Grok Bot Explained: AI Employees With Their Own PC

Grok Bot landed this week and a surprising number of people have already switched to it. Always on AI agents, each with their own computer in the cloud that talk to you like real colleagues in Slack. If you run a business and you're wondering whether to hand over 200 bucks a month for that, this episode is the honest answer. We get into what it actually does, the lock-in that makes it a bit of a poison gift for some, and the open source alternative that does most of the same thing for a fraction of the price but with a few trade-offs. Then ChatGPT's new computer history feature records everything you do on your Mac and then tells you what it can automate for you. We turned it on and let's just say the results were quite surprising. I'm joined today by Gael Breton, my co-founder at Authority Hacker, where we tell business owners what's worth doing with AI and what actually isn't. But before we get going, I want to have a quick word about the AI Accelerator, which is our membership for business owners that want to learn and implement it. AI and automation in their companies. Almost a thousand of you have joined now, and we've been adding a ton of new content and features recently. We've got our new ChatGPT and Codex course, which sits alongside the Cloud Code course for business owners. We've got WorkOS, which is our operating system, which we've built for business owners and sits inside your AI and makes you 10 times more efficient. We've got our MCP, so your AI can connect into your membership and get all of the context that all the discussions courses and discussions courses, everything in here and implement it directly inside Cloud Code or Codex, whatever you're using. So if you sign up, you don't have to sit there and watch hours and hours of videos. You can have your AI do it for you. And now absolutely is the best time to sign up because the price is going up. It's going up on the 1st of September. It's currently $129 a month. It's going up to $149 a month. But if you join before the 1st of September, so about a week and a half after this podcast goes out, then you can lock in that $129 a month rate forever. There's no long-term commitment here. You can cancel any time inside your dashboard. So if you want to try it out now, head on over to authorityhacker.com forward slash AI accelerator. Bot has landed, Gael, and it's taken the internet surprisingly by storm. A significant number of people have switched to using this now. You're a bit skeptical, though. You want to tell us why? I don't think I'm skeptical. I think it's cool. I mean, let's just explain what it is first. And I'm going to share the page for it so you will see it. And it's like, basically, Grok Bot is this application that is released by XAI. So XAI now is also Cursor. So Cursor was bought by Elon Musk. I think it was for $60 billion or $60 or $90 billion. A lot of money. A lot of money for a code editor. But the point is, this was built by the Cursor team using the Grog models. By the way, Grog models actually are good now. If you actually check out typical analysis, they released Grok 4.6. It matches GPT 5.6, basically. So actually, XAI is now a significantly better model provider than, say, Google, for example. Just to put things in perspective. So basically, they released this Grogbot thing, which is open cloud slash Hermes agent without the complexity, without having to do all the setup, etc., which I think was a big part of why people struggled with it, especially normies, people who are not technical, etc. Yeah. So before you had to either, you know, you had people buying those like Mac minis and setting up rules of them at home and installing it, or more often they would buy rent like a remote server and they had to, you know, go in and there was a lot of code. And of course, you could use AI to help set up, but it was a barrier. And this effectively removes that barrier. Yeah, kind of. And it's like, what's really kind of interesting is that it has its own computer. So actually, it has a cloud computer and has a browser on it and you can open it like it's basically a Linux desktop, but like, you know, there's Chrome installed and everything. So you can use computers, but without using your computer, your computer can be turned off, like etc. It works on your phone. And I think the big paradigm shift is you're talking to agents, like you talk to people and you have just one continuous chat with them. And the same way as like, I don't make a new session when I send you a message on Slack, for example. And so that allows you to kind of give them a personality and so on and to be like, okay, this is my account manager. This is like the sales guy. This is the guy who manages inboxes, etc. And you kind of give them roles and then they kind of go and do things for you and they have their own computer. So it's the closest thing to kind of like a plug and play AI employee. And that's why people are loving it. We've seen this quite a lot, actually, where people kind of sometimes struggle to get their head around different workspaces and folders and even the fact that their information is stored in a folder. But when you explain it to them, like, oh, you've got your chief of staff, your marketing manager, your virtual assistant. They're like, all right, I get how that works. I know how to interface with that setup. Yeah. And that's, I think for like knowledge workers and people like that, they love it. And so like, it's almost like a Slack where you have that and you can make group chats. So you can actually group multiple agents and work together on a project. And, you know, it's like each of these agents also has different access to different tools. So like, for example, like the inbox guy has access to the inbox, but maybe the account manager doesn't have access to the inbox, but they can work with each other. The account manager needs something to be sent. They can message the inbox manager that can help do it. And so they work like a team. You give them different access and they can do different things. And so that's overall, like, why it's worked so well. Now, let's talk about the negative of this, which is the pricing, because there's a price to this ease of use. So you can see, like, we're checking it out. But if I go to pricing, if you actually want to get access to this, you're minimum paying $200 per month. So it's not exactly a cheap thing to buy. And that comes with its own usage limits. They're pretty generous, to be frank. It's like it seems like the new Grok model is actually very cheap, much cheaper than GPT 5.6. So, you know, as smart, cheaper, pretty cool. But you're kind of locked in when you're out of usage. The only thing you can do is just buy extra usage through them. You cannot plug an API key. You cannot do any of that. You're locked in. And especially I can imagine people building, like, complex setups, et cetera. Like, it's hard to move out, basically, once you've built all the plumbing, et cetera. And so it's a cool thing that they basically made it a lot easier. But you're going to pay for it via the fact that then they can do whatever they want with the pricing and how they sell it to you. So interesting concept. But then what happened is Hermes, the guys who made Hermes, they actually released Bot Mode. And Bot Mode is... We're talking about the Hermes agent, not the bag company, right? Not the bag company, no. No. That would be weird. They released something called Bot Mode. So, like, Hermes used to work a little bit like Codex and Cloud Code where you'd make threads. But they always had profiles under the scene, right? It's like they always had... You could make multiple profiles. In our Hermes setup, we have, like, three profiles that are actually on our Slack. And then we can interact with them. And they're completely independent from each other. But now they kind of, like, they build that inside their desktop app. They are building a mobile app as well. It's coming soon, apparently. And so this is kind of like the open version, the open source version of this. Because you can connect any API, any provider. You can buy a TransGPT subscription. You can put your Cloud subscription. You can use Open Router. You can, you know, deep seek if you want to spend very little tokens, etc. And you can do whatever you want. You have to use that because there's no, like, Hermes model. Is that right? Actually, they do. Like, if you go on their site and you go to Hermes Agent, you will see that if you go in their product, they sell, like, Hermes Cloud. And basically, you can buy, like, a subscription. And basically, they handle everything for you. And they will, for 20 bucks a month, you can see they give you, they host it for you. And then they let you still plug whatever model you want. So it's like, they have a version of this that is cheaper, gives you some inference as well. I mean, like, looking at that website now, like, it looks like someone's. You compare these to Grok Bot. It's a bunch of, like, ASCII art. And, like, it just doesn't look like something. That's the problem. This versus this is a problem. But it's essential. And then the thing as well is, like, Hermes is very, like, you tinker with it. You play with, like, how it uses memory. It's kind of like a geek tool that you can do a thousand optimizations with, but you have to maintain. A little bit like WordPress versus Webflow, for example, if you're building websites. And then people, a lot of people are like, I can't be bothered dealing with this. I know a lot of entrepreneurs who love Grogbot right now because it's just plug and play. The memory works. It works. Yeah. And you don't put your credit card in and it works. And sure, you pay more. And sure, you're locked in. But they don't even care. They just go for it. But it's still interesting to see that there's basically going to be two types of people. The people who are happy to get locked in in exchange for ease of use, basically. And there's going to be the people who want to tinker. And they will go to Hermes and essentially do the same thing for cheaper and with more control. I want to talk about that lock-in conundrum, though, because that is a big potential risk. I mean, we've seen basically every month like a different AI model or a company kind of shoots to the top. And it's quite important to be flexible. And, you know, if you had tied yourself to Gemini a year ago, you wouldn't be able to take your time right now. So, I mean, who knows what's coming? Maybe Grok does become the best one and everyone else. There's a chance. Who knows? But also there's a decent chance that that doesn't happen. So if someone is tied into Grok in this way, GrokBot specifically, how do they get out of it? How do they switch? Like, is it even possible to get your agents and history? I think you can ask it to prepare reports. And, like, you know, again, for example, our Hermes. So we run Hermes and Hermes is, like, very technical and there's lots of things to do. But I never use it that way. I have a Hermes project on Codex. And Codex handles all the tech. And I just talk to it. I'm like, oh, I wish we could connect to this memory provider. I wish, like, the Hermes agent is doing this thing. I don't like it. Can we change the settings so it would do this instead, et cetera? I'm just speaking in natural language. I'm like, you go and read the documentation and then you operate the VPS and you do everything. Even the VPS was set up by Codex for us, right? So it's like, I never logged. We use DigitalOcean. I never logged on it. I just used the terminal tool. It prepared the server. It installed Hermes on it. The only thing I had to do was it opened the browser tab where I had to authenticate my OpenAI account so I could use the subscription for it. And that's it. And after that, I just talk to it. I'm like, oh, I wish my Hermes agent would do this. I wish my Hermes agent would do that. So there's kind of ways around it. And it's kind of like, to me, it's the best of both worlds because you get a much higher degree of simplicity in terms of I'm just saying in natural language what I wish this would do. And then the tech is handled by the bot. And at the same time, we're not locked in because, again, OpenAI is supposed to release a very good model early September called Astra that will probably leapfrog the Grok model. And it's like you'll be with your Grok bot and you'll be like, fuck, I wish I could plug Astra into this. And you won't be able to arise on Hermes or just be able to switch that, for example. And same if Grok takes over. I can just buy a Grok subscription and plug it into it. And boom, we're using Grok, you know? Yeah. So, I mean, essentially, people are paying for that, like, for the perception that they don't have to deal with complex technical setup, even though what you're saying is not actually that technical. And I'll show you how it looks on the Hermes agent, like, just to show you, like, how close it is. All right. This is the desktop up on Hermes agent. And this is kind of, like, the same thing. And I'll just say hi to one of these bots. Like, I started a new session because I don't want to show, but, like, you can start multiple sessions or you can keep going on the same one up to you, basically. And we'll just connect and answer. But, again, you see it's okay, but it's not as clean as this Grok bot experience. But it's mostly the same, basically. So, you get the idea. You can do that. Yeah, yeah. So, for anyone who's watching or listening and thinking, well, I already use Cloud Code. I already use ChatGPD Worker Codex. Yeah. You know, should they be considering this category of tools, either Grok bot, Hermes agent, or any of the other ones out there? Like, what are the use cases for business owners for this? I think I had this question on a consulting call yesterday, actually. And my answer was, like, if you're a soloprono, don't bother. Like, just do everything on your Codex or your Cloud Code. I like these tools better. Like, to me, it's like, you know, if I try to do real work and I have the same prompt, I feel like I get a better output on ChatGPT or Cloud than I get from Hermes. Because there are different harnesses with different sets of instructions and etc. However, there is a use for shared bots. Like, and the way I view Hermes is it's a bot that lives in your shared space with other people. So in our case, it lives in Slack, for example. I never used the desktop app, for example. Like, that would make no sense for me. I just talked to it on Slack. And we use it on Notion as well. We can set tasks to Notion. So let's say you set myself a task. You send me a task and you're like, oh, girl, can you go and find, like, five customers that said something positive about the latest calls? Let's say I received that. I can set a subtask to Hermes. I would just go and do it and then grab the result and just give it to you, basically. I still take the responsibility. But as a way, I can just call this sub worker without switching, without opening another app, etc., directly inside Notion, which is really nice. Another example of how we use it, for example, is, like, our social profiles. Like, we have a Hermes bot that has access to them. And if anyone in the team has an idea for a social post or something to share that we should share as a company, they can DM that bot. And it will be able to draft and post a post based on an idea. You don't have to give it to me that has my codex connected to it or something like that. Like, it can go directly to the bot without talking to me. So it's pretty cool. And these bots, I actually gave them coding agents access and they can submit PRs on GitHub. So they can do coding. They cannot publish it, but they can just prepare the code. So let's say you have an idea of something we should do on a website, right? You can message the main agent that we have on Slack. Be like, hey, I'm on my phone. I'm at the airport. But I'd like to kind of work on this. And you drop it like a voice message. It will prepare the code. And then you can just run it locally on your computer. Or I can run it on my computer and review it. And then it has done most of the work. I don't have to wait the 30 minutes it would take for a codex to do the work, basically. I can submit it for review. So it's like, yeah, for me, it's like as businesses, these are interesting as shared agents. I see them as like AI employees. And then your codex or your codex code is like an exoskeleton you get yourself in to kind of get more done. And I tend to give more access to my codex or codex. It has access to my entire computer, basically, because I'm using it interactively mostly. And I'm here to make sure it doesn't do something stupid. Whereas these agents, they're more autonomous. Therefore, I'm more careful on how much access I give them because obviously they could do something while I'm sleeping. And I don't want them to do something stupid. So there's a lot more consideration on safety for these bots because they are shared and they can wake up in the middle of the night and start doing something without you being aware of it. Yeah, I mean, I'm not sure I entirely agree with the statement that like if you're a solopreneur, you don't really need to use that. I think that I think it's a very good description of how you explain it. Like Claude Codex is an exoskeleton. Hermes Grockbot, it's like creating employees. Yeah. But, you know, it's very interesting, I think, for a solopreneur to want to almost create employees to work with, like before they would realistically need to hire one in a team, you know, their business is that size or whatever. You know, I see this more as like a lot of the kind of autonomous stuff that NAN would do in the background. Yeah, but then you have the scheduled tasks on Claude Codex and Codex that could arguably do the same thing. And so it's a little bit more of an abstraction thing because it's not like an employee, it's like a task. But to me, it's kind of like it's more likely to be successful when you already have set up all the access on this bot that you're using already versus you need to reset up everything you have to set up to maintain, etc. Yeah, I totally agree. But just to push back on the scheduled tasks thing, I think that's just fundamentally different to having someone who's like basically there, not even nine to five, but 24-7. Like a lot of these kind of like virtual assistant-y type tasks. So, you know, like remind me before this podcast that we need to do something for it or, you know, like monitor something overnight and see if anything comes in. Yeah, but it's all scheduled tasks behind the scenes, right? Like it's exactly the same thing. I understand. But there's like the always-on aspect of this. And I get that you can do that with your computer and stuff. Just have a desktop computer and run everything that way. It's not quite the same. You know, classic example. Like, oh, well, I need to take my laptop when I go on a business trip. For that one example, yes. It's offline for a bit. So it's like all of these tools, like all AI tools, like they kind of do the same thing. They're kind of the same thing, yeah. It's a chat with tools. But the perception of, I think like it being always on, I think that's material. So do you think that perception is worth the hassle of maintaining two setups? Yes. Okay. And I think like, I'm still waiting for you to set up my Hermes agent. It's set up. You just need to talk to it. We need to do it. All right. Yeah. Yeah. So there's one last bit, but I'm really excited to just start building these little like quality of life routines into it. And I think that, yeah, I think this will be a big deal. What's really cool with these agents is they have really, a really cool concept of memory. Although I would say tragedy is really catching up and I'll show you some new memory features a little bit later that are really cool. But like the way they approach memory and overall the way they approach like being a distinct entity rather than just a bot that does what you ask it to do. It makes the interaction fundamentally different basically. And that's the interesting part. And so, yeah, but like Hermes is a very viable alternative to Grogbot. And I would say my solution of like how an actual coding bot setting it up for you to me is the best bang for your buck at this point in terms of flexibility and in terms of also value. I do understand for people who can't be bothered though, that Grogbot is appealing. Is this going to be like a PC versus Mac thing long term? Like people prefer one setup or the other? Every provider will have some kind of version of this is doing quite well right now. So I would not be surprised. OpenAI has a version of that for kind of business accounts. Didn't they buy OpenClaw though? They did, yeah. More or less the same thing. But the thing is OpenClaw is going to stay open source and it's not going to be baked into Grogbot, et cetera. Like they have it 80% built in business accounts already. Like you can create agents that live in the cloud and you can connect them to tools and things like that. So I would not be surprised if they started rolling this out following the success of Grogbot. And especially because it's XAI and it's Elon Musk versus Salman, they hate each other, right? It's like they're suing each other right now, et cetera. So obviously any kind of market share one can gain, the other one's going to do everything they can to kind of go after them. So yeah, it's like, I would not be surprised if you get a cloud version of this, a chat GPT version of this within the next few months as well. Yeah. I mean, for all the kind of like bitterness of the competition there, it's good for the consumer when these companies are kind of trying to match each other. Cause I assume that there'll be like a price war on that to an extent. And I really see Hermes becoming kind of like the WordPress of agents. Like in the sense that WordPress is also an open source project that allows you to kind of build custom plugins and do all of that, et cetera. And Hermes is starting to be that. It's like, it's doing a good job at kind of keeping up with the trends and reacting. And it's going to become kind of like this de facto harness if you don't want to commit to any side and you want to be able to connect all of them, which has a ton of value, especially as we don't really know the outcome of what's going to happen, which companies are going to take the crown in this market. It's one that is worth paying attention to. Like I was kind of, if you look at the podcast from January, I was against open claw, like very much, et cetera. And the reason why is because it was difficult to set up and doing the access properly was a pain in the ass, et cetera. But now that there were genuine security risks as well, there still is right. If you set it up wrongly, it's still dangerous, but first of all, the models are a lot smarter. They're a lot less likely to do something dumb. And the second is now your coding agent can do all the technical side of things, including setting up like custom API. Like I use codecs to like go in the API settings and block a bunch of stuff and so on and sync granularly for every single API key. And it's all documented in local documentation, et cetera, as a way to make sure that it just gets where it's needs and it cannot do too much damage if it goes wrong, basically. This and also obviously prompt injection research has progressed a lot, basically. So overall it's going in the right direction. And I think that now it's something that's worth paying attention to, but not until you've kind of like mastered your code code or your codex, et cetera. I think it's the thing that you do after that. All right, let's move on then now. And I want to talk about a new feature in ChatGPT, computer history. Yeah, that one is. This has riled a few people up and it opened a whole privacy can of worms, really. I totally don't. There may be an interesting payoff. So just to explain what this is, if you update the ChatGPT app on your Mac, there's a feature which you can opt into and turn on if you want called computer history. And this will record basically everything you do on your computer. So like clicks, typing, shortcuts, app switches, accessibility context. Apparently it does not capture screenshots, audio or private browsing activity. I know there might be one or two people worried about that, but it basically stores everything for two days locally. And it does pass a lot of this info to the ChatGPT, the OpenAI servers. And then it kind of like creates summaries or memory files based on this. And importantly, it doesn't promise that it's going to like filter out your banking passwords or any kind of like really sensitive data like that. So yeah, there's a trade-off here where you're exposing yourself potentially to like a lot of surveillance. But there's an interesting payoff. Do you want to tell us what that is? An interesting payoff. And that's why I scroll down. I might blur some of this on the final podcast, by the way. Just I want to be sure there's nothing compromising. But probably I'll show you this one because there's nothing very interesting here. But the point is that you can see that it will sometimes suggest you to create skills. And so it's super interesting because you can literally just use your computer, do your normal work. It will act like almost like a consultant who would observe you working and be like, hey, I think you can automate this. And you could create a process here where you could do that. And you can click on this button. It will launch a new chat and create a skill. And essentially, it will try to replicate the task that you've done. And you can give it context, right? If you click, like let's say if I click here, you see it just writes a new prompt, basically. And I could add to it. I could write, I could dictate and explain what the task is and that kind of stuff. And it's really powerful because not only can you create a skill, but then you can schedule it, right? You can put it like say, let's say it's like, you know, updating a report or something like that, whatever. And then you're like, okay, well, every Monday at 9am, just run that skill that I just run. And it's like kind of a seamless process where you just do your work. You go and read that report and you're like, yeah, actually, I can automate this. You click the button, it makes the prompt. You give some extra data, it makes the skill. And then you're like, do it every Monday at 9am. And you never have to do this task again. Obviously, you probably need to review it a few times in reality. But it's interesting as a setting, basically. And it's like, the trade-off in privacy is insane. It's like, I don't know if I'll keep it on. It's like, I have to try it, right? It's like, it's kind of my job. You see my Hermes project, by the way. That's how I manage Hermes. But yeah, the trade-off is insane. But I think it's also for a lot of people who, you know, we teach this AI stuff to people. And I can see people struggle to get their minds around how these things work. Most of the time, they're unable to identify what could be automated in their day-to-day. And this is going to be incredible at pinpointing. You just turn it on. You're like, hey, you go do your work and let's check in one week. And then you just kind of press this button and just create all these skills for them. And to me, like, I'm going to recommend some people do that. I won't necessarily recommend that they keep it on. But I'll be like, okay, just know you're being recorded for this one week. Don't do anything stupid on your computer that you wouldn't want anyone to see. And then let's use this as a way to improve your productivity. And you can pause it or turn it off temporarily. We need to log into the bank or whatever. You can add some apps and sites that you want to remove tracking, etc. If you want to. Like, there's some granular control of it. But I think you're right. Like, doing that just for a week, most people will discover like, holy shit, I can automate all of this. Yeah. A lot of people recommend time tracking yourself for a week. Like, every 15 minutes and see what you're doing. And then look at the end of the week and say, oh, I spent so much time doing this. Can I hire someone to do this? I mean, this is all advice. But now the AI can automate probably a lot of what you're doing already. It does make sense, I think, for probably, arguably most people to turn this on for a week and try it out. And imagine you're like a manager in a company and you've been struggling to get AI adoption, for example. That's like a no-brainer. It's like, hey guys, we're turning this on for a week. And then we're reviewing the logs one by one in a one-on-one meeting together and see what we could automate. Oh, God. I mean, think about it. This is going to be like a big flashpoint for corporations. Like, if they can turn this on for all their employees. Exactly. It's not even just to like get AI adoption. Can you imagine? It's like 5,000 employees. All right, we'll turn it on. And then just let it run for a month. And then AI just automates 50% of the work. So they lay off half their staff or something like that. Yeah. Like, that's the reality of, wow. Yeah. Yeah. That's what it's for, right? Think about it. Let's say you're a manager and you have performance targets. It's like, it's a no-brainer. Let's be honest. It's one of these things, like, if you don't do it, every other company is. And they're going to be ahead of you because they can be more efficient. Like, yeah. Yeah. So, yeah, a lot of employees are going to hate it. But in reality, it's going to bring a whole lot of efficiency. And just these kind of tools are really powerful because I can see the main obstacle to implementing these things is often people. It's not like the tech is much further than the way people use it for 95% of people. Maybe not you if you're listening, but for 95% of people. And this kind of is a very powerful bridge of, like, kind of, like, pushing things forward very fast by automating the process of observing and translating that. Now, I don't know which models they use to do that. But when you actually open individual memories, you'll see that it actually has quite a lot. And it's surprisingly smart. So very, very, very controversial feature. But I think if I am an OpenAI sales rep, I would promote that to the executive team and they would love it. And that would be a very strong switch from Entropic to OpenAI thing, just because of how easy it is to do. Like, I'm sure Entropic could do it, but it doesn't exist right now. You know, the next play is like, well, OpenAI suddenly, like, know what everybody in your company does. So they can also just, like, do, like, when their models get good enough that they can just start rolling out companies, especially software companies or something. So when you pay API rates, like, which businesses do, like, people who are not on individual plans, they do not, they're not allowed to train on your data. Like, they actually released this week a new thing on how they cannot train on your data and they have full privacy. And so I would expect that the company, like, if it was provable in any way, they would be in big trouble because that's what they're selling to people. Agreed. But, like, who knows? Who knows what's actually going on behind that? Is there some, like, loopholes? Oh, we didn't train, but, like, you know, we did something else and then we trained off of something else or, you know... But actually, that's one of the things. It's like, OpenAI, they're struggling incorporate. That's why they're behind. They're behind Entropic on sales right now. There were some numbers on revenue and, like, they are actually, like, officially behind, basically. And you can see that it's actually a problem people have with Fable, particularly. Fable has data retention because Entropic is so afraid that you're going to hack the government, basically. Is it? And so now they released a new system that you see, like, they have a zero data retention for frontier models, and that's one way for them to get banks, for example, to use them because banks cannot use Fable right now because of data retention. Same for a lot of medical stuff, for example. Like, it's not allowed. And so they're really, really pushing for that. They're even, right now, discounting Sol by 50% on, like, OpenRouter, Vercel, and all these kind of resellers. So you can buy Sol for almost the price of Solnet right now, except Sol is competing with Fable almost. So they're very eager to actually show their hand and give a lot away to these corporate customers because they are behind Entropic. So I'm just saying, like, I'm sure they would love to know what every employee in the world does 24-7. And it feels like they could use that in some very, very nefarious ways. I think what they will do is the normal price will not train, and then they'll offer you a discount in exchange for training on your data. And then budget-conscious companies will give away their data, and then they will be able to do it, and something like that. Like, that's what they do is the bargaining chip, basically. I could see that. They have a high sticker price, then the sales rep goes to the company. He's like, oh, you know what? I could give you a 30% discount if you let us train on your data. And then, like, the managers look at their AI budget, and they're like, sure. And then they sell themselves away, you know? So this is going to happen. This I can see. And it's not advertised, and you only see it when you talk to the sales rep, et cetera. I'm sure. So anyway, that's computer history. That's mega controversial. I don't know if I'll keep it on. It's my job to test this thing, so I'm testing it. But it's an interesting one. But I'm excited to see if it will actually... I cannot report on it, like, in two weeks, and I'll see if it actually helped me in any way. It's also very powerful in the sense that it builds memory, right? Which means, again, you know, I talked about how Hermes has good memory. You can say, hey, I was working on this video the other day. Like, what did we say? Like, what did they say I would do? Like, oh, there was this one page I checked. I can't remember. What was it? I will know, basically. Which is kind of cool, too. Yeah. Powerful stuff. So I want to talk a little bit about the codex limits being slashed as well. So that's been a thing lately. But just before we do, I want to let everybody know something about AI Accelerator, which is our course membership on AI and automation for business owners. So we've actually just released a new ChatGPT means business course. And there's more lessons coming out today. There's a lot more coming out. For this, a lot more, in fact. So if you've been using ChatGPT codex or if you switched over from Claude recently, or, you know, if you've never used any of these tools that we're talking about and you run a business and you think, oh, I need to get onto this, this is the place to start. Because we take you through the full setup process, show you everything it can do for your business. If you've never touched anything like this before, we'll teach you how to set it up. But also how to actually use it in your business to do stuff. This is not like theoretical. Oh, let's go book up. There's like case studies as well. Like if you look at the business trick, for example, like it's like there's actually files you can download and then you can run the same prompts as me and so on. And it's like, it's always kind of like business related stuff, marketing related stuff. So it's like we're not teaching you how to make cute kitten pictures or stuff like that. It's more like real stuff, basically. And just to update folks on some more things that we've added to this recently. So we have developed, we're calling it WorkOS. It's basically like an operating system that lives inside Codex or ChatGPT. And it's kind of like a second brain, but for a business user. And so it's going to kind of like remember everything you do in your business. It's going to interview you at the start. It's going to organize all of your data, all of your work into different areas and projects. And it's automatically going to learn kind of what you're doing and how you like to do things. So yeah. So like when you come back to use AI again and again, again, it kind of, it learns what you like and how you do things. So it gets better over time. So that's been really, really popular with a lot of our members. Plus we've also just released the authority hacker MCP, which means that your AI, so your codex, your Claude code, your Grokbot even could connect. Yeah. Hermes could connect into our membership and into all of the knowledge, the community posts, the technical detail about like how we build our skills and processes. And you can just query it and you know, your agent will be able to interpret that and use it in a way that's useful for you. So for example, you could say, Hey, how do authority hacker do LinkedIn carousels? And then it will give you the information about how we've done that and say, you know, would you like me to build this for you? And then your Claude code can just implement the thing that we've structured and organized and you're going to have it up and running in minutes. So really, really powerful thing there. Also, we have expanded our range of skills. So skills are like SOPs. Plugins now. Plugins now. A difference between a plugin and a skill. Plugins like a collection of skills, but you know, some code in there, some other bits and pieces. It's like a skill on steroids. Is that right? It's just more advanced collection of skills. Basically. So that enables us to do really full end-to-end processes. So you developed this outreach, AI outreach team skill. Do you want to tell us a little bit about that quickly? Well, this one is interesting because it's like a lot of people want to do call outreach, but it's kind of difficult to set up. And there's a, you know, it's kind of like a bit technical in terms of you need to do good availability. And then there's also kind of an art to writing good emails, et cetera. And then also finding the right people and not being a spammer. So like these three things are quite difficult to do and kind of like handles all three. So it will actually connect to a tool called instantly and just handle everything for you. Set up the campaigns, warm up the emails, do all of that, et cetera, for you. Then it will actually use a bunch of scrapers and combine them together to try to find your ideal ICP. So for example, I think in this one, I show how I find potential companies that would hire the recruiting company that we're working with. And the way it does it is it finds stale job offers on the LinkedIn offer. And then it finds people's profiles with another scraper. And then it actually with another scraper, it has a scraper that goes from LinkedIn profile to email, for example. And it's changed these three. It finds like people who have job offers that are out for like 20 plus days, not filled, et cetera. And then emails them because based on the rules that they have. It tests them first as well. Yeah. So it makes a tiny list. It tests it. It makes sure that this actually works. And if it works, it's like, hey, we can scale this building system. Let me build it for you. It does the whole thing. It reaches the profile, gets the email, makes sure it's valid. And it prepares the campaign. It uploads everything into Instantly that handles all the sending and, you know, the dripping, not going too fast, et cetera. And it will also manage the campaign for you. So let's say you started your campaign. It didn't do perfect. Like sometimes, you know, maybe the list wasn't built properly. And there's some issues in terms of the liability where they will kind of adjust based on that. Or maybe people don't respond to the template. So it's going to rewrite it, et cetera. So it kind of manages it. So you can set up a recurring task like every week or something. And it will wake up and manage the campaign for you and try to re-optimize it as it goes. So that's why we said like it's your outreach team because it literally does everything. We actually have a member who started using it and already got his first client. And I released this like a week ago. SEO agency, I think. Yeah. Yeah. And so like, yeah, it's kind of cool. Yeah. And like, you know, this is just like the tip of the iceberg really for what you can do when you start chaining different skills together and plugins like this. But we've actually already built dozens of them in here and you can get over the air updates to all of them through our plugin library. So then you can get them to basically stream them directly into your cloud code or codex now. So yeah, this is what an AI membership looks like in 2026. It's not just some courses and some prompts. It's much, much more than that. And you can get all of this for $129 per month. And that's important because in 10 days. So on the 1st of September, we're actually increasing the price to $149 per month. So we're giving everyone fair warning. You're a week and a half left. If you want to join AI Accelerator and lock in this $129 per month price for as long as you stay a member, then I suggest you go to authorityhacker.com forward slash AI Accelerator and you can join one of the plans there today. So yeah, thanks everyone. And we hope to see some of you guys inside there. We've already got almost a thousand authority hacker listeners and subscribers in there. So you'll be in good company with other business owners as well. All right. So let's talk now about codex limits. What's being slashed? Because, you know, we're always hearing about, you know, like next week they're coming down, but then there's an extension and then they reset everything. That's, that's entropic. That's entropic. You know, they have this plus 50% and then they extended it twice already. But there's basically like kind of like a Reddit thread that's actually going very popular right now where people are basically, they compare like the $20 plan and basically they said that they used to get $160 in API value per week. And that is dropped to $80 according to their tracker. And Thibaut, who is basically the guy who resets the codex limits at OpenAI and the investigators basically posted today that they investigated and they didn't find anything wrong, basically. And people are outraged, et cetera. I do feel the limits are lower for my usage. Like I easily burn more than 20% in a day on a $200 plan if I want to. It's still usable. Like I, when I burn 20%, I'm running four or five threads most of the day. So it's not exactly. You're a pretty intense user of this. So yeah, I would say you're, it's probably not typical of like a business owner or knowledge worker. No, most people probably, most people can make one account work for sure. And I still make it work. But I'll tell you a story that happened to me this morning that really pissed me off. So every month I kind of reconsider my AI usage. So I kind of like lower the plan on which I'm going to renew. And then I kind of remake the decision. Like, am I happy with this tool when it renews? Right. And so for example, on Codex, I renewed this morning from the $200 plan to the $100 plan. Because I expect I will still have some usage, but I wanted to make that decision. And so before the renewal, I had 77% left on my weekly limits. They don't have five hour limits on Codex right now. They only have weekly limits, which is a blessing and a curse in a way, because you can burn your whole usage and then be out for multiple days. But then what I did is it renewed to $100 and my usage went from 77% left to 7% left. So it basically assumed. And so I just paid, right? I just paid like $100. And then I basically could send a few more prompts. And right now, if you actually check my Codex, I'm going to have to re-upgrade because I literally have 1% left after literally just a little bit preparing this podcast and doing a few things. The reason for that is because you're doing the shenanigans where you downgrade and then you upgrade so that you can get a reset. You know what really is annoying? I don't think that's the intention. No, no, no, no. There's one thing that I can defend. There's one thing that I can defend. And what I can defend is the fact that they happily pushed back my reset time by seven full days while at the same time essentially cutting me down to that usage. So essentially, they've robbed me of some tokens because if the limit had like the limit was 25th of August. Which back as in like brought it forward or like put it back and forth. No, brought it further away. So it was at 25th of August and now it's at 27th. So they've pushed it as if my quota had resetted, except it hasn't. And therefore, like in some aspects, they essentially stole me some tokens. If I'm already... It's not crazy, but I'm just saying it goes into the feeling. It seems to be some kind of bug there. But I don't think people who downgrade and then upgrade every month are like, you know, a priority for people. They should... Right. Like you're gaming the system a little. I agree. But all I'm saying is that OpenAI is getting a little bit stingy with tokens. They used to give them a lot away and they had to do these results all the time. And you can see that now it's getting kind of like a little bit less. It's interesting in that strategy because they're giving less in the subscriptions and then they're giving more to API buyers. So they're going after corporate. Like, as I said, they have this 50% off, et cetera. It makes sense. I agree. As a business, it makes sense. I'm not mad. I'm just telling you what happened. I'm just reporting here. What happened? And it's like, yeah, that's what's happening. They're basically like, OK, we're doing well with subs. I think most people on sub... A lot of people are switching from Cloud2Codex right now. But I think on the corporate side, like especially because corporate, you know, makes like six months contracts and 12 months contracts, et cetera, it's lagging. And they're really kind of... I think they've decided to reallocate a bunch of the stuff they gave away from subscriptions to API and try to kind of push that, basically. So that's what's happening. I'm just reporting a lot of people on subscriptions are now whining. I still think it's decent on a $200 plan. I still get a lot done. And I struggle to kill it. So I'm sure most people would struggle to kill it. So yeah. And I think if you go on X these days, like people talking about this stuff, they get very invested and very emotional about like, oh, my limits feels like it's down 10% or something like that. But if you just zoom out for a sec and you like go back... You get a lot. Two years, three years, four, like before ChatGPT was around, you think like, hey, 200 bucks a month, you get a tool that can do this. Like everybody would buy it. You know, like it's such an incredible thing. And I'll buy more subscriptions if I want more tokens and I still say it's good value, basically. And especially compared to API price. I mean, people are whining that they get $80 per week for spending $20 per month. It's like, it's still a good value, basically. Yeah. All right. So let's move on. Let's talk about this new mysterious model that seems to be going around right now. Is this a fable killer? It's apparently being called, is it 0xAlpha or OXAlpha? OxAlpha. Yeah. OxAlpha. Yeah. It's like a model that's been released on a bunch of, you know, providers we can buy API. It's free. So I think they have like 1 trillion token per day to give away for free or something right now. But this model is kind of like people are going crazy on it. So it might run out throughout the day. But you can use it for free and you get 1 million context. And allow like, for example, these guys like a developer and run a bunch of developer tasks against fable. This is GLM53, which is a Chinese model, GPT-56Soul, and Grok 4.6, which is really good, as I said. And essentially that's the pass rate, right? So fable passed at 65%. GPT-Soul passed at 52%. Grok passed at 62%. And this passed at 80%, for example. Again, that's like a sample size test, but it's not too crazy. And then on knowledge walk, like same, this guy uses it for editing videos. A lot of people are starting to edit videos with LLMs. He uses fable law in general to do that. And he basically said, it's not as good as fable, but it's the closest thing. Basically, it's like opus five fable-ish level. And you can literally run this for free right now. And nobody knows who makes this model. That's kind of the beauty of these mystery drops. Labs, usually they are a marketing stunt. So like if a small lab has a very good model, if they release as a small lab, people are like, ah, it's a small lab. I'm not going to pay attention to it. If there's a new stealth model and nobody knows who made it, it almost becomes kind of a game in the industry on like trying to decide who made this and trying to figure it out. And overall, there's a lot of usage and a lot of PR. And then when the model comes out, the lab says, hey, it was us. And so that's what's happening now. And so probably a smaller lab has made this model and is going to get a lot more exposure based on that. So if you actually are set up on open router, we talked about open source models with DeepSeq for writing last week. Well, you can actually just call this model for free right now. And for a week, it's going to be free. So have fun. If you just create a free open router account, you can use this completely free right now. Unmeted. Yeah. That's pretty good. So if you have any big batch tasks to do or something like that, it could be a good use case for that. Yeah. Yeah. You can either have like your codex or code code call it for like a simple task or you can plug it into an external harness, like open code or something like there are like free harnesses that you can run. I could run the main model. So you could almost code for free or whatever if you wanted. Very cool. Okay. Let's move on to our final story today, which is the new LinkedIn AI slot reporting button. So just, just so we're clear here, LinkedIn, a company which for the past year has done everything it can to put AI comment and AI post generating features inside its own platform is now seemingly a bit surprised that people have been using them to create posts and comments with AI and has now added a button. So if you go on any LinkedIn posts, there's little three dots at the top, right? You click on that and there's a little option now that says seems like AI slot. I'm not going to click it on Eric's view, but yeah. I mean, first of all, I get why they've done it to an extent because LinkedIn is just full of slot these days. Oh, LinkedIn is the worst. This and Pinterest, like LinkedIn, Pinterest is just all AI slot. Instagram is starting to get there too. But, you know, at the same time, I don't feel sorry for them because I feel like they caused it on themselves by, you know, pushing all these features so much. But what do we know then about how they're using this information? Is this essentially a dislike button or is it going to be aggregated? Do you know? No, I mean, they won't tell you, right? But it's definitely a negative signal. It's just next to I'm not interested. So I guess I'm not interested is more like, don't show this to me. It doesn't mean slow quality. It just means that this is not what I want to see. And it's more like, like if a lot of people report you, you probably will lose reach eventually and they will use that against you. I think it's interesting because it reminds me of like, it's like, you know, this thing where companies kind of go with the trends and follow what's happening. So it's like right now in America, a lot of people are against AI a lot, mostly because data centers, they don't want them in their communities. There's this wide belief that it uses drinking water as well, which for the most cases is not really true, but people don't like that. Energy costs are going up and the American grid is not very good. And therefore it's competing for that. So overall, there's just, you know, there's a lot of these graduation speeches where people are like against AI and stuff. And I think that's kind of what's happening. And it's like, you know, there's always kind of like a corporate reaction to these kind of things. Like, and that's what this is. Basically, this is like, oh, look, we're listening to you. We care about you, et cetera. To me, it's fake because I know that Microsoft owns LinkedIn. Microsoft sells AI to like Azure is a massive AI provider. Yeah. They resell entropic models. They resell open AI models, et cetera. So yeah. They must have seen some kind of like, you know, user statistics or like reduced engagement or something like that, like platform wide. And they've identified this as an issue. So they need to, first of all, collect data on what people perceive as being AI sloth or not. And I think it's important that they're not specifically going after anything that's AI content because it's AI content. Then it's, they're going after AI content that is bad. So if you're using it to create something good, it shouldn't really be an issue. People are not going to feel bad about it. Yeah. Yeah. So for now, I think it's more of a sign to a wink to their users. Like, Hey, we're listening to you and we're going to do something about it, but I don't think they've decided what they're going to do about it. It's going to be interesting. Like a lot of people like on the Vergecast, for example, they're like, Oh, we think there would be an AI content only feed, for example. And that eventually the goal of the platforms is to make you like it and consume it so that they can then produce the content themselves and completely bypass the creators. Like basically like they become just a media machine. And that, yeah, it's like, it's something that could happen eventually. The problem is it's very difficult to separate what is AI generated or AI system. Well, less and less with watermarking, right? That's the goal, right? It's like everything's getting watermarked and then you can bypass it. And maybe we'll talk about bypassing watermarks in a podcast, but like you can, I'm definitely making, we're working on our SEO plugin actually for the membership. And that's one of the things I'm building in is like bypass watermarks. So make sure you subscribe. So you don't know if that episode we cover it. The goal is that all these providers eventually align on some kind of watermarking. And then it's very easy to tell this was AI generated. This was not AI generated and potentially let the user choose. And then they work very, very hard to try to make the AI feed actually not shit. And then eventually they can start eating away at creator revenue. In this case, on LinkedIn, they don't pay creators, but like you could imagine YouTube, you could imagine TikTok, you could imagine any, even X, right? They pay the creators quite a lot and then just eat away at all the advertising revenue, basically. There's a whole industry going to spawn up around like for passing off AI content as human content in this era though. You can see. There's going to be big bucks in this for... You are not sure last week, but see, like you've talked to some people and you can see people hate the watermarked idea. All the marketers want to kind of pass their content as human created, even if it's super obvious. And so, yeah, that's going to be a thing. Any final words of wisdom this week before we wrap up? I would say going back to growth bot, I think if you're really non-technical and you want some kind of Hermes type bot, et cetera, in my opinion, it's a bit of, as you say, it's a bit of a poisoned gift in the sense that I think it's great right now. And there's tons of reasons to use it, but the lock-in is so strong that I'm a bit afraid a lot of people are going to invest themselves a lot in this platform and eventually want to use something else. And it will not be easy. And you'll be using something subpar or overpriced. So it's like, it's not that I don't like the platform. It's more like I don't like the sales conditions. And I would recommend to try the setup I mentioned earlier, like try to have your codex or your cloud set up a Hermes on a VPS for you. So you don't have to do any of the tech and do that. So maybe I'll talk about that at some point. All right. So thanks everybody for watching this week. I would also encourage you head over to authorityhacker.com forward slash AI dash accelerator to check out the AI accelerator program before the price goes up on the 31st of August. So everything that we talked about on this episode, you can learn more about what's inside the program there. So thanks again for listening to this episode of the Authority Hacker podcast. Please do us a favor and leave us a review, hopefully a five-star one on whichever podcasting app you are listening to this on. That helps us out so much. You can hardly even understand. And it's basically the main way that podcasts get distribution is the more reviews you get, the more distribution we get on these apps. So that would really help us out if you can do us a solid there. We appreciate you. And we'll see you this time next week for another episode.