chatbots, the second level is coding agents, and the third level is personal agents. And that is a so-called motion graphic, and I think the results have turned out really well . And these are my agents here, meaning my team that I put together for video

31:05editing, for thumbnails, for content, for data analysis, for research, and so on. That means these are real agents that work for me and, yes, then get the work done. Then I also created another such video here for the basic functions of Crockbot, and I think the whole thing is illustrated nicely here, namely how does Crockbot actually work, our AI assistant that really does work for us. By the way, I said that this style should be used. The whole thing

31:37was well implemented. We write a task. Crockbot plans and works in multiple stages again like an agent. It works primarily in the background, which is why a virtual machine can also be used here, and then the finished result is delivered. We can use our own computer or browser. We can deploy connectors, set up work routines that are triggered by a schedule, for example, and there is also a memory component so that Crockboard can remember our likes, preferences, and other facts. We can use skills, which are basically Markdown files that simply record how a task is to be completed according to a specific scheme. And that is how the

32:21whole thing becomes reusable. The bots can be integrated or combined into a team, and there are also pages for reports, for dashboards to share, and so on. Yes, overall these transitions were created really beautifully. This is my personal style, and I could, for example, tell Remorets—my motion graphics bot—at any time to capture this in a skill, and then I can, of course, reuse it in the future when I

32:47want to create a new motion graphic. This is how you can work to build up this knowledge step by step. I can, of course, also show you what my other bots are doing here. For example, I also have Thea; she is responsible for thumbnails and has created such thumbnails, and I must say I am, of course, still in the learning phase with these bots. So, I'm not really happy yet. I wouldn't use these thumbnails just yet. What has been done here, however, is quite interesting, namely that there is also VVN, for example, and VVN is responsible for video editing when I have a raw file where, for example, this video is

33:29recorded directly in picture-in-picture mode in a screen recording, because I also have Eddie, for instance. Eddie is my somewhat more specific editor, who then has to assemble separate files and integrate effects during video editing. When it comes to simpler video editing, then VVN handles that. And the exciting thing is that the bots can work together. You simply click on the search icon. Here I can then see all the bots in the overview. But I could also say "group chat," and then I simply create a group chat like in Microsoft Teams or Slack and bring them into a conversation, so VVN and also Thea. And then you could, for example,

34:14just say, "Hey you two, I want you to work together because I need a new thumbnail." Thea, please coordinate with Vivian. She edited a new video for me here. That was the video about Cloud Docs, which has since been published. And yes, just look for the most important, um, parts of the video where I'm looking nicely at the camera, based on those. A thumbnail suitable for the topic of Cloud and Cloud Docs should be created based on that. Yes, I'll try to transcribe that now. In fairness, I have to say that it's actually even better, especially with Codex and Cloud Code. So, it shouldn't be called Cloud

34:59Dogs, like the animals, but Cloud Docs, this new Office feature. And it would be nice here, like with Whisper Flow, if one could change or improve such specific spots or words, and have it all learned. The "um" could, for example, also not be transcribed; that might be nicer. But what it's really about is the point that I can assign such tasks to multiple employees, namely to Vivien and to Thea, who is responsible for thumbnails. And so they can coordinate with each other; they can also chat with one another. And yes , then could. For example, um, Vivien could simply provide suitable segments to Thea, and ultimately a new thumbnail

35:46should be created based on that, and this was done here as well. It wasn't implemented that poorly at all. I think the point is simply that I didn't look into the camera that nicely without, um , movement and so on. Um, I will perhaps simply incorporate that in the future, that I'll just take a few more seconds to really look into the camera, and then that will be picked out by Vivien and passed on to. Here, it didn't work out so well. That would perhaps be the best, uh, version, and with that, I can really save an extreme amount of time and generate such

36:22thumbnails directly from the videos. Now I'll show you what, uh, Eddie did as well. He is also one of my video editors. It was the case that I had, yes, various files, specifically raw files that were incredibly large—more than 10 GB from a workshop I once did —and I also had a matching screencast recording for it. You also have to make sure, um, that the audio tracks match up, that you only use one audio track, so that, um, yeah, the spoken part in the video really matches the picture-in-picture recording, so that my lips move in sync with the audio. He then immediately provided me with several design suggestions. I think

37:08that's the one I liked best, and I found it nice, especially since before he started working for hours with several tens of GBs—that can really take a long time—he gave me a preview first. So, just choose which templates you like best, and I really liked that. Many YouTubers use this too: a screen recording on the left or right, maybe a title next to it, combined with the face recording. And that's ultimately how the entire video was produced. And the best part is, I didn't really have to worry about it while he was working for over an hour, because the whole thing was executed on a virtual machine

37:46and I just looked at the result at the end. Yes, let's also talk about the data analyst, Daniel. And the cool thing about Daniel is that I simply gave him access via the Google Cloud Platform, via the API, to my YouTube account, because I just want to learn a little more from my YouTube, um, analytics. And so, through a scheduled task, he creates some analytics for me time and again, for example, which videos performed particularly well in which months. For example, in December 2024, which was quite a long time ago, I got more than 10,000 clicks on the Microsoft Copilot in Excel video; maybe the timing was just right. There was

38:29also a time when Power Query did well, or content on Copilot in Office applications like Excel and PowerPoint. Copilot also fit well in terms of timing, or for example, now in September and October, GPT-4o Astra performed well, GPT-4o images 2.5, or product updates regarding Copilot. Yes, and AI updates is a good keyword, because if you don't want to miss any updates, want to learn more about artificial intelligence, AI agents, automation, and how to successfully use tools like CrewAI in practice, then definitely check out my free AI community on Skool. And as you can see, we are already ranking at the very top on the Skool homepage. And by the way,

39:16I've updated the layout a little bit here, included the Crockboard logo, and even adjusted it a bit for Halloween, as you can see. Yes, this group is completely free of charge. We already have over 600 members and, as I said, we are one of the most active communities in the entire German-speaking region. So why should you join now? As I mentioned, you won't just find other AI enthusiasts here who are experts with AI tools or perhaps just starting out—where you can easily network and learn together—but there are also free learning resources available. For example, on Cloud Code, Codex, Cloud Corect, the Hermis Agent OpenCla, or n8n AI automation. So

39:59definitely come join us; we look forward to having you. And now, let's move on to Connie. Connie is my content bot and helps me, for instance, to structure my video courses or with general content creation, and you can, of course, combine that perfectly with apps. To do that, you can simply click on "connect apps" here, and you can already see how many, well, apps there are in total. Password managers—you can, as I said, use this 1Password vault to log in to different services using just one account with Crockbot, and that makes work very, very comfortable. In the productivity sector , there are many apps, such as Notion,

40:43of course, the Google Office products, but also Zapier, Craft, Upwork, and so on. So really hundreds of connectors that you can link up. And, of course, for communication as well. I prefer using AGS, but you can also link various communication services to control your bots. For the design area, there is a lot available, and I like to use it for motion graphics or even for coding. As you can see, a lot is really possible with Crockbot, and to wrap up, we'll look at further use cases for how you can deploy Crockbot. For example, you could also use Crockbot for the support area. I found this tweet quite

41:29exciting, where more than 200 tickets were processed in 10 minutes. Or, for example, someone set up a Crockbot here that manages their reservations. For instance, just to call their favorite Italian restaurant and organize an appointment—essentially making a reservation. You can also combine Crocbot with Croc, using Croc for research, to compare options, to work through a problem, and then send instructions back to Crocbot; Crocbot can then, for example, open apps, make changes, send messages, and ultimately get the work done. That means Croc, because Croc is basically just the brain. As I said, you can also use Claude, or any other large language model, and Crocbot is basically the

42:31hands that act for us and get the work done. Or, for example, if you're in e-commerce, you can also use Crocbot to manage your Shopify store. Someone even has their shopping planned by Crocbot, getting a list created, and has expanded on that, or here's an interesting one, using it to replace software subscriptions. For instance, one person killed off subscriptions worth $ 1,000 a year by simply telling Crocbot to rebuild those functions. And yes, he basically went into Gmail, said "check all the subscriptions I have and give me feedback on where you think you can build solutions yourself or compete against them," and just implemented it all for me. You never know if you can

43:27believe all of this, but it would certainly be insane if you could save that much money, considering you only have a few costs a month for this Crocbot subscription. There is also a whole list of workflows here, for example, a Personal X Assistant, just as I use it myself. A prompt is also shared here, for example, or a competitor monitor for leads, or even a group chat manager. Also a subscription auditor here, which has the task of rethinking existing subscriptions, perhaps building alternative products, or a morning briefing, just like I use it with X. And it can create entire agent swarms or, of course, be used to

44:14build a second brain. Then, of course, in the field of marketing. Someone here talks about how they achieved 1.5 million impressions in just one week using Crocbot. Um, then of course also in the creative field, where a music video was created. There is also the integration with Suno here. Overall, I think what was created here is very well done, a video that was also combined with sound. It can also be put to good use in creative AI. Someone even built their own mail app here, and the whole thing in less than 30 minutes , right? For example, an automation to upload YouTube videos to Spotify without having to do it manually, right

45:10? For example, also for managing emails : if. For example, someone writes an email, you could have the whole thing summarized by your Crockboard. That would be a rather simple automation, and my Crockbot just notifies me and can then immediately create a draft, and I can then decide for myself whether it should be sent out or not. Now, however, we are also coming to a bit of criticism at the end, and yes, the whole thing is currently super hyped, these personal agents, be it Crockboard, but also Mus Dots, Cloud, Doet, and so on. And that, of course, raises the question: do you absolutely

45:54need a Crockbot or a personal agent? Because if you look at exactly these workflows that I just presented, you can of course implement a lot using a simple LLM, which already has access these days. If you work with, for example, Chat Desktop or Cloud Desktop via all the plugins, apps, and so on, you can already set up access to all tools. And that, of course, leads to the provocative question: couldn't an LLM do all that too? So, we might not always need a personal agent, and the whole thing naturally reminds me of the year 2025, when everyone was working with Netn, the workflow automation platform. But there are still many use

46:38cases where the whole thing makes sense . Especially the fact that you have a virtual machine in the data center, or that these entire workflows work even when you are not working at your computer yourself—you can just let the work be done. Then there is also talk of Rigen here. This can of course quickly become expensive if, for example, Crocbot simply performs tasks that are perhaps not strictly necessary . And I have to say for myself, the whole thing is just a bit too opaque for me—which models are really running, which skills, which tools are being used—that's not always clear to me, I have to say, when working with

47:21Crockboard. That would definitely be a point of criticism. And in conclusion, the question: what should one actually use now? This was about Crockboard, but there are also Dots from OpenAI with GPT6 Astra. There is Meta Musw. My experience is that Meta is very good for getting started. You only have a single thread there. But Meta's models

47:40aren't really frontier models either. Sure. Um yeah, the new Meta Muse model has definitely gotten much better now, but I don't think it can keep up with Claude Opus 3.5. GPT-4o Astra. I only tried it out briefly. It didn't really appeal to me that much, but maybe I just need to invest more time. I just find that handling-wise, Crockbot is super comfortable. Above all, it’s getting exciting now because you can use all kinds of AI models under the hood. But Claude also responded with Claude Duet. Regarding pricing, you have to say that Duet is at $ 20, so a bit cheaper than Crockbot at $ 30, but

48:18you always have to look at what you’re actually getting for the performance. That one is more expensive at $ 100 and Muse is actually free, but you can't do that much there. So for productive workflows that are meant to become more complex and so on, Muse isn't really suitable, but rather for simple planning. And there you also have the integration, yes, into the Meta ecosystem with Instagram, with Facebook and so on. That doesn't really appeal to me. For my AI area, I prefer the integration with X, and so, this is more suitable for me. Regarding integrations, Duet has, as they say, 6,000 integrations. Dot also has around

48:574,000. With multi-agents, you can say that Crockbot is very good at building various agents, orchestrating them, and so on. And above all, the whole thing is also partially self-learning. You had this recording function. I don't think it's that good, but it is true that Crockbot gets better over time. I wasn't necessarily satisfied with the video editing right away, I have to say , simply because I want to have it clean or in my own style, and my own character is quickly lost in my videos if I just let the AI do all the cutting . But I gave feedback, and by the second, third pass and so on, it did

49:35get better. With GPT-4o Astra from OpenAI, it's the case that computer use has become really good, which is why it's so strong in this area. Duet can't even display its own browser correctly, but it will probably get better too. It also offers the possibility to customize approval roles, so you can set them according to your needs, and it will always ask for approval, especially when sending emails, making payments, or when deleting things. And this is of course super important: security for Do It is currently still on a waiting list in the USA. For us in Europe, it will naturally take even longer, so let’s move on to the

50:21conclusion. So, Crockboard is my recommendation for the personal agent sector for now. I have already introduced the Remy Agent and OpenCla as well. I have to say, it is more suitable for simpler users, and you can get the hang of it very quickly. It is not as technical as OpenCla or the Hermes Agent, and there are really many , many use cases. And yes, I would simply recommend doing your own onboarding with Crockboard to brainstorm which areas you can use Crockboard for and just test it out. I think everything will get much, much better in the coming months, and I am curious to see how it looks then in a

50:56few months, or what Cloud Do It will look like. Yes, that's it for this video. Thank you very much for watching , and see you in the next tutorial. Bye .