← Creator Science with Jay Clouse
Creator Science with Jay Clouse
#291: 48 Hours With Clawdbot: How I’m Using It and Initial Reactions
00:47:49
My experience building a personal AI assistant
Loading summary
Transcript3 lines
- [00:14]
A
Hello, my friend. Welcome back to another episode of Creator Science. I had a full episode outlined and planned to record and share this week for you, talking about what's on my mind as of January 2026. Some general updates about how I'm thinking about content. But over the last 48 to 72 hours I found myself spending a lot of time playing with a new tool called claudebot. And if you've been on X at all over the last few days, you might have seen some chatter about this. I've really fallen down the rabbit hole and spent a lot of time looking at this. And so I thought today what I would do is record an episode that talks about my first 48 hours with the tool, how I found it, what I thought about it, how I set it up, some ongoing thoughts and how I'm using it concerns how you should think about it if you're considering claudebot and just general takeaways for you in case you are looking at this saying, maybe this is something I should spend time setting up, maybe it'll save you some time, maybe it'll speed up your interest in setting it up. Totally up to you to decide, but that's what I want to talk about here today. I first came across this because I saw a video on X from a user named Alex Finn. I've started seeing a lot more of Alex's stuff on X over the last few months. He talks a lot about Claude code and vibe coding and AI in general. And he made a video showing how he bought a Mac Mini to run this claudebot and talked about some of his use cases for how he's using it. And something about that video just made it seem both futuristic and interesting, but achievable for me as someone who I would say it's very fair to call me non technical. I don't have an engineering background. I have a little bit of practice. I've used terminal before, but not to like build full applications or anything. So I'm mostly non technical. But this video made it seem like I think I can set this up. It seems like the creators of claudebot have been thoughtful about how to make it easy to set up. Whether or not that's the right thing to do. We'll talk about here in a second. But it felt like something I could probably dive into. So after watching Alex's video where he talked about having this assistant that he can just chat with in Telegram or WhatsApp and tell it to do things and set up these workflows so that every morning he Gets proactive messages from this assistant saying, here's what I did last night. I thought, okay, let's give this a shot. So what is claudebot? Claudebot is an open source AI agent gateway that connects large language models like Claude to messaging platforms and tools running on a server. So that sounds very technical because I actually asked claudebot to explain to me what it is and how to explain it. But think of it this way. If you've used an LLM like Claude or OpenAI, you see the power of what AI can do for you. Claudebot basically allows you to plug into those LLMs but let it run your machine. So some people are literally taking claudebot, giving it access to their laptop and therefore all of the apps and programs on their laptop and saying, do these things for me. Claudebot can do that. There are, of course, some concerns and considerations with that. If you think about an AI assistant as a human assistant, would you grant a new human assistant who you hired that you have no experience, experience or background with? Would you grant them access to your personal work computer and all of the documents, apps and things that are on it? Personally, for me, the answer was no. I wanted to, if I set this up, create it in an environment where I have very strong control over what it can actually access and who is able to access it, you know, so the reason a lot of people are buying these Mac Minis is because they set that up as its own dedicated environment and they say, I'm just gonna put on that machine the tools and apps and whatever that I am willing to let it control. But there's a step further than this. So once I started understanding what cloudbot can do, I started looking up other videos of people who have set this up, and I actually looked for more technical people who have set this up. And I came across two videos on YouTube. I one is by Neal Stephenson that was published a couple of weeks ago, setting up cloudbot. And the other was by a user named velvetshark and actually really liked the Velvet Shark approach because he was clearly a technical person and he took a security first approach. But both of these users, and I actually suspect that Neil might have watched that same Velvet Shark video. They both set it up on a vps, a virtual private server. And what that means is it's not running on your machine at all. It's actually running on a virtual machine in the cloud, on a server that you pay for. So this is what I did. I dug out an old MacBook Pro that I have since graduated from, and it's Just been sitting in my desk for years at this point. I formatted that disk and I started fresh with a brand new operating system. Then I set up Claude, claudebot, I should say, in a virtual private server on a website called hetzner.com and I could only access that virtual private server in the beginning from that cleaned old laptop. So claudebot only had access to whatever I put on that virtual server. And even if I screwed something up, it only had access to that laptop, not my personal machine. And so the key characteristics of this, by setting it up this way, this assistant is accessible and can run autonomously 24. 7 on the server. Not just when a laptop is open, not just when I open the app and it connects to for me, Telegram, you can connect it to other messaging apps to communicate with it, but Telegram seems to be the easiest and most reliable. So I have it saved as a Telegram chat and I actually use a tool called beeper, beeper.com that aggregates my different chat tools. So inside of Beeper I get direct messages from Instagram, from LinkedIn, from X, and it also pulls in my WhatsApp threads and Telegram threads. So I use Beeper, I have a pinned chat for my Telegram chat with my claudebot and I can communicate with it. It's powered by Anthropics language model. I can communicate with it all the time. So what's cool about this is I can just give it selective access to things that I want and ask it to do stuff. And what's really cool about it is it's an open source project that has built in skills that as people develop, they add them to the project. And you can just ask claudebot to install this skill, whatever that skill is. And anytime I want cloudbot to do something, I say I want you to do this, can you do it? And it will tell me if it can and if it needs things from me. So right now I have it integrated with a few tools. Those tools are Notion, my Oura Ring, my Fathom, Notetaker X, Dropbox, where I keep a lot of my work files kit, my email sending system and my Google accounts like Google Sheets, Google Docs, my Google Calendar. All of those things are accessed via API, not my personal login credentials. So the great thing about this is instead of just using Claude Desktop, which you have to set up like projects, if you want to give it some instructions or memory, it remembers context across all of our conversations and inside of the files inside claudebot. So it's as if it has near limitless memory and a persistent knowledge base to tell you whatever is within it if you need it to reference things like the reason that I connected it to Dropbox is I wanted it to be able to access all of my podcast transcripts, 300 transcripts. I also uploaded a JSON of all the blog posts that I have written historically that I exported from Ghost. So now claudebot can reference any of my past writing any of my past episodes and it runs a weekly process to check and see are there new transcripts, are there new essays, are there new broadcasts in kit? And what it will do is something like okay, new podcast transcript. I'm going to dig through and pull out some insights in short form, draft them inside of notion and score them based on how good or how interesting I think this topic is. And so proactively, anytime I create a piece of long form content, I now have several drafts of ideas around short form that I can look at, take as inspiration, edit them and say okay, this is pretty interesting and publish that. And it's based on my long form content. So this is a taste of what is possible here with this claudebot because what the unlock has been for me, as someone who has not tried to automate a ton with Claude in the past, this feels so much more user friendly to ask Claude to do things on my behalf. I can just text it. I just text it in telegram and say hey, I want you to send me a report every morning going through my X timeline and I want you to tell me what is trending. I want you to do a vibe check of what's like the current conversation on my for you page of things that I pay attention to. I want you to pull interesting threads from people in the creator economy. I want you to pull interesting threads from people in AI and tell me the engagement metrics on those threads. So I just get a curated list in the morning of these different tweets and I can go in and reply to them. I can educate myself. It also gives me a readout of my Oura Ring scores that morning. Which, you know, historically I've opened the Aura app and looked at my scores, but now it just gets served to me in the morning through cloud with my morning report as well as going in and I can reply to these different threads on X. So that's really exciting because essentially what I want to do as a creator is spend more and more time building relationships, reading and creating long form content predominantly through writing. But as a creator, especially a creator who educates creators, I feel so much pressure to exist on social media and short form. And this helps me stay on top of kind of the vibe of what's going on proactively and lets me react and edit, revise short form ideas derived from things that I've already written in long form. So that just feels very interesting to me. So the setup was not the easiest thing in the world that I've done, especially going the route of the Virtual Private Server on Hetzner. The hardest thing actually was connecting it to my existing anthropic account, not via the API. It really wanted to connect to the API inside of my anthropic Claude account, which I think can get very expensive. But I already have a Claude Max account which is $200 per month, and I found how to set up Claudebot to connect to that subscription rather than the API. So my subscription cost stays the same. I was already paying for it and the Hetzner Virtual Private server is about $6 per month. So for an additional $6 per month I have this set up is the takeaway and it's so proactive and user friendly. The learning curve is real, I would say. I came across that initial video about Cloudbot at 8pm on Saturday night. It was 1:30 until I actually had it working on this laptop that I also had to format and start over from scratch. And I slept until about 7:30 when the baby woke me up. Once we had kind of our morning routine down, I spent all day Sunday playing with this, setting things up, giving it access to different APIs that I wanted to be able to access and do things with. And it's been very, very fun. The ironic downside is like I've spent a lot of time talking to it now and because it is so accessible I can just pull out my phone. I've actually created a shortcut in Apple to push the Action button and it opens my thread with my cloudbot so I can just talk to it just by long pressing that button. So I spent a lot of time talking to it. Has that made me more productive or has it made me actually less present? We'll see. I'm going to give another week and I'll update you on that thought because right now I've probably spent more time working over the last 48 hours working quote unquote, because I've been excited about this, I've been in flow and I've been setting it up. But here are some tasks that I have asked Tubi I named my cloudbot Tubi because Tubi is a little creator science mascot. Here's what I've asked Tubi to do so far. So because I gave it access to Notion, I created a connection, an app in Notion, and so I can connect Tubi to specific pages in Notion to give it access. It doesn't just have access to everything I have in Notion unless or until I want to give it that. But right now I have a page called Tubi Dashboard. And within Tubi Dashboard I have a kanban board called Tubi Tasks. I have a content ideas database called Tubi Content Ideas. And Tubi can also add pages to the Tubi Dashboard so it can't touch or read all of my stuff yet. Because again, I'm taking this slow and I care a lot about security, privacy, access, but I can ask Tubi to do things on my behalf in Notion and it will create a page in the Tubi Dashboard. So some things it's done so far it's done guest research for my upcoming three podcast interviews, I asked it to create this structured prep document and we've iterated on the format of this, where it goes through their career timelines, it finds the links to their different social profiles, it looks at other interviews it can access for some ideas. I give it the direction of what I want the conversation to cover. So it will do research in that lens. And this won't necessarily replace all the research I do for this interview. Like typically for an interview, I'll go listen to other interviews with that guest and I'm still gonna do that. But this does serve us a lot of factual information that otherwise I'm spending time on doing. And within that document, I ask it to source with links where it found the information that it's presenting to me. So the next several interviews, the initial research done, incredible. It now has stored all my podcast transcripts, I told you that. Where it can pull quotable lines, ideas for short form, put that into the Tubi shortform database for me to react to. I had it pull all my previous essays like I said, so it can create short form content ideas from that. I connected it to a Google Sheet. So previously when I was just using Claude, I created a fitness project and that desktop version of Claude was creating spreadsheets for me every week for my fitness, where I have a cell that I fill in. Here's the weight for this exercise, here's the number of reps I did. Now that I'm several weeks in, it recommends my weight progression based on my goals. But when I was going out to my gym, I was still pulling up the Google Sheets app. And then Entering in that information in the cell, which is okay, but it's a couple of steps. So I gave claudebot access to that spreadsheet and now I can long press my action button, click the voice transcription and just tell claudebot what I've done for my exercise and it will fill in the sheet for me and then it will generate on a weekly basis a new sheet based on those goals in the same Google Doc. And I don't have to do this prompting anymore because we've created what's called a cron job to do that automatically. So that's been really fun. It just feels so much more natural. There's less friction. And now I can ask claudebot anytime anything related to my fitness because it's offed into that sheet and the progress that I've made. Some other one off stuff that I've done that I think is interesting. As I think more about long form writing, I had a sense for a few writers that I really admire and respect their style and I kind of feel like I'm a combination of those styles. So I gave it the name of these four writers and I said, analyze these writers styles and tell me what it is that makes their style them. And then how can I try to write more like them, but also a little bit more like this? It was a very cool exercise to read through the analysis of these four writers and then workshop with Claude Bott. Okay. How can my writing take more of the form of elements from these different writers? And it literally gave me exercises to write. Okay, so a couple of examples from this pile. I really like how dense and practical James Clear's writing is. I think a lot of people would say he's an incredible writer. It's very practical in that way. But I also like how warm and encouraging Amy McNee's writing is. Her Instagram account is called Inspired to Write. So I had IT do a little bit of analysis of these two writers to see what is it that makes me feel like Amy's writing is very encouraging. What is it about James's writing that feels very practical? And so it gave me some writing exercises to practice to try and embody what it would be like to combine these strengths inside of my voice. A long time ago I had a voice guide created from ChatGPT where I said, here are a bunch of my essays. What makes a Creator Science essay a Creator Science essay. And it gave me kind of a guide. So now that lives inside my cloudbot assistant. And so with that voice guide and with this analysis of these different writers. I said, okay, here are some of the essays that I've written on Substack. Tell me how well I'm doing in embodying some of the different style elements of these four writers. And it did that. And it said, you're actually like 80% of the way there. Here are the ways that you've done this. Here are areas you could push further and gives me very tactical direction on how to do this further. Again, this is all powered by Claude from Anthropic. So you can do this in Claude Desktop. You can ask it to do the same thing. So you can set it up a project file. You can give it instructions to say, you are a writing coach. Here's my voice guide. I'm gonna upload that. And here's some writers. You can do a lot of this without claudebot. My point is the friction now for me to communicate with the same technology that's underneath Claude is just so low. Cause I'm just texting in Telegram. The other great thing though, is that all of this lives within my own system. It's very private. You know, it's just in this virtual server that only I have access to, that I'm only allowing myself to really access from this machine that I don't use otherwise. So it's very private in that way. Another thing I had it connect to was my Fathom notetaker. I said, every day, pull down transcripts from my latest Fathom calls so we can reference that in the future. And in particular, what I did. We had our town hall in the community last week. And in the town hall we talk about ideas and initiatives that we want to do in the future. We had an open discussion. So I said, go to Fathom. Pull the transcript from that call. Then I uploaded the Zoom Chat file that I had saved to my computer. I said, look at the transcript, look at the Zoom Chat file. Create a document in notion with the action items for what I proposed and what the community agreed with on that call. It did that in moments. And it also pulled testimonials from the chat call of people saying, I love the lab. The lab is the best investment that I made in myself. It was because of the lab that I made this decision at the offline event last year. I did this, and it's made a huge difference in my year. It just pulled these testimonials that because I was leading the call, I didn't even see this happen in the chat. And it just did that on its own volition, which was so awesome. So Great. So the recurring tasks, as I said, we have this morning briefing that shows me my aura scores, plus a curated X feed to kind of give me a pulse check and recommend some posts that maybe I could weigh in and give my thoughts. It pulls the Fathom meeting notes every day, saves to memory. So I can go and say, okay, I have a VIP call with this member of the lab today. I can ask claudebot to look at our last three VIP calls with that person and remind me what we talked about. I can see a future very soon where I create my own CRM that is updated all the time from transcripts from call logs, because that can just be so, so useful. It does a podcast transcript sync, the kit broadcast sync once a week, and it's also doing a weekly security scan because I've constantly been asking it security questions to ensure that it is not accessible outside of me right now. A lot of the conversation on X is like, here are the security vulnerabilities of cloudbot, which you should take seriously. And a lot of the reasons people are not secure is because they're running it on their main machine and not thinking about security at all. So having it on a separate machine, that's a good step. Having it in a remote virtual server, that's a good step. Today we did something called implementing tailscale, which makes it even harder to access. So anytime I see a thread on X about security, I send that to my cloud bot and say, hey, look at this and tell me if we have any of the vulnerabilities that are mentioned here. And it will do its own security scan and say, here's where we're already ahead of this. Here's where we might be able to harden this a little bit more. And so probably the most common conversation I'm having with it is, how do we secure this? Are there any risks involved with doing this thing? I'm asking you, and it's constantly telling me what we can do and it will just go and do it on its own. I just say, yep, please do that. So I would recommend, if you are going to set this up, keep that security mindset and keep talking to it about security. Ask itself to do an audit periodically. The biggest threat seems to be code injection. You know, malicious code injection, because you give claudebot access to files and the ability to search the Internet using the Brave browser. I created an API key to Brave. It can search the Internet using Brave, but if it comes across a prompt trying to tell it to do something malicious to my machine, if it thinks that prompt came from me. It could make bad moves, which again, if that's on your personal machine and has access to your credentials, it could go and log in as you to one of these websites that has your credentials, email those credentials to the malicious actor. Which is why I just don't give it access to my personal machine or give it any of my personal login credentials. It's all API access, but it's very intuitive to ask it about security threats. But then you're also trusting the AI to act in your behalf, which hopefully you can. You know this does not come without some level of risk, which is why just keeping as much distance from, from things that are very personal and risky is good. So here in a second I'm going to let you know how I'm thinking about using claudebot in the future and also break down whether I think you should use claudebot. But first we're going to take a quick break for our sponsors. Have you ever had one of those conversations that feels like it completely changes your perspective, maybe even your whole life trajectory? Those are the types of interactions you'll have in the lab. Our membership for creators like you, I know you are a good fit for the lab just because you listen to this show. If you're drawn to the people and the ideas we showcase here, many of whom are members of the lab, by the way, you would love the conversations we have inside the community. Inside the lab, we all share the results of the experiments we're running. The good and the bad. By learning from the successes and failures of others, you, you will grow much faster. I'm running my own experiment for new members of the lab. I'm offering $200 off your first year using the promo code POD. $200 is 29% off the basic tier, by the way. So take me up on this. That's $200 off using promo code podpod at checkout. Just go to creatorscience.com lab there's also a link in the show notes.
- [25:32]
B
This episode is brought to you by FX's the Beauty Official Podcast. Join host Evan Ross Katz on the Official podcast for FX's Hotte the Beauty, taking you behind the scenes with its amazing stars as they discuss the show's most jaw dropping moments. Featuring Evan Peters, Anthony Ramos, Jeremy Pope, Ashton Kutcher, Rebecca Hall, Bella Hadid, Meghan Trainor, Isabella Rossellini, Jessica Alexander and Ari Grayner. Search FX is the Beauty wherever you listen to podcasts.
- [26:04]
A
All right, so how am I thinking about using cloudbot in the future? I Think about it from the perspective of research ideation, automating some stuff that is annoying. You know, this episode right here, once it's recorded, I'm going to send it to my audio engineer. Once he masters the episode, we'll have an audio file that typically we would put into a tool called CastMagic to get the transcript. And I have a lifetime deal with CastMagic, so that doesn't cost me anything. But I think I could probably just take this file, give it directly to claudebot to transcribe and make show notes out of it. But I think what I'll, I'll do, I'll take this a step at a time. I'm probably gonna put it back into castmagic, but I'm going to then ask claudebot to create an ongoing prompt for show notes to give us things like, hey, here are things that you mentioned that you would put in the show notes. Send me the links. You know, at the beginning of this episode, I said that there are a couple YouTube videos I watched to set this up. I'm going to set it up. So claudebot prompts me to find those videos to send it to it so it can create show notes. That includes all the things I promised to give you in the show notes. I think our show notes have a ways to improve. Right now we're kind of using the default Cast magic prompts. I know I could go into Cast magic and make my own prompts, but I'm going to set this up now as a workflow in claudebot to pull down the transcript, make show notes out of it with timestamps in a certain format that fits to make my assistant's job even easier of posting this in Megaphone, our host. And maybe I'll be able to connect it to Megaphone and actually just have it fill in that stuff itself. I don't know. We'll find out. But I'm gonna take it one step at a time. I like it as a research partner a lot because it can identify things, sift through them, or present me a huge pile of stuff that I can sift through and make strong decisions. I love that I can build this institutional memory inside of it in this private server of all my podcast transcripts of all of my research documents, because I can go to it and say, hey, I want to create a carousel about some of the most referenced books on our podcast. Go through the transcripts, find all the books mentioned, rank them by number of times they're mentioned, make a document in notion, and then propose an order for the carousel, you know, and then I can have my team, my designers, create a carousel around that. It does a lot of this formatting and research that none of us particularly enjoy doing. But as a step of the process to make the thing or get to the part of the process we do enjoy doing, I think it can do that and it can do it consistently in an automated way. If I just set up the process, it can be very proactive. I saw something that Alex was doing, was giving an access to his personal to do list and saying, what are the things that you can do for me on this list? And just letting it run overnight. Maybe I'll get to that point. But again, I've been keeping it kind of away from the majority of my personal documents and stuff. But I like the idea of saying, here are things I need to do. What can you take off my plate? Claude Cowork, I think also can do this if you're using the Claude desktop app. So if that feels more reasonable and accessible to you, try it. Because again, I think a lot of this can be done just with Claude desktop if you know what you're doing or Claude code. Claude Cowork claudebot is a more intense and risky, I think, implementation of this. But there are some upsides in terms of institutional knowledge, the context window and privacy being on your own machine versus within the confines of this app. Other stuff I've been doing in Claude in the past is I will give it drafts of my essays and say, where can this be stronger? Where can it be weaker? I like doing that inside the Claude desktop interface. Doing that in Telegram feels a little weird. Like I know that actually in Claude it will create some artifacts for me as a PDF, as a markdown file, as an image or a spreadsheet. And I'm sure claudebot can do that as well. But there is something about the Claude app that feels more intuitive for that. So I do think I'll continue to use the Claude app from time to time. And in fact, sometimes I've broken my connection to claudebot and I've had to go into Claude the app and say, hey, fix this. What's broken? What's going on? And it helps me get it back up. So I don't know that this will completely replace Claude for me, but I can tell you I haven't really touched ChatGPT in a long time. We do have the OpenAI API tied into Claudebot because I wanted to send voice notes so I can send a voice note. As I was saying, with the Fitness example and it will transcribe it and answer it for me. But man, I'm just so impressed by everything Anthropic has put out. Making stronger tooling, I think than OpenAI. So should you use claudebot? Ultimately, I think this is a good fit for you if you're comfortable and willing to learn some basic server and terminal skills. It certainly makes you feel confident and comfortable because you ask it questions and it gives you an answer that's way more technical than you understand. You're like, okay, that sounds smart, go ahead and protect yourself in this way. But I do recognize that being non technical, I am ultimately trusting the AI underneath this and the team that is contributing to the open source CLAUDE bot project. But it's cool that it is open source because as people use it, they make it stronger, they make it better. And then my claudebot can ping documentation of the claudebot project and update itself with security patches and things like that to continue to make it stronger. If you want an always on assistant that integrates with your tools, this can work. But again, I think you can probably do this with CLAUDE code. I just never did. Personally, this is kind of the key thought for me that I would like to pass on to you. CLAUDE desktop, CLAUDE code, Claude Cowork or OpenAI or any of these tools. Ultimately they're all tools, they're what you make of it. And if you can set up your tool stack to do the things that you want, you don't need to implement a new tool stack because this was quite a bit of setup to get it up and running, like a full 24 hours, probably of just getting things set up in a way where we could communicate more frictionlessly. But 24 hours is also not a huge cost to get this going. If this continues to save me time and do these little tasks for me, it's been pretty awesome so far. I do think, you know, you want to keep thinking from a security lens and so if you don't want that on your mind, maybe don't use cloudbot. If you want to put cloudbot on your personal computer, I personally just wouldn't recommend it. I would say definitely use a virtual private server. The one I used again is hetzner.com it's a little less convenient. Like what I see about people using the Mac Minis, that does seem attractive is you can see the code base on your desktop and you can just drag and drop files that you want CLAUDE access onto that computer. You can airdrop it from Mac to Mac in that way and there's something attractive about that for sure. But the Virtual Private Server seems way safer. There's a step removed from getting your files into it, but it feels way safer. So that's what I would recommend. If you're a little intimidated about Terminal and Command line, maybe hold off. You know, if you want a plug and play user friendly solution, this is still pretty diy, still pretty hacky. I think it's the direction everything is heading is in this direction. So I think we will see more user friendly versions of this. And again, I kind of feel like Claude cowork is a lot of this combining Claude cowork with Claude code. You know, I've talked about this being proactive, claudebot being proactive and doing stuff for you, but ultimately that's just code. Those are just what's called cron jobs. These are hooks and workflows. I think you can set that up with Claude code and CLAUDE cowork. So if you're looking at this and you're like, I don't know, it seems like another thing, but you're feeling fomo. If your systems are working, just keep using them, you know. But if you're drawn to it and you want to try stuff out, try it out. Who am I? Who am I to stop you? I'm having a lot of fun with it. And again, if you're taking a security first approach, I think it could be fun for you. So some questions that I asked claudebot to add here to anticipate what you may be thinking. How technical do you need to be to use this? I think you need to just feel comfortable using Terminal and comfortable talking to LLMs like Claude or ChatGPT about using Terminal. They have come so far at just giving you step by step. Here's how to set this stuff up. Can you run this on your Mac instead of a Virtual Private Server? Yes, but then your Mac has to be on all the time. The Virtual Private Server is on all the time. My current cloudbot Mac is closed, it's asleep. And I can text cloudbot right now because it's on a Virtual Private server. The Virtual private server costs 5 to $10 per month. Then you have your CLAUDE subscription or Claude API costs or whatever API you're using to plug this in. I'm using Claude OPUS as the model to do most of this, which is heavy powered, it goes through a lot of tokens. But I'm on the Claude max subscription. It's $200 a month. I did hit a limit yesterday. It's a five Hour context window, so I had to wait about an hour to use it again. But then I had claudebot set itself up to have Gemini as a fallback. So in the future if I hit my Claude opus limit, I can fall back on other models and you can set up multiple models. A lot of people are recommending a tool called Minimax, which I don't know a lot about, but apparently it's very capable and uses far less tokens and energy. So you could do that setup took a few hours to get it going, 24 hours to fully feel like it's up and running and capable of doing things I actually want it to do. What can go wrong if your server is exposed, if your credentials get leaked, if there's unauthorized access. Those are like the horror stories. Which again is why everything you can do to harden the security and ask it like, where are we exposed? What can we do? Here's an X thread. Are we at risk here? I'm just constantly asking those questions and I feel like my setup now is pretty strong. Should you connect your email or banking? Personally? No, I would not do that. I would not give it access to like my 1Password. I haven't even given it access to my email. People love giving it access to email and having it clean up, but it seems like it's mostly just adding folders and tags and archiving things. I already have a pretty good system in my email inbox with my assistant who I do fully trust. I don't really want it out in the world communicating as me, so I'm taking that slow. Can it write content for you? Definitely. And this is kind of my existential question of where is content heading now that it's becoming so easy to generate stuff. I think at the end of the day, anything with AI, you've got to ask yourself what am I outsourcing here and am I outsourcing difficult things that I want to get better at? Because if you want to become a better writer, but you're asking AI to write everything for you, you're not going to become a better writer, you're going to become a worse writer. So I'm always asking myself, is this a skill that I want to personally get better at? If the answer is no, then I feel more eager to use AI drafting posts for X based on my essays. I'm going to keep trying to see if I can get that to work. Honestly, if I can get short form drafts that are derived from my long form work, that are my ideas that sound like me and I don't have to think about it all that much. I don't hate that because ultimately, like I said, I want to spend more time reading, writing long form, creating long form. Could be podcasts, like this, could be YouTube videos as well. And so if AI gives me more time back to read more, write more, practice my writing, I think I'm comfortable letting it write some short form for me. If I feel like I want to keep showing up on those platforms more consistently and repurposing the IP that I'm putting into long form. Curious to hear from you, listening to this, how that strikes you, because I'm still a little, a little on the fence about it. Because if my X feed is just AI created short form content, how do I feel about that? I don't feel great. But ultimately, if it's the writing someone has already done in long form, I also don't feel bad about it. I don't know, I'm kind of on the fence. You've probably seen over the last few months that I've pulled back a lot on short form because it just simply hasn't been as big of a priority for me as long form. But there's value there and I feel like the future belongs to people who have distribution. And if my long form writing enables me to create short form content that I can publish and build distribution and reach more people, help more people, it seems like I shouldn't say no to that. The line of integrity I draw is, is this my writing repurposed? Is this my ideas? Yeah, I don't know. I'm still, I'm still thinking about that. So you have to figure out where you draw the line, what feels right to you. But that's currently how I'm tossing it around. Will this replace members of my team? I don't think so. I do think it will displace some of the work on their plates and give them more time back to do more creative tasks and thinking. But I have no plans to change my team makeup because of this. Any integrations that I wish existed. One problem I did run into because of running on a virtual personal server. There was a Skill built into Cloudbot that was called YouTube Watcher, where you could drop a YouTube link, it would pull the transcript and let you talk to that video. It seems like YouTube is blocking that from VPS's, so I can't actually do that. I can manually copy the transcript and paste it in, but that's a research aspect that could be a little more frictionless. I don't know the thing I wrote in my essay this weekend, my newsletter this weekend was in a world where we're using AI to summarize content and or using AI to create short form content, what is the role of content? What is our role as creators? And when I think about my own behavior, I'm not consuming less content, quote unquote, but I am consuming differently. So the question I'm trying to keep top of mind all the time is what types of content am I personally proactively seeking out and placing a premium on? And I came up with 1, 2, 3, 4, 5 different buckets of content that I'm still personally consuming and placing a premium on. And those are long form writing books, substack articles. I love seeing a window into how people think. And I am assuming when I read that long form writing that it was written by a human effortful art, things that are verifiably human effort. This is a lot of like the reels that I watch. Artists who are just doing crazy artists and engineers, really all kinds of people who are doing crazy things with their time and capturing it on video. I love it. I love seeing what humans are capable of, especially when they've dedicated months to a project or years to their life to getting good at it. Demonstrations are becoming increasingly popular. A lot of the long form YouTube videos I watch now, they are either interviews, which I'll talk about here in a second, or live demonstrations of how people are doing something. Because I can go to Claude and ask for a step by step on how to do something. But it hits a little bit different when you watch somebody demonstrate how they do the thing and why they do it that way. That's a verifiable experience that I really enjoy. So demonstrations are definitely still popular. And personally, I think in our interviews I'm going to be doing more of probing and asking people like, show me how you do this. So if you watch on YouTube, you'll actually see a screen share of people showing how they do a certain process or thing in their business. The fourth bucket that I pulled out were interviews with people who have unique, verifiable experiences. I love the direction that Tim Ferriss has been going on his podcast where he's basically like, I'm opting out of the book tour Ferris wheel. And he's just following his own curiosity and talking to people who have unique human experiences. And I don't know what I'm going to get out of those episodes, but I find myself trusting certain interviewers like Tim Ferriss, like Rich Roll, knowing that whoever they bring on, there's a reason they discerned that they should bring that person on and they're going to guide a really interesting conversation. So I still watch a lot of long form interviews. And the fifth bucket is good hangs people who, when I watch their content, when I consume their content, they make me feel comfortable, at ease, more confident, like I'm having fun. Some of the podcasts that are good at this, Amy Poehler's podcast literally called Good Hang is great at this. Vulture's comedy podcast, good one is great at this. Mike Birbiglia's podcast is great at this. Pete Holmes podcast is great at this. So I enjoy spending time with creators who I enjoy spending time with, but what I'm doing less of is getting clickbaited into a video because they made me uncomfortable with the package. Not knowing some answer to something and then suddenly I just watched a 20 minute video. I'm able to reclaim my attention back from that because if I'm baited into watching a video that I don't actually want to dedicate 20 minutes to watch, it's easier than ever to pull out and close whatever curiosity loop was opened. But I do think as creators, we have to think about how do we show up and positively contribute to people's lives now that we have this unbelievable technology in front of us? I just don't think we can deny the technology that is here and the fact that people really value their time and they want to get time back, they want time savings. And so I think we have to be intellectually honest and amoral about this and say if people are going to use these technologies to compress and compact our content because they want the answers held within it more quickly than we're giving it, maybe that's not the way we should be creating content. Maybe we need to find ways to show up that people don't want to compress that experience and they want to experience it fully. That's partially why I'm really excited about a future where I'm spending a lot more time writing long form and writing books, frankly. Because the time that I get back from AI and tools like cloudbot, I want to repurpose into activities that I really enjoy, like reading, like spending time with my family, like having human lived experiences that then I can reflect back into content that is unique and different than just straight up how to step by step that can be generated based on my particular experience. But it is a big change. It's a big change for educators like us. I think we cannot put our head in the sand and hope that it doesn't come for us. I think it's coming for us. And we have to ask ourselves, how do we show up and contribute and add value and serve in the way that people most want us to hang out to show up? So that's how I am using Cloudbot. That's how I'm thinking about AI in the season and content in the season. I hope this is instructive to you. I'm going to try and link everything I mentioned here in the show notes. Cloudbot's going to help me on that. If you have any questions, you can Tag me @jclaus on social media and ask if you enjoyed this episode. Leave a comment here on Spotify and let me know. Leave a rating or review on Apple Podcasts if you haven't already. Thank you for listening as always and I'll talk to you next week.