# Grok Bot unlocked my business

Channel: Brian Casel
Video: https://www.youtube.com/watch?v=RM8ZwhoJdeg
Duration: 22 min
Language: English
Words: 4264
Transcript page: https://viewrankai.com/tools/youtube-transcript/RM8ZwhoJdeg

---

[0:00] When I first set up Grokbot, I had the same question that a lot of people have. How is this going to be useful for me? Now, my team of bots run about half of all the work across my portfolio of products, and I almost never even chat with them, but they are working for me. So, two unlocks helped me figure out exactly how to make Grokbot useful in my small product business, and today I'll show you both of those and how they work in my setup. And most people treat their bot like an assistant, you know, something comes up, you open a chat, you tell it what to do, and then you check

[0:32] its work, and then you tell it the next thing, and you do that all day, and then you become the bot's full-time manager. So, the unlock is to treat your bots like employees. So, when you hire someone, like a real person, you don't just hand them one task at a time, you give them a job, and they're supposed to show up and do that job day in, day out, week in, week out without you prompting and without you needing to look over their shoulder every 2 minutes. Now, when I talk about the work that my bots are doing, I tend to use the term background work, and it's not like their

[1:04] jobs are not important. I mean, they're working on marketing and growth experiments for my products, and that is very important to work in the in my business, but I think of their work as background work because I hired the bots to do that sort of in the background. I design the processes and I I make the key calls, but the bots are doing the legwork. And so, that's about half of what drives my business. The other half is the work that I do directly on my products and on my operations. That's my work. So, I'm not running my own work through the bots. I'm mostly in Claude

[1:39] all day. I'm working on code bases. I'm working on agent skills that then power the jobs that my bots work on. All right, so I'm looking at one of my GitHub repos that I call HQ operations, and this is essentially where I go to work on my business, right? So, whenever I'm working on the operations across my entire portfolio, this is where I'm going. And I keep a lot of different stuff in here like, you know, we collect reports in here. We have some growth strategy stuff, some, you know, general like docs and SOPs, but the agent skills folder is where I maintain and I work on

[2:13] the collection of skills that I ultimately install into my bots. And so, this is that folder and here and I actually have some retired skills. That happens quite a bit. I'm constantly like shifting new skills in and out, but I I'm going to go into the skills folder and these are all of the active agent skills that I have built. And this is this list is constantly growing. These are all private, internal, custom-built

[2:40] skills that I really carefully maintain. And then I'll show you in a minute how I install these into into my my Grok bots, which, you know, operate in here in in Grok bot. But over here, you know, lately I've been working on these growth skills. So, I have my bots running like weekly growth experiments across all of my products. Sometimes they do market research, sometimes they're running growth experiments, they're doing SEO audits, we're doing like general strategy reviews, but then I have like other stuff like operations keeping like giving me like KPI metrics reports on on different stuff. We've got like quarterly bookkeeping in here. A whole bunch of

[3:22] stuff. And so, each one of these is essentially like a process, right? So, if I go in and I look at the skill, you know, I'm not really going to go into each one in depth because they are quite quite advanced. You know, I like to break them out into multiple steps. This one is for running actual growth experiments. I'm going to sort of back out here and let's look at market research. So, this one has a skill and and it's basically just instructing, you know, the bots on how I want market

[3:51] research to be done. Um and again, like a lot of stuff kind of goes goes into that. And you know, the other thing that I have my bots do is I have them open up Claude sessions. So, um you know, when they're working it uh in here, this is where like where they orchestrate tasks and they receive different tasks and and stuff from the the skills, but then they use a skill that I call Claude management and they use that to then actually open up uh my Claude account and they and and these things down here are these are actual sessions that my bots are running. Um

[4:25] and sometimes this like fills up with lots of stuff on a on a really busy day. Uh but yeah, they So, I actually have my Grok bots do their coding work in Claude so that I can use Claude models and and and not like use up the Grok bot usage. But getting back to what I wanted to show you here is I have these skills. These essentially are jobs or these are parts of jobs that I give to my bots. And then what they do is they have routines. Routines in Grok bot are like recurring scheduled things that just always happen without me needing to prompt and remind them to,

[5:01] "Hey, do that thing, right?" So, over here on the right side are the list of routines. This is for my chief bot. Um and if I look at any of the bots that manage my individual products, like here's my Builder Methods bot, you know, that one too has a bunch of scheduled routines. Now, these routines are really really simple. Like the actual descriptions of these, you know, it's weird. They they got rid of the ability to actually directly edit the descriptions. You have to actually ask the bot. So, I'm going to I'm going to ask BM bot, "Let's do the

[5:35] daily growth run experiments." Tell me what the description is for your daily growth run experiments routine. Okay, so the daily growth run experiments routine runs every day at 7:00 a.m. It's whole prompt is just this. It's just that one line. Log this run in our bot run log and then run the growth run experiment skill end-to-end. So, I do have a system in place for my bots to log every every one of these and it's sort of like a record for Chief to be able to check in on who's doing what and

[6:10] what the current status of things is. And so, the second sentence here is just run the growth run experiment skill end-to-end. That's all we need. And I'm not putting any other instructions whatsoever in the actual description for the routine because all of those instructions live inside of that skill and we maintain that skill very carefully in in here, right? So, this is where I want the source of truth for all of the instructions. I don't want to have some of the instructions living in here and some of them living in here and then they sort of start to conflict with each other over time. So, that's that's

[6:47] really important, right? So, every one of these routines is just saying, "Hey, run this skill." Right? So, then the question is how do they actually get those skills installed into Grokbot? Well, we have another skill for that and another routine for that. So, if I look at Chief here over here, nightly agent skills sync. Basically, that is just an instruction and and a skill, sync agent skills. So, that's this one. Very simple one. It's a it's a skill and I think that they run a script, a Python script, which basically just tells the agent to open up this GitHub repo, pull down all of these skills and install them in the location

[7:28] on Grokbot's computer where they store and read agent skills. I think I also have them pull down our docs. So, these are sort of like just like standing like policy docs about how our operations work so that, you know, this is sort of like defines how my whole company works and how our operations work. I want my team of bots to be aware of this stuff as well. It's almost like a company handbook, if you will. Um so all this stuff also gets synced in and

[7:55] stored on a location on their computer. Now, that does mean that those files are living in two places. They live in the GitHub repo and they live on Grok Bot's computer here. And uh you know, you can open up Grok Bot's computer and you can go like this. You know, so we've got like agent data and stuff. So you can sort start to like dig into here and and try to figure out like where Grok Bot stores its own files. And you know, I've spent some time like asking uh Grok Bot exactly how it organizes things in here and I I've come to learn like, you know

[8:27] what? I I better sort of be a little bit more hands-off and and let it manage that how it's going to manage it. I don't want to get too in the weeds. Mainly what I'm concerned with is having Chief go through and get the latest version of all the skills and all the docs. And every single night it's just syncing those to make sure that they're up to date. And and that sync involves uh pulling down the latest updates to all the the skills, retiring any skills that we've, you know, moved to to the retired folder, and deleting any skills that have uh or like adding any new skills that that

[9:01] we've that we've created. And so a lot of my work day-to-day is actually in Claude uh working on those skills. So like here is a big session that I've been having with Claude where I'm working on our growth skills in that repo. And we're sort of walking through like how these skills uh sort of like send me, you know, emails. Um and how the the bots send me emails. And I'm going to show you that side of things in the second part of this video. So like day-to-day I'm coming I'm coming in here and I'm working on my business. I'm working on my operations by tinkering

[9:35] with the actual growth skills or any of the agent skills, I'm having them push those skills to this repo into the agent skills folder. And so, I'm always making making changes to this. You can see, literally today I was making some changes, 2 days ago I was making some changes, 3 days ago. Um so, you know, these things are constantly in in flux and I'm always tinkering and and and really working on my business, making things run the way that I want them to run and I'm always adding new systems that can add value the business. That's That's what working on a business is, right? So, that's sort

[10:10] of what it looks like now when I'm working on a company of one with a team that I scale out with, you know, with with Grok bots and and you know, that's how I think of these bots, more like team members who take on these jobs, if you will. Oh, by the way, if we haven't met yet, I'm Brian Castle and if you want to write skills like these for your own bots, I made a free workshop called from AI dabbler to AI native. And in there, I show you exactly exactly how I design agent skills that you can hand to any agent or any bot. So, you can watch

[10:40] that by going to buildermethods.com/workshop. All right, if you're running a business and you're looking to a platform like Grok bot and trying to become a power user with it, that first unlock that I talked about is to think of this as a team of employees with jobs and not so much as a personal assistant. Now,

[10:59] here's the second unlock. Grok bot, Muse, dots, Open Claw, all these platforms, they can all handle just about any tasks that you throw at them. The limit is you. So, your bots can only be as productive as you can be at managing them. So, if your bots are burying you in chats, in approval requests, and things that you need to review, and super long paragraphs of text that you have to parse through just to find the one thing that matters, then your own productivity vanishes. You You if you have any management experience with people, then some of these things sound familiar, right? I mean, it's eerily similar when managing a team of

[11:37] AI agents. So, as I already showed, the agent skills take care of the instructions going out to the bots. But, I put the same level of design and effort into what comes back to me, you know, the approvals, the status reports, the performance reviews, the things that require my decisions. You know, I need to craft how that stuff comes to me so that I can manage it all quickly, easily, without friction, so that I'm never the bottleneck, right? So, I also need to manage my own mental energy with this stuff and the overhead with this stuff. So, the unlock for me was good

[12:13] old email. Now, once you start to use a platform like Grokbot a lot every day, and once you start to create multiple bots like I have, like I have a chief of staff, and then I have a different bot that runs each one of my different products uh that I'm overseeing, and each one of these has their own slate of growth experiments that we're running and product updates and new features that we're pushing out. So, you know, there's a lot going on across my my portfolio here, and uh and so, what I found when using, you know, Grokbot uh the app, which I I love the app and I love how it

[12:45] works. I love the simplicity. I love all the personality that they put into uh the bots and all the tiny interactions. It's all great. I I really really think it's well done, and I find that the app keeps getting better and better. They they keep adding like tiny improvements. But, I did find that like the whole chat interface doesn't work for me from a management standpoint, right? Like, I can go in here and I I can like sort of troubleshoot and ask quick questions if and when I need to, but I don't really do that all that all that often. Like,

[13:15] what does work for me is my email inbox. Believe it or not, I am an email person, and you know, it's not really chaos in there. And what I want is my bots to actually send me emails for things that need my approval or it when they need to send reports and things. When I have all that stuff sort of flying around in chats, I like to have an email inbox that I can work through methodically whenever I need to. So, what I decided to do was give each one of my Grok bots their own email address. To do that, I'm

[13:47] using a service called Agent Mail. They're not sponsoring this video or anything. I just found this tool to be really helpful. I I think that there's a a few different competitors to this. You know, all all are, you know, seem perfectly fine. But, Agent Mail has worked pretty well for me. So, basically, I give each individual Grok bot their own unique email address. I set up you know, like a domain email address on on my domain for each one of these. And then, you like I can literally email to them, they can email to me, and we can go back and forth. And so, for example, here is one email that

[14:21] I received, you know, 2 days ago for my product fly rate. came from my FW bot. FW bot is here. And so, this particular email are some growth tasks that it was working on. Six of them need me, meaning like six items need my approval, and they came in, and they're just sort of like showing you know, what these things are, and I just needed to sort of like quickly glance through this. And then, all I needed to do was reply with something

[14:49] simple, like one, approve. Two, approve. Three, I have a question about that. Four, yeah, that's done. Five, yeah, just keep it simple, blah blah blah. Six, done. Okay. So, you know, and then it just comes right back to me. Checkmark, done. And just one more thing, this one still needs your open thing, and you know, then I come back. Yes, approve. You know, so like, you see how productive I am with this? Like, it's just we're just firing off emails back and forth. Now, it might seem a little chaotic when you look at this because, you know, what's going on with all these different emails flowing

[15:18] into my inbox. And and yeah, there there are days when I have multiple emails coming in from multiple bots. But I spend a lot of time iterating on exactly how those emails come to me. So, um if I come into Claude, here is uh here is that Claude session where I've been working through this with with Claude. So, this is where I come back to working on my business, working on the operations. And yeah, there were days where I was like, "These emails are too much or there's too much information in these emails. I I'm having a hard time processing it all." And so, I went

[15:53] through and I worked with Claude to improve that. And that and and like we we worked through that problem. Like, these emails are too hard for me to process. Let's try to find a solution to that. And we sort of worked through, "Okay, we we need we we have the bot sending me four different kinds of emails. We need to have each email have a very particular job. We need to establish some conventions when it comes to email subject lines. And here's the life cycle of a task." And so that I can easily process them. And then what Claude ended up doing as we worked through this is it

[16:25] actually mocked up like what an email inbox might look like. This is all just a mock-up. This is not like real information. Uh we sort of like mocked up what these emails would look like when they come to me. So, here is what my inbox would look like with the conventional subject lines and the um like the names of each product, what type of email it is. And by the way, we we've grouped the types of emails into one is like a KPI pulse email where it looks like this. Um this comes from

[16:55] Chief. This is for the whole portfolio. These are like fake numbers. This is just a mock-up here. But like for every one of the products, I get some key metrics. And this is just an informational email. Tells me which experiments are currently running and which programs are currently running. And uh we kind of just go through the portfolio. It gives me all the key numbers. Great. But the other types of emails that I receive are experiment emails like this one where okay like this month for builder methods

[17:23] we've just finished this experiment. This is going to be our next bet and these are like a B and C option and all I need to do is reply with yeah let's go with bet A or bet B and simple as that. And so I receive an email like that once a month one for every product and that way we are continuously running growth experiments and all I need to do is receive that email reply yes or no or I might give a little bit of feedback and then my bots are off and running. They're going to fire up Claude sessions. They're going to you know do all the things. They're

[17:55] going to push stuff into my websites and and then my multiple products can continue to have marketing experiments running all the time. The other types of emails that we designed in here let's see tasks like I just showed you like you know two things needed for me really clear right in the subject line. This is for Flywire right in the middle of doing some tasks. Two things just need your quick approval and then it just goes boom boom and and and then I can you know go through and either approve and you know it does give me the link and that links me directly into the Claude

[18:29] conversation where that work is happening. So I can go in and steer the conversation if I need to but yeah it's all here in the email and again it just literally tells me how to reply to the email. Number one looks good change one you know it gives it gives me these easy [snorts] action. I often do them just right from the iPhone right on regular

[18:49] email you know. The last type of email that I get is a strategy review. We do this on a quarterly basis. So we're sort of doing like an overarching view of all the different experiments and we're also doing some market research and and like kind of proposing or like the bots are pitching new ideas to me. I receive them in in a nice email like this and and then I can go through and we can approve and we are constantly improving and iterating on our actual strategies. I mean, this is what it looks like for a real marketing team to be working on products and pitching ideas to the head

[19:25] of marketing or to the CEO and the CEO signing off. Like that's the kind of dynamic that I'm trying to design and craft for my company of one. So this is literally how I can run multiple products, we're doing multiple things, and I'm taking a very careful approach to how I'm processing this information in an orderly way. Now, one more thing on the on the email stuff, you might be wondering like why don't I just give GrokBot's access to my direct Gmail? I mean, you you've heard people doing that like, you know, giving Muse access to your email or GrokBot access to your

[19:59] email. Again, that gets back to more of the personal assistant kind of thing. So if if I were to have a personal assistant to help me manage my email, then I might consider doing that, but I'm not really interested in that, frankly. I I just want my bots to be their own autonomous employees, if you will. And again, if we come back to like what would a real human employee need? They would need a computer, GrokBot gives them that. They would need an email address, and I gave them that. And so we go back and forth, and they do their work, they send it to me, I I approve. I can manage my email

[20:33] inbox just fine, thank you. So, um yeah, that that's the dynamic that sort of works for me when I'm actually running a business and I want to use GrokBot's um as a way to help me scale that business. That that to me is more of like a power user approach than thinking of it as like a personal assistant. So we know that these bot platforms, they're just going to keep changing and churning, right? Today it's GrokBot, Muse is blowing up, that's from Meta, Dots just came out from OpenAI. We've heard about Instinct, you know, so next month there's going to be something

[21:05] else. But the jobs that you design and the skills that run them and the way that they report to you, all of that is yours and they can move with you to whatever comes next. So, if you want to learn how to start writing those skills and really crafting them, my free workshop helps you go from AI dabbler to AI native and that's how I show you all of my favorite design patterns when it comes to, you know, writing agent skills. So, you can watch that right now by going to

[21:33] buildermethods.com/workshop. And subscribe to the channel so you don't miss my next video on being productive with AI. Let's keep building.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt