Build & Sell Grok Bots (2 Hour Course) transcript

Grockbot is one of the most powerful AI tools that I've ever played with because it gives you an army of AI agents right in your pocket and you don't have to be technical at all to use it and to get value out of it. I've been playing around with Grockbot since it came out and it has actually changed the way that I'm working rather than just being like a fun AI tool, a bunch of agents that's just good for like a demo or for social media.

So, in this video, I'm going to break down everything that you need to know about Grockbot. I'm going to show you the way I'm using it. I'm going to do a full setup. I'm going to talk about the mindset and how you actually connect this stuff so it becomes useful. I'm going to be spinning up a completely new account and building out a team of agents right here in front of you guys. So, you can basically just follow along.

And I'm going to be showing you how you can use it to obviously provide value to you and to grow your business, but also how this is a really good offer to go into other businesses with and help them get set up with their own Grockbots. And real quick, if you guys don't know me, my name is Nate. I've been building AI automations and agents for about 2 years now. I've got a global community of AI builders that's over 440,000 members.

And my business in the past 12 months has done just under $5 million of topline revenue. And it's because everyone's able to move quicker because of using systems like these. And that's across things like education, communities, certifications, events, newsletters, and more. So anyways, by the end of this course, you'll be completely set up with what you need to know about Grockbot and scaling your business and other businesses.

So let's not waste any time and just get straight into this one. Okay, so Grockbot, it's AI teammates that you can give real work to and they can sign in, they can use your tools, and they can take action for you. You can see that you're going to get a nice interface that's sort of like a Telegram style thing, so it feels very familiar and you're able to talk to them like teammates. They can talk to each other. They can delegate work to each other.

They can do things while you sleep. It's super super cool. And just to start off with like a 30,000 foot view and to see where we're going in this course, I'm obviously going to sign out of this, spin up a new account, and set up everything with you guys so you're not confused at all. But this is kind of what this could look like by the end of this course for you. So, right here, you can see that I have a group chat.

This is actually a channel similar to like you have Slack channels or ClickUp channels or whatever you're used to. And you can see that on the left I basically have DMs with all of my different agents whether they're general or their content or the ones that I've pinned up here which is kind of like my seuite. My executive assistant I call them chief. I've got my COO, my CCO which I call a chief content officer and then my CFO.

So anyways, what we're looking at right here is I have these four agents in a group chat called leadership. And what I said is based on how this past week has gone and our long-term goals, I want you guys to ideulate. I want you guys to help me figure out what should we be doing in the fourth quarter? How do we continue to differentiate? and how do we continue to reduce keyman risk as we keep scaling up the business?

So, you can see that they come together, Becky, Dan, Chandler, and Klouse all come together and talk about the different areas that they're involved in, and they work together like a leadership team to help me figure out what we need to be doing. They're tagging each other, they're talking to each other, they're reading each other's answers. And then right here, they finally came down to five key decisions that they agreed on, which was no new business units in Q4, spend only to get Nate off the critical paths, second phase test off on the main channel, Becky owns comment mind pick, so Nate isn't the idea gate, and CIC is the 2027 unit that scales without Nate on camera.

So, obviously, there's more here to dig into, but that is what they were able to come up with in just a few minutes based on their areas of expertise in my business. And the reason why I have these four main agents pinned up here is because what's going to happen is as you start to use Grockbot, you're going to spin up a ton of different agents. You can see that I mean I only have 10 right now in here, which isn't actually that many, but I have each agent having one very specific purpose.

This one just looks at X. This one just creates animations. This one just talks about my content strategy. This one just checks Fireflies. All of these have a very, very specific thing that they do. And that's going to get overwhelming quick unless you have kind of like your own leadership team up top that delegates work around. Because you can see if I click into Klouse, not only can I talk to Klouse, but he is able to talk to everyone else.

He's able to delegate work around to all of these bots. You can see that Claus can send messages to Dan, Becky, Chandler, Dev, Eyes, Views, Minor, Motion, Money, Dash, Cal, Coffee. All of the agents can talk to each other. And that's when it gets really cool. As long as you're making sure you're scaling it up in a way where the agents don't get confused about who's doing what and how to talk to each other and stuff like that.

You don't want to have them duplicating work or leaving holes in your plans, things like that. And I know this interface may seem a little bit intimidating right now, but trust me, it's not going to be bad at all. Just to show you some other stuff that's possible in each agent, you basically are going to have a name, which is what you see over here. You'll have a label, which is optional if you want to give them like a role, like a job title.

And then you have a description, and that is actually what tells other agents what does this agent do. So, for example, for Klouse, this description is you are Nate's chief of staff. Klouse is the only bot that Nate talks to. Before doing any task, check whether another Grockbot owns it and delegate first. Only do the work yourself if no specialist fits and then bring the result back here. You can have notifications turned on for specific bots.

And then also what you can do is that you can see each bot has their own computer. Now they they have their own screens but they do share the same computer. So if you authenticate into like school.com or if you log into something then other bots will be able to access that too but they have their own screen. So right here you can see it was literally in its own CLI authenticating into GitHub. You can see that it was looking at my Herk 2 repo.

You can see that it was looking at our school community and it's logged in as me as you can see. So, it can help me take action inside of AIS or AIS Plus and things like that. You can see it was also searching for Chicago Cubs food. I'm not sure what it was doing, but the point is I can control the browser and so can Claus and so can the other agents. You can also see there's something called routines. This one for Claus is called a Sunday work log archive.

It runs every Sunday night. And what this does is it takes everything that we did this week in ClickUp because I have every project that gets logged or created inside of my Grockbot be stored in ClickUp so I can view it later and comment on it and stuff like that. It basically takes anything in here that we marked as complete for the week and then it pushes them into a different one called the archive. So I can actually see what we've done and make sure that things on our actual main work log are staying clean.

Now all of the agents can contribute to this and push things around, but Klouse is kind of the one I would say Claus and Dan are both the two that are in charge of keeping it cleaned and keeping things like moving forward. But they could also be like every 5 minutes, every 10 minutes. They could also be certain other types of triggers where if I go to here, you can see it could be like on a Slack message or on a Teams message.

There's not a ton right now, but I know that they're working on adding a ton more triggers like off of emails or calendar events and things like that as well. So, anyways, just wanted to give you a super quick interface tour, uh, a super quick 30,000 foot view of where we're going. And now what we're going to do is we're going to actually dive into getting this set up if you've never opened it up before. All right, Grockpot, let's do it.

So, this is kind of the dream, right? People are looking at this one main AI agent that just controls a swarm of agents. And this is cool and it's possible, but the problem with this is that it gets a little bit confusing because this agent becomes overwhelmed because it has so many different agents to pick through and things as you scale can just get a little bit messy. So, what we want to try to set up is more of this sort of style where we have you that just talks to a few executives and we'll scale these up slowly as you realize you need them.

And then all of these executives are basically aware of, okay, these are the three agents that I talk to the most. These are the three agents that I talk to the most. And these are kind of more of like your operator agents that do something very specifically within the overall vertical. So if this was more of like your, you know, let's just say the marketing team, then these are little marketing operation agents. If this is the finance team, then these are like accountants and like just things, just little agents that do one very specific task.

So that's something that I'm trying to communicate through is you don't want to just have one mega agent that does everything. You want to have a bunch of specialized agents that do one thing very very well and then they just get called on whenever it's time to do that thing by like their manager essentially. And what's really cool is that even though these are operators, they can still talk to each other. So if they realize, oh shoot, like in order to run this weekly report, I actually need to get data from um this team over here real quick.

And so it'll just reach out over here, get that data, come back and keep working. And so that's why things like having a description for each agent be very very clear of what this person actually does. Now, real quick, what I wanted to talk about is the difference between something like Hermes or Grockbot or Codeex or Cloud Code because if you guys have been using my channel or using if you guys have been following my channel for a while, then you know that I really like to use Claude Code and Codeex.

But I really see these in two different buckets. I kind of see these two tools being in the same bucket and then I kind of see these two tools being in the same bucket as well. And what I mean by that is when I'm sitting down here at my desk and I'm doing sort of like my my daily knowledge work and I'm building things, I'm always going to use cloud code or codecs. You know, Grockbot is basically being used by the XAI, the Grock models.

And these are being driven by GBT models, cloud models. And I think overall I like these models better for, you know, production work and building things. Um, obviously you can switch out the models, but the point being, these are the harnesses that I like to use when I'm sitting down at my desk and driving work and projects. These are the agents that I like to use because I can have them on my phone and because I can have them on the go and because I just really like spinning up automations with them.

So, I think of these as like, you know, sitting down and you're doing work at your computer. And I think of these as building automations and agents that you can talk to while you're at lunch or when you're on a vacation or when you want to set up something that just kind of runs overnight while you're sleeping or runs in the background based on some sort of event trigger. So if I wanted to build my own little team to run like an Instagram account fully autonomously, I would probably use a Grockbot.

But if I wanted to build myself some skills and a production pipeline to generate a bunch of, you know, Instagram reels or carousels at scale, I would probably just build that within something like this. So anyways, hopefully that paints a little bit more color here. Don't get too overwhelmed about that. I will kind of talk about while we're building out stuff in Grockbot what I think are good use cases and what might not be, but just wanted to call that out in case you guys were thinking about that.

And I know that there's a lot of AI tools going around. So I don't want to add more like overwhelm or confusion to your head. So we talked about this. What I want to talk about now is my framework for how we actually do this effectively. And I call this an AI operating system. And I call this my four C's of building an effective AI operating system. So the four C's are context, connections, capabilities, and cadence.

I think of these two as sort of like the information about this Grock agent or whatever you're building that makes it specific to you. And then I think about these as the things that help the agent move faster. So context is like giving it information about you, giving it information about your business, giving it information about your goals and what's important to you and how you work. I then think of connections as giving the agent access to the things that you need to pull data from.

So obviously connecting it to your email, to your calendar, to your Slack, to your, you know, your bank accounts if you want to, connecting it to things so it can read what's going on in real time in your business besides just giving it like static data, but also allowing it to take action. So creating drafts if you want, you know, creating ClickUp tasks if you want, doing all of those sorts of things, creating Google Docs, you know, generating slides, generating assets.

That's actually a big part of the power is connecting it to different sources. But if you don't give it the context first, it's not going to understand what do I do with these connections? Why are these important? And how do I actually make deliverables for you that are good and like valuable for your business? And then we talk about capabilities. This is basically combining the previous two things. So now we're going to give it capabilities, which a lot of times we're going to be calling skills, um, to actually give it instructions on how to do something well.

So, if you want it to be able to create you good proposals, you need to give it the context to make those proposals and the connections to actually do that in like a Google doc or something. If you want it to be able to scrape LinkedIn and help you write LinkedIn posts, you have to give it all of this context and connections to do so. So, capabilities is where you really start to give it basically like more tools and more skills.

And then from there, we have cadence and those are the routines I talked about. These are building the automations that help this thing scale besides the fact of you just like saying, "Hey, can you do this? Can you do this?" making this thing more autonomous and intelligent so it knows, oh boom, this happened in Nate's business. I need to do this or hey, this agent reached out to me to do this. I better check in here and I better do this so that when Nate wakes up, he's good to go.

And once you kind of like keep having this loop of creating more of all four of these four C's, then you get in a really cool spot with your crockbots. Okay, so the rough agenda of today was this. We kind of already took a little bit of a tour and as we get set up, I'm still going to be pointing things out and calling things out and then we're going to go through those four C's. I'm going to show you what it looks like to start adding context.

I'm going to show you what it looks like to start adding connections and capabilities and cadence. And then we're going to pretty much end off once we've got a bunch of different agents running. And I'm just going to go over some hacks and some things that I like to do that really help you get more out of building your Grock bots. So, that's kind of the rough agenda of today. There's going to be timestamps down below if you want to kind of skip around a little bit or save parts and come back to them later, but let's just go ahead and get set up with a brand new Grockbot.

Okay, so what you want to do is go to Google and just type Grockbot and it's going to be probably the first link. AI teammates that finished the work. You're going to click here. This is the page that I showed you guys earlier. And what you're going to do is just go ahead and download this for your operating system. And then also go ahead and install this on the iOS app if you want to be able to use them together. I'll show that off later in a bit.

But when you're here, you're going to go to pricing. And basically, you can get access to Grockbot for just 30 bucks a month. So it used to be when it first came out, it was just the 200 bucks a month, but now it's 30 bucks a month. So, you're just going to go ahead and upgrade your plan to Super Grock. And then whatever email you did that with, that's the one you're going to sign in with when you open up the desktop app.

So, I'm just going to go ahead and upgrade to Super Grock right now. And I'm going to create a new account for the 30 bucks a month plan. So, you can see how far does 30 bucks a month actually get you on your, you know, usage. So, I'm going to go ahead and sign up for this. Okay. So, I just signed up for the 30 bucks a month subscription. We are now good to go inside of this account and I've downloaded the app. So, what I'm going to do is go back into the app.

I'm going to come in here and I'm going to log out and then we're going to log back into this fresh account. So, sign in and then sign in with the email that you just used obviously to create that account. I'm authorizing Grockbot to read my profile, read my email address, do things like that, and then we're going to go ahead and continue through the data sharing so that we now get into Grockbot. And I'm going to click sign in.

And then we should be able to go right back into our desktop app and get set up. So, it's setting up Grockbot's computer. And this is all basically going on in Gro's cloud. Meaning it automatically is going to sync to your phone. You can turn off your computer, you can close out of the desktop app, and it's still going to be able to work. So that's why I like it so much is because with other things like Hermes or setting up like Cloud Code on a VPS and all this stuff to have that like always on feel, even something like OpenClaw, it's not that hard, but it did feel a bit more technical because you kind of had to set up like a VPS or you had to have a manage app.

But this is basically all taken care of for you right away. So meet Grockbot. We're going to go ahead and click next. It says that Grockbot has its own computer and works just like you. Cool. Give each bot a job. So, this is what I talked about. You don't want to build one mega agent. You want to have each agent doing something very, very specific for your business that provides value. Um, what do you use every day? So, if you want to click through here and connect some stuff, feel free.

I definitely use Google Workspace. Definitely want to connect something like, you know, GitHub, LinkedIn, ClickUp. You can connect whatever you need to. But hey, if you forget about it here and you go ahead and click next, don't worry because you can connect things easily anytime. And I'll show you guys of that, of course. But now we have to create our first bot. So this one, I want you to just think of this one as your personal assistant, your executive assistant, the one that you're going to talk to the most.

So give it whatever kind of name you want. For now, I'm honestly just going to call this one friend because he's going to be my best friend. And we're going to go ahead and just make this one a blue square. So let's get started with friend. It's going to spin up that first agent. And what's going to happen is before I even say anything, it's going to start talking to us. It's going to say, "Hey, my name's friend. What can I help you with?" Like, "What's important to you?

What is my job?" That's what it's going to start doing at first. And this is where we really start to get into that first C of the context. We're going to start giving this thing our information for who we are. Hey Nate, good to meet you. What do you actually want me in your day for? As you can see, so what I'm going to do is I'm just going to start giving it some information. Hey friend, my name is Nate. I am not technical at all.

I don't really have a ton of experience with setting up Grockbots and things like that. So, when you're explaining things to me and when we're setting things up, I just need you to explain it to me concise and an easy way. But, I run a home services company. It's called Summit Home Services and I want you to help me because I have a small team, but I want to be able to do more and scale my revenue. And so, just start off with something very similar like that or very simple like that.

Messed up Grockbots here, but friend should be just fine. Now, while this is responding, you can see if I go in here and I go to my settings, we have a name. It doesn't yet have a label, but you can see that based on the thing that I filled out earlier, it says the user works with Workspace, GitHub, LinkedIn, ClickUp everyday. Start with these tools when suggesting connectors or taking on work. This is where we could change the description and add a label.

Anyways, now it says you're running Summit with a small team and you want to grow without drowning. That's a thing that I can actually help with. Where's the pinch right now? Now, here's what I'm going to do. I'm just going to say, give me one second. I'm just going to give you a document that has a bunch of information about my business already. And then we can talk more about like bottlenecks and stuff like that. And so why I'm doing this is because I want to show you the first connector.

And if you already have information like this in your business, which a lot of you guys probably do, give it to Grockbot. So this is just basically an internal operations and knowledge base. You can see we have our company snapshot. We can have hours of operations, services we offer, pricing. And so this is just going to be easier than me explaining to Grockbot everything about my business. So, what I'm going to do is I'm just going to go ahead and copy this link.

I'm going to go back into Grockbot and I'm just going to give it this link and say, "Hey, go ahead and read this doc and just, you know, save this to your actual memory so that you know this about me and my business." And what's going to happen here is this thing currently has no arms at all. This is basically just an LLM that we can talk to. So, it's going to try to read this and it's going to realize, wait a minute, I can't.

And then I'm going to show you how we actually connect it to something like Google Workspace so that it can actually read through this type of stuff. and then we'll continue going on about figuring out other agents to make and things like that. The first thing it says is that the link is locked because if you can see here, this basically says, hey, you know, only people with the link can access to it. So, it should be able to open it up and it's going to try to do this in its actual computer.

So, if we go in here, you can see that it's going to paste in that link and it's going to go to that account, but now it's going to try to sign in and it's not going to be able to access that doc. So, there's going to be a much easier way to actually give it access to Google Docs, Gmail, Calendar, all that. So it comes back and it says, "Hey, the easiest fix is to come back and connect Google Drive so I can read it." So sign in with the same Google account that has the doc and then tell me.

So I'm just going to click on this connector and it says add from plugins. So I click on that and all I have to do is click on add. And then what makes it super easy here is that all we have to do is authenticate. And one quick tip before you start authenticating a bunch of stuff is to label them because if you want to authenticate into a bunch of different Google accounts, you want to be able to know which one's which.

So, I'm just going to come in here and I'm just going to call this one Nate Herk8 because that's the one that I'm authenticating. And then it changes the name. And then I click authenticate. And it just pulls up the Google window for you to actually sign in. So there's not a bunch of having to deal with all these little API keys and technical things. You're just signing in like you would on your own computer. So I'm going to go ahead and sign into my account here.

I'm going to select what Grock can access. So right here it says, do you want to let it see and download or do you want to let it see, edit, create, and delete? And I'm fine with both of those. If you don't feel comfortable with letting it edit and delete, then maybe you want to be careful about like what you give it access to. But in this case, I wanted it to do all that. So now I have authenticated my Google Drive and I will say go ahead and try again now that that is authenticated.

And now friend is going to go ahead and keep working pulling up the doc. It's going to read it and then hopefully what we're looking for is that it's actually saving this to memory so that it knows later if we are talking about other things about our business, it actually can answer that kind of stuff. And you can see it's actually telling you what it's doing. It's connecting to our Google Drive, the Nate Herk8, and then it's going to kind of, you know, reason and think and take action and come back to us.

And that is kind of called an agentic loop within these agent harnesses. You can see now it's working. So, it has this loop of it looks at something, it thinks, takes action, looks, thinks, takes action, and that's how it decides to stop or decides to respond to you, things like that. So, anyways, it saved it. Summit is HVAC in the Twin Cities, small team. Fix it, not upsell. The money leak is slow email replies and busy season.

That's the job you actually want me for. And if you want me on that, then tap Gmail and tell me and I'll look at the inbox. Okay, so here's where I want to pause a little bit. We want to get set up with some email stuff, right? I put a bunch of fake emails inside of this account. This was our receipt from Groai. But anyways, I put a bunch of emails in here about our business, right? Fake emails about from customers, you know, like, hey, this charge or hey, I want an appointment. all this kind of stuff and it's hard for me to manage that.

But what we want to do is we might want to have one individual agent that's job is just to help us like triage emails or respond to them and create drafts and things like that. So what I'm going to do is I think you're right. I think it would be good for me to connect you to Gmail so you can help me out with like responding and triaging. But because I'm treating you as like my chief of staff, I think that we should create a separate agent and its job is basically only helping out with these emails.

And while that's happening, I'm just going to go into my friend, my settings, and I'm just going to call this one chief of staff. And what I'm going to do is also just going to go ahead and pin this agent because this is the one I want to talk to the most. I don't want to get overwhelmed with all of these different agents that it is going to create for us. Now, look at this. It actually said, I'll make an inbox agent whose only job is to summit emails, and I'll get the Gmail hooked up for it.

So now, right from here, it messaged inbox. We can click on this message and see what it said. This is the message from friend to inbox agent and it says, "Hey, Nate created you as the dedicated email agent. Here's information about the business. Here's information about the pricing. Here's how you triage. And here's what you do." It also gave us the source of truth, which is the actual Google Drive. And then it said, "Okay, cool.

Connecting Nate to Gmail now." And then if I actually click into inbox agent, it says, "Hey, I got the brief from friend. I'm going to save this and I won't do anything until you tell me to." So that's already a demonstration of how these agents start to work together. Anyways, let's go ahead and authorize our um Gmail now. And it's the exact same process as we did earlier for connecting our accounts. We click authorize.

We click the email we want to connect to. We allow these different permissions. And then we hit continue. And now when we go back into Grockbot, that should basically be added. Now, what I want to do real quick is give that a label, right? We talked about that earlier. All you have to do is you're going to go to plugins. You're going to see that we have two installed. And right here we have Google Drive. We have Gmail.

I'd click into this one. And I would just change the name right here by calling this Nate Herk 88. Making sure that name has been changed. And while you're in the plug-in section, you can see how many other ones there are. There's calendar, granola, AWS agents, there's plane, there's ampify, there's so many different things in here. And you basically can search for them, or you can click on these different categories.

And so I'll talk more about these other plugins later, but that's basically how you start building in the connections like we talked about to give your Grockbots access to actually do things. So that's where it is. It's called plugins. Anyways, let's see what's going on here. So, friend has a message for us. It says, "Playbook is saved. Gmail's connected. Want me to start triaging the inbox? I'll pull emergencies and warm leads first, and I still won't send anything until you say so." So, it's okay for us right now when we're setting up inbox, when we're setting up these sort of automations and things that we want it to do, it's okay to talk to inbox.

The point I was trying to make is at some point you just want to make sure that if you need to delegate workaround or if you need for um you know automations to go through, you should make sure that friend understands that it knows where other agents live and what they're good at. And the way you do that is by having um that knowledge set up. So if I go to inbox and I click on the settings, you can see that it says here already your only job is to summit house services customer email.

This is what you do and friend can read this to make sure it knows what inbox should be doing. I'm also going to give this one a quick label, which is basically our um email agent. Cool. Now, what I'm going to do is I'm inside of inbox already. So, we're going to just help him get set up with the what we want to do in the inbox. So, the inbox is pretty messy. I think that there's probably a few different tasks and automations that we're going to set up for you.

But first, what I want you to do is just read the emails that we have in there and make some suggestions to me. I'm basically the only person at the company that responds to company emails. So, help me figure out what would be the highest value. We probably need to have them labeled. we probably need different automations set up. What do you think makes sense to help me run this company more efficiently? And it's going to go through that agentic loop of understanding our message, looking inside the the inbox, reading things, and then kind of giving us some recommendations.

All right, so inbox came back with quite a bit of information here. I actually don't really know if I like the name inbox, but we'll keep it for now because it's pretty clear. But one of my favorite quotes that I really want you guys to internalize is that you can outsource the thinking, but you cannot outsource the understanding. And what I mean by that in this context is that you can see I said, "Hey, look at my inbox and help me figure out what makes the most sense." Now, what that means is inbox went out, read everything, and now it's giving me some recommendations.

But that doesn't mean I just blindly say, "Yeah, do all of that." Because one of the big mistakes I see with people getting set up with agents and automations is they just build so much right away and then they lose track of what's actually going on and then it starts to just not even provide value because what they were building might not have actually solved the problem. And for this example scenario, we know that managing the inbox is an actual bottleneck in the business and it is a problem we want to solve, but we want to solve it, right?

So, anyways, got that out of the way. Now, let me look at what this response is. So, it read through the whole inbox. Two of these need you today. So, Marcus has no heat. You know, he's got a newborn. Here's his phone number. Hannah has a water heater leaking in the basement. Those have both been unread for a couple hours this morning. The inbox also has 5200 emails. It's unread. So, I'm just going to ignore that because it's just this is just an old email.

Um, so the highest value isn't cleaning those old emails, but it's making sure that the next emergency never sits and that you only see the mail that needs you. I agree. So, it wants to create six labels on customer emails. Emergency, needs you, quote, schedule, billing, and warranty. Vendors and SEO pitches get ignored. Skip the old labels already in there, which I made for different demos. Anyways, three new habits that it wants to create.

So, it wants to have a routine that watches new mail and basically just like pings you if there's anything really immediate like emergencies or refunds and angry customers. Everything else waits for a short weekday morning list. And then what it wants to do is draft the easy replies. So, things that it can answer based on the documentation I gave it. A lot of this is super easy kind of just like customer support FAQ answers that the bot could use to answer with.

And then it also said the rest of the 30 is quotes, reschedules, warranties, and a tech application. Those can wait behind the two emergency and the money fights. So where should it start? So in this case, I think that all of these recommendations are good. I want it to draft emails for me. I want it to ping me with important things and I want everything to be labeled as well. So, I'm just going to say do all of it because one decision that we might have to make in the future.

Probably not yet because I think that at the moment inbox can handle all of this. But let's say we were getting a super high volume of um refund sort of requests. Maybe at some point we would want to have inbox agent triage all of those to an agent that specifically deals with refund requests and we can train that agent really well on here's what you do in X case, here's what you do in Y case. And then maybe we even give that refund agent the ability once we really trust it to actually process the refund.

So I'm not trying to overwhelm. I'm just saying like thinking multiple steps ahead. That's where we could go at some point is each agent takes care of one of those labels. And so that's why I think it's important to have like this hierarchy where we mainly have to communicate to a few and they can delegate work to a bunch of agents below them. So you can see it's setting up those six labels now and then it's going to draft the two emergency replies and turn on a watch.

Still not going to send anything. We're keeping everything as drafts for now, which I think is still a good way to start. And then when we fully trust it, we can maybe start to scale up those capabilities. So, first thing I want to call out is it created a routine. It created a routine called inbox emergency watch. It says you are inbox. You are Summit House Services customer email agent. You're going to check the connected Gmail account.

And also, I gave it a label just in case later we add on a bunch of different email accounts for new inbound customer mail since the last run. You're going to look up the Gmail tools. You're going to blah blah blah. You're never going to send. You're just going to create drafts. Here's what you do on each run. So, this is basically if you've ever created some sort of SOP for a specific process in your business that you're going to give to a human.

You're basically just onboarding and training how a human would do it, but you're just telling an agent how to do it. It's just very specific step-by-step instructions as you can see. And what you see here is that it says when to run every 30 minutes. So, every single 30 minutes, every day of the month between 12 a.m. and 11 p.m., so pretty much the entire day, this is going to check. And if there's no emails that fit into this specific category, then it's just going to do nothing.

And it might even say, "Hey, I just did a check. No actions." But if there is something urgent, then it will go ahead and actually run that. But now, let's take a look at what it said. It created the routine, but then it created drafts for these two. So, if I actually go into my inbox, you can see that we still have all of these emails, which are right here. But if I go to drafts, we have two more right now. And the agent left these as unread so that we can still see them.

It also, as you see, it just tagged them. And if I go back to my inbox, you can see all of these are now labeled with quote, emergency, schedule, needs you. This one even has two billing and needs you, needs you, and quote, emergency. And so, anyways, what we did is inside of our drafts, it only took in the two that were labeled as emergencies, and it went ahead and created drafts for me. So, let's click into one of them, right?

This one said urgent, no water, or sorry, no heat. The house is freezing. It says, you know, please help. We woke up this morning. I'll pay the fee. Here's my number. And then the response is, "Hey Marcus, sorry you're dealing with this, especially with a newborn in the house. This is an emergency. We'll treat that right away. Someone from our team will call you shortly at blah blah blah. Sunday emergency visits have $150 emergency fee on top of the diagnostic.

Please reply with the service address so we can get a tech out to you." Cool. So, good response, right? Um I also can see in this one we have a response as well that highlights what um Hannah said at this point. Now, one thing that you might want to think of is if you have some sort of way that you as a business like to communicate or if you have some sort of special sign off that you like or you know, like I said, just a way that you like to communicate.

That's where you could come back into Grockbot and say, "Hey, by the way, you should always be responding like this. Here is basically our tone of voice so that everything you draft is already kind of like on brand." But anyways, what I'm going to do is I'm just going to go ahead and say, "Yeah, go ahead and send both because we went in there and we approved them and now it's going to shoot those off." You can also see that it created those labels which was really good.

And what I want you guys to notice is that that happened super quick and it also wasn't hard at all. We connected Gmail. We gave it pretty simple instructions on what to do and then it just went ahead and took action. And now we get into this thing called the feedback loop which I actually call this um riding the bike or the bike method. So what I mean by that is when you first start building agents or you first start building capabilities, you have to think about it like you're teaching a kid to ride a bike.

Meaning, we can't just expect that everything's perfect and that it's going to work on the first try perfectly. These labels look pretty good, but if we really looked into them, we'd probably notice a few things that we don't like. And that's the whole idea of when you use AI for something, obviously, the end goal and the the motivation is that it's turning out to something that's 100% done, 100% accurate. But more realistically, what happens is as you start building, it gets you like maybe this much of the way there.

And then you have to say, "Hey, this is what you did good. this is what you did bad. Try again. And then you get here and you keep giving it feedback over time until you get to the point where you're like this much confident, right? And so why I call that the bike method is because you would never just put your kid on the bike and then say, "Hey, just just pedal. Just pedal. Stay balanced. You'll be good because they're going to fall and it's going to be bad." What you do is you put them on the bike, you hold the handlebar, you have one hand on their back, and you're watching them and guiding them.

And then slowly over time, you build more trust and you give them feedback. You know, hey, you're leaning too far to the right. You have to keep your weight in the middle. try that and let's try again. And then slowly you start to take your hands off and you're still walking next to the the kid and then you know you you put the training wheels on and you're not touching at all. But even when you feel really confident about their ability, you're probably not going to say, "Hey kid, go ride your bike." And then you're going to go inside and take a nap.

You're still going to like sit in the driveway and watch. You're still going to like be in the kitchen and just be near in case you hear them crying or whatever it is. The point I'm trying to make here is this stuff is amazing and it makes you so much more capable, but it's not magic. And people that treat it like magic, and that's why I said earlier, people that come in here, they build 15 agents on day one, they don't test it.

They just say, "Okay, cool. Now I have an agent swarm." Those are the people that run into trouble when clients get messages that shouldn't have gone out. And the agents start inter, you know, uh, conflicting with each other and things die in the middle of the night. That's where you get into these scary situations. Awesome. So now those replies have been sent. Let's talk about some capabilities here. This was obviously a capability and cadence because it already created this automation for us.

And so really what that would look like is I would come into here and I would like maybe grab five of these emails and say, "Okay, so for this email, this actually should have been labeled as quote for this X, Y, and Z reason. And this re this email should have been labeled as, you know, needs you and schedule for this X, Y, and Z reason." And every time it does something and you give it that feedback loop, it literally will say, "Okay, cool.

I've updated the instructions, so next time I can do this better." And you have to keep doing that feedback loop. And that's where you usually start to turn something into a formal skill. So, let's just go ahead and build a skill. Hey, inbox, that looks good. What I want you to do now is we're going to build a skill. And I want the skill to be weekly email report. And the deliverable that I'm looking for out of this skill is a Google sheet.

So, what I want you to do is set up a new routine on Fridays at 5:00 p.m. So, the end of the week. And when you run this routine, I want you to invoke this skill that we're building. And what this does is it looks at all of the emails that you labeled in the past week. So, Monday through Friday. And then what I want you to do is generate me a one-page Google sheet with analytics on um how many requests were in each of the labels.

Maybe look at things like average response time and just give me the analytics on like the types of requests we're getting because this data is going to help us build more automations in the future. You know, kind of tackling things that are the highest priority. So if you have any questions about that process, ask me. But otherwise, let's go ahead and start building out the skill and, you know, give me an example deliverable.

And so one of the things I did there is pretty key. I obviously told it what I wanted in the skill. I told it, you know, what the deliverable is, but I always like to say, you know, like ask me questions about this until you understand. Because if you don't give it a specific enough definition of done, then it's going to give you something that's really bad. And if you just said, hey, you know, ask me questions until you understand what I'm looking for and understand what you should be doing, then it will basically keep interviewing you until it has a lot better understanding.

So this is a pretty simple request. It's not too difficult and so it probably doesn't have a lot of questions for us. But a lot of times when you're planning out like a new automation or you're planning out a new project, it will ask you a lot of questions to make sure you guys are on the same page. And also what you'll notice is when I said I want you to build this new routine, invoke the skill. So skills are basically just instructions.

Skills are, you know, if you it's it's a recipe. So let's say you want to build um not build, you want to make some chocolate chip pancakes. And I'm just going to do a real quick terrible drawing of a person. Um, let's just say they're looking at some sort of recipe. So, this is the the chocolate chip pancakes recipe. Let's say you read the recipe and now you're making these chocolate chip pancakes that are just, you know, absolutely incredible.

They're so delicious. Now, what happens if you want to make more the next week? You probably want to reuse that same recipe because it turned out so well. Or what if these turned out really bad? you would probably update the recipe and say, "Okay, you know, I think that I cooked these each for too long on each side, so next time I'm going to do a minute less on each side." And so this is basically the IP you're building here is the instructions.

Oops. The instructions of this recipe is what you're building. So that every time your agent says, "Okay, I need chocolate chip pancakes. All I have to do is I just have to go over here and read the chocolate chip pancake skill or recipe." And then you're going to basically just turn into a resource library where you have a skill for every single process and every single automation in your business. And then all that you need to do is make sure that your agents, just pretend this is like the Grockbot logo or something.

Every single one of your agents is aware of all of these skills so that all it has to do when it understands, okay, I have this new instruction. I have this new goal that I need to work towards. All I have to do is find the right skill, read it, and then take action. So when I say skill, that's basically what I mean. So switching back into Grockbot, you can see that we have a new routine. So now inbox already has two cadence set up.

It has the every 30 minutes inbox watch and it has the every Friday at 5 email report. It also said here's a sample. So if I click on the weekly email report, I go into here and I get a Google sheet. And how do I have this Google sheet? Well, because if you remember in our plugins, we already connected Google Drive which already had Drive, Docs, Sheets, probably Slides as well. So we already have this deliverable that I could once again open this up on my phone too.

But now you can see we have the week, we have a note, we have the total of emails, we have how many are still waiting, we have average time to response, we have all of these stats, we have a breakdown. Now let's say we actually want to have this be visual and this is really ugly. Well, we come back into Gro and say, "Hey, I think that this is good as far as the data, but I need you to make two changes. What I need you to do is um add some color in here." So add like formal headers and, you know, columns and bold text and use colors. our brand guidelines as far as colors, we really like to use green is like our primary color.

So add some green in there. And then also I want you to add a chart. I wish that you had like a pie chart or maybe even a bar chart as well on the actual breakdown by label just so we have a better way to visualize this data. So what I want you to do is make the changes to that Google sheet and then update the skill so that next time you run this on Friday, we already have these changes baked in. Cool. So, I'm going to shoot that off.

And what I'm going to show you while this is running is what this skill actually looks like. If I open this skill, it pops up right here. It has a name. It has a description. So, use this when producing a weekly labeled email analytics report, blah blah blah. And then it also has instructions. So, this is basically very similar to the instructions that we read earlier in the routines, just step-by-step instructions of what to do.

You know, here is the input, here's what you're building, here are the steps, here's what you should be doing. And what's going to happen is this is what inbox is going to update. So it's going to come back in here later and say, "Oh, you also need to have like a pie chart and a bar chart." And also make sure that the columns are always colored in this way. So this is basically the living recipe that it will continue to update over time.

And if you want to be able to access it rather than just like clicking on it right here, you can go to your plugins, you go to installed, and then you actually just see them right here as private plugins. That's kind of what they're calling them here. But they're very much skills. And the cool thing is if you have skills from cloud code or codeex, you can bring them in here and you can also invoke them with slash commands as you see very similar to the way you do in cloud cod or codex.

Right here you can see that we have this weekly email report skill which is the one that we just built together. Cool. So while this is actually building out, let me just talk about the actual um way that you're paying for this. So if you guys remember, I'm on the 30 bucks a month plan. And if I go to my name, you can see that we have something right here called our usage. And this resets in 7 days. Right now we're 6% of the way through out of 100%.

And this is where if we wanted to change our limit, we could come in here and we could upgrade to 100 bucks a month or we could upgrade to um you know 200 bucks a month or whatever it actually is. But this is where we can see our usage. And so I'll keep checking in on this so you can see out of everything that we're doing today together how long this will probably last you. Once again, while it runs these routines while you're sleeping or when you're on your phone, all of that contributes to your email and your usage right here.

Now, real quick, while this is running, I wanted to show you guys on my phone. You can see it's pulled up right over here. This is our account. I can go into inbox, and you can see that it's completely synced right here. If I said awesome, let me take a look. As soon as I send that, that'll also send on my computer. So, all of this is completely synced. I can click into the skill and look at that on my phone. I can go to the actual Google sheet if I want to over here, although I'm in a different account, so I'd have to switch.

But the point being, all of this is now accessible on my phone as well, and it stays perfectly synced. I could come into here. I could message this agent as well. So, really, that's all it is. It's it syncs in real time instantly. And now you have this in your phone in your pocket as well. Okay. So, I'm going to go ahead and click into the sheet on my actual desktop here. Well, what happened? We don't yet see any changes.

I even do a refresh. Let's see. Did it say I'm checking whether the green headers and charts have actually landed yet? The numbers are in the cheat. The color and charts didn't stick yet, so I'm putting those in now. Okay. I'm not exactly sure what happened, but what's good is I didn't have to give it feedback. It basically proved to me that it verified itself. It opened it up. It realized, oh, you know, I didn't actually fix anything.

So, let me go ahead and fix that now. Aha. Interesting. Okay. So, it said the sheet is still plain. To add the green headers and charts, I need you to sign into Google on my computer once. So, what we need to do now is we're going to take over the computer. This is going to pop up this interface. We just need to sign in with our account and then it will be able to use this computer to do formatting and stuff and it will save this login across the computer that all your other Grock bots can use.

So at this point I'm just going to go ahead and sign into that account real quick. You can see like I said I'm able to drive this as well as Grockbot being able to drive this and then we'll go ahead and get signed in. Okay, so now I've signed in. You can see that we're looking at the weekly report. So I'm just going to go ahead and click I'm done. And now Inbox should be able to get in there and fix that. And you can actually see right here if I open this up, it's mouse is going around and highlighting things and taking action, you know, right here.

If I actually I'll go into the the one in the web. This is me, but this is actually Grockpot right here taking action using the browser. And so there is this function in here called teach a task, which basically means that I could record myself in this browser doing something. So, I could record myself formatting this with different fonts and different colors and highlighting and then I could stop recording and then that would teach Grockbot the skill and then it would watch the recording and it would understand what do I click and where and how do I do it.

So, that's something we could do. I honestly don't like this function. Like I don't use it very much. I'd rather just instruct it with my natural language. But it is a nice built-in feature that's pretty helpful, especially if you do have a couple tasks that you do have to do by opening up a browser and clicking into different places and maybe downloading reports and things like that. It's a nice feature to know about, but I will be honest, I just don't use it very much.

Awesome. So, that took about 7 minutes and now you can see that it has created those elements. It also was taking screenshots to prove to me that they are built and that the colors work and that everything looks good. So, if I open up that actual doc, you can see we now have these green colors. We have everything organized more like an actual table. It's a lot more readable. It's a lot easier to digest. And now we also have this pie chart, emails by label.

And then we also have replied versus waiting. So there's a lot of emails waiting right here. We have some work to do. But now next Friday at the end of the week, we're going to get a report that looks just like this. And if there's anything that we want to add to it, any tweaks, any changes, all we have to do is come back into inbox and say, "Hey, make these changes. Update the skill for me." And I did want to double click on this whole verification loop idea.

It's really important that you are thinking about these agents as humans. And what I mean by that is when a human gives you work, you are typically going to review it before you like, you know, consider it done. At least if you're managing that human. And however you normally verify that the work is done and that it meets sort of like the definition of good, then that's when you would either accept it or you would give feedback.

And the idea I'm trying to make here is most of the time agents can verify their own work before they give it to you. So rather than them giving you V1, why don't you have them give you V4 or V7 that they've already basically said, "Hey, before I give this to Nate, let me review it. If I notice any changes, I'll have to fix it and then I'll review it again." And so every single thing that I delegate to my agents, I always have a verification process in there.

If they're making me a video, I tell them to screenshot all the frames and make sure everything looks good. If I'm having them create me a research report, I'm telling them to double check and triple check every single credible source to make sure that there's no conflicting information and to make sure that the sources are like all relevant within 3 months or something. Whatever you would do as a human to verify the work and consider it good, just tell your agents to do that first.

And it doesn't come back perfect like I said earlier, but now the first thing they're handing you is a lot closer to done than what they would have initially handed you because they would have just kind of quit too early to be honest. So that's what we got. Now, what I want to talk about is one of the things that just happened. So, when we sent off this um email, if I go to my sent and we go to um this one, we basically said, "Hey, someone from our team will call you shortly." Now, here's the thing.

No one actually did that unless I as the founder of this business am going to reach out to my team and say, "Hey, you need to follow up with this person." So, why don't we have a system where we have a Grockbot that can delegate the work out? So, what I'm going to do now is I'm going to go back to friend because friend is my chief of staff. Whenever I'm confused about where to talk or who to help spin up things, I'm just going to go back to friend.

So, I just worked with inbox a little bit to set up a few automations around triaging and managing the inbox, drafting replies, things like that. Now, what just happened was friend or sorry, friend, not friend, inbox helped me draft two emails and I sent them out to two customers that said, "Hey, we'll have a team member follow up with you." Now, what I'm thinking is we probably need to have an automation where when inbox delegates work out like that, we need to have someone actually follow up.

So, what I'm thinking we should do is we use ClickUp for our task management. What I want you to help me figure out is how do we actually take the emails that inbox is looking at and create tasks in ClickUp so that we know that this work is actually getting assigned to people automatically. So, right now, this is just kind of the brainstorming phase. help me figure out what agents we should spin up and then let's actually go ahead and build these automations.

And so what I'm probably going to suggest we do is similar to what I did with my other Grock bots is have some sort of list in here where we can track tasks and assign people work. Now obviously I'm not going to be assigning anyone work because Summit is a fake home services company. But if everyone in your company lives in ClickUp, then you could have your Grockbot actually creating these tasks, assigning them to people, giving them a due date, and saying, "Hey, this is the email that just came in.

I'm I'm assigning this one to you." So let's see what friend says. Don't add a pile of agents. Inbox keeps the email and we add one new teammate called tasks whose only job is to use ClickUp. So when inbox tells a customer someone will follow up. It will ping tasks. Tasks moves the ClickUp card with the name, phone, what they need, and how soon. That's the whole loop. So I want to get to this, but I want to be thoughtful about the way we do this inside of ClickUp.

So what I want to do first is let's connect ClickUp. And I like this idea of having an agent called tasks. Although I don't want to call them tasks. Let's just call him Fred. And what I want to do is I want to have an overall list inside of my ClickUp. There's a space called internal automations. And I want you to create a list in there for all of the Summit Home Services Grockbot work. All of the projects and everything that you're working on and everything that other agents are working on.

We need to have everything being logged in here. But yes, then we will have Fred, this new agent, um, take a request from inbox and then delegate it to the correct team member inside of ClickUp. So that was basically me taking the suggestion from friend. We now have Fred. And honestly, those two names are kind of similar, so maybe I should change that up. But I just added a little bit more color to it to make it more specific to how I actually want to work.

So you can see that friend messaged Fred and inbox, and it's probably saying, "Hey, this is the new teammate. inbox when you're, you know, delegating workout, you should be messaging Fred. But anyways, what I have to do now is authorize ClickUp. So, I'm just going to click authorize. It's going to take me to my Google ClickUp. And all I have to do is connect the account. So, I went ahead and connected the workspace. Go back into Grockbot.

That has been added. And now, what I want to do once again is just make sure I go in here to my plugins and actually name that because you might have more than one click ClickUp account and that's totally fine. And this one is actually just my Nate Herk main. So, I'm just going to call it Nate Herk main. Make sure I'm keeping myself organized here. And now let's see what's going on. So ClickUp is connected. I'm making that list inside of internal automations.

So you can see it doesn't yet currently exist. We just have list Grockbot worklog and Grockbot worklog archive. So we're looking for one that's maybe labeled like summit archive or something or summit log. And now we see that Fred and inbox are starting to talk. They were communicating with friend and they're getting caught up to speed on this whole uh process. Now let's say I'm doing a lot of different work with friend.

What I can do is if we're talking about lots of different things, you can actually reply to a specific action or a specific item in a new conversation. So if I shot this off, it would open up a different section where I could, you know, have two different conversations with friends. So I can basically have like two different work streams to keep things more separate. So it created a list called Summit Home Services Grockbot.

I'm going to go into the ClickUp and see if it's there. Perfect. It's right here. It's called Summit Home Services Grockbot. You can see that we have call Hannah Whitmore and we see call Marcus Duller. It said that I signed both to you. This ClickUp is your space and you're the only person blah blah blah. And then it messaged Fred and inbox once again. So, let's just see what it sent over to Fred. It basically said, "Hey, Nate created you as Fred.

Here's what you should be doing." Fred said, "Okay, I got it. I'm going to cue these for ClickUp once it's live." ClickUp is live. List saved. I'm on it. Blah blah blah. And then we can also see if we go back, what did it say to inbox? It said, "Hey, Inbox, you have a new teammate called Fred. Whenever you tell a customer a team member will follow up, you will message Fred with the customer name, phone number, address, and then you will put it in ClickUp as you can see.

And now, without me having to really micromanage that, we have this new process where next time Inbox does a sweep of an agent or of emails that need delegation, it will go ahead and tell Fred to create that task in ClickUp automatically. And you'd obviously need to come in here and you'd say, "Hey, Fred, you know, we have these five teammates. Maybe each of the teammates has a different type of specialty as far as like some are good with AC, some are good with other types of things.

And if we gave it a little bit of a directory as far as who to assign to what stuff, or maybe we had to give it an a database to look up to be able to look at their calendars to see who can actually um you know, fulfill it. Then we can start to add more color into Fred's processes. And just a quick check on our actual usage, we are now at 13% used up for the week. Real quick, guys, just wanted to say if you're enjoying this video and you're enjoying the way that I teach, please consider subscribing and joining my free school community.

The link for that is down in the description. We're building a global community of AI builders and entrepreneurs and business owners. We're almost at 450,000 members. We're almost at half a million, which is super super exciting. And it would just mean a lot to me if you joined. We have a lot of resources in here. We've got agent skills. We've got every YouTube video where I give away free resources. Like I said, skills, guides, templates, docs, GitHub repos, I give it all away in here for completely free.

We also have a 7-day challenge. We have a build your AIOS course. We've got a lot of fun stuff going on in here. Events, certification programs, tons of cool things that we're building towards. So, if you're enjoying it, if you like the way I teach, please definitely check it out. Hope to see you guys in there. Let's get back to the video. Okay, so so far we've gotten pretty familiar with all the stuff. We've talked about all four of these four C's.

We've kind of used them interchangeably, and now you should hopefully be understanding how you're going to keep growing up all four of these kind of together, right? like you keep building all of these different context and connections and capabilities and cadence as you go and as you add more bots. We've also talked about a few hacks like logging things in ClickUp and building you know like systems around that. And so I wanted to do one more piece of that so we get to the place where we are um with the way that I do it.

So we have a list now in ClickUp that we can use but right now it only basically gets delegated whenever an email comes in that needs to have some sort of you know person take ownership over it. But what I think is really helpful is when you're doing tasks with any of your agents ever, any project that you start working on, you are having it be logged so you can understand what agents are working on it and what the status is.

So now that we've connected to ClickUp, I want you to take ownership over something which is basically just managing projects inside of our ClickUp. So what I mean by that is not only when we are delegating out emails for one of our team members, but also when we're doing any sort of projects, if we're building new agents, if we are building new Google Sheets or building new processes or whatever the project might be, everything that we do needs to be logged inside of ClickUp for visibility.

I'm talking I want to see things like what agents did you assign this to? What time did you do it? Um is it in progress? Is it complete? When do we have the due date on it? just basically so I can at the end of each week track what actually is being moved and you know making sure that projects don't go stale and they don't die and that we don't forget about things stuff like that and so what will happen here is we're able to turn this into a skill that just happens automatically without us having to say oh put that in ClickUp put that in ClickUp so then after you've been using Grockbot for let's just say a month you can then go back and see everything that you've actually done with these bots and it gives you way better visibility into how often you're actually using these.

You could also, if you wanted to, you could add at the end of every routine. So, by routine, I mean these things that happen automatically. If you wanted to, you could go in there and you could say, um, at the end of every single run, if you took action, add this to ClickUp just so we can see what happened. Now, I just realized when I switched into here, we had this little error. So, Gmail signed out. I can't see new customer mail right now.

The only reason this happened is because I had to switch my password because I was signing in on so many different accounts and I switched my password. Otherwise, that wouldn't have happened. These should save, but if you switch your password, it will probably have you just reauthorize. So, I'm going to reauthorize real quick, but I wanted to address this so not to scare you guys that you have to like go sign in like every hour or every day only if you like actually switch something on the back end.

So, what's happening is the list is already set up and now friend is just understanding how to populate this with actual projects and tasks. Now, remember what we're going to do is we're going to turn this into a skill so it just happens automatically and that doesn't mean that it's going to be perfect. we might see some cards come through and be like, uh, this probably should have been split into two or, uh, you know, you didn't give us enough detail.

So, for example, if I go into here, we can see that we have some in progress, like the inbox emergency watch every 30 minutes. This probably doesn't deserve a task because this happens every single 30 minutes. We also have something like this. And we can see we have an owner, we have agents, we have asked, we have links. I like this. If we didn't like this description, then we would ask it to change it. But right now, my only feedback is these routines that run so frequently, like every 30 minutes, probably shouldn't be a task.

I think that having this end of week one is good because then when the end of week report comes out, we can mark this as complete. But we're just going to go ahead and get rid of that. So, I'm just going to say that looks great. Two things I want you to do. The first thing is I want you to turn this into a skill. So just like a log to ClickUp skill so that every single time you actually delegate a project or you understand that we're building a new project, you can do this as a skill and just like it's always consistent because I like the way that you have the description set up right now.

But I will say the inbox check every 30 minutes, I don't think that deserves to be a task. So if you could just go ahead and archive that and then update the skill because we're going to have to as we build more projects help you build your understanding of what deserves a task in ClickUp or not. But I do like how you are basically being safe rather than sorry and logging more rather than less. So take that feedback and make those changes and build that skill.

And what you'll notice on the lefth hand side is that we just got a message from inbox. You can see here that it said, okay, this basically was my scan for every 30 minutes. No new emergencies. The morning pile is still there and the rest of Google security notes from reconnecting. I'll keep watching. And what you can do now is you can see how often these routines have run. So, if I click into this one, you can see that we had one run 34 minutes ago and then we had run one run 5 minutes ago.

So, this is where you can just get visibility into if these are actually going on and if everything's running as it should be. If you want to test them out, you could click test run up here and it will just fire right away. If you want to deactivate them, you can just basically pause them by turning this off. As you can see, it disabled this routine. But I don't want to disable that. So, I'm just going to go ahead and turn that back on.

And if you wanted to delete them, you can just go ahead and delete them right there. But now we can see that friend actually created a skill called log to clickup which is perfect. We could come in here and make little adjustments if we need to. And we can see that the 30 minute watch card is archived. So that no longer exists in our board. Perfect. Everything is assigned here. We've got priority. We've got due dates.

That is all looking good. Now one thing you guys probably noticed when we were inside of my actual Grockbot is I had different like folders and I had some group chats and stuff. So really all I did there was if you right click on an agent and you put move to new section, you can kind of organize them by section. And in here I can call this, let's just say um I'm just going to call this general for now. So then as you build up teams of like sales agents and um marketing agents and recruiting agents, you can kind of box them up just for organization and then you can close them out because like I said, a lot of times you won't be in these threads.

You're more so going to be dealing with like your seuite or your executives. And then something else you can do is you can actually create channels. Let's say you wanted to have one channel just for all your bots. You could put a new channel. You can have a group chat with everybody. And this is just going to say all hands. And now you can talk to all three of them at once. Now, I don't think this is necessarily needed in this example because these bots can all talk to each other either way.

But maybe if you wanted to have like all of your different leadership team in one channel so that every week you could get a report on like what happened in their vertical or whatever the use case might be. You want to see everything they've done that week or you want to get everyone's opinion on something and give them a better environment to talk as a group chat rather than just like one-on-one DMs between all the agents.

Then that's how you do it. You just have a channel like this. Hey guys, this is currently everyone that we've have on the team. Is there anything else that you think right now we need to build? Are you guys identifying any opportunities for new skills or new routines? Or more specifically, based on what you know about me and our business and our current workflows, what other agents might be helpful for us to keep growing our business?

Now, one thing I'm curious about is if they are going to be aware that we haven't yet connected our calendar because that's like a big part of it, right? Like we have um where is it? Where' it go? We have this Google calendar here which is Nate Herk8. So, it should be like, you know, my personal calendar, but they haven't had any visibility into it yet because if I go back in here, the plugin for the calendar is actually different.

It's not the same as Google Drive. It's a separate plugin. So, I'm just curious if they're able to identify, hey, you know, that might be one good gap to save you more time. Nate is if we could see your calendar. So, you can see here inbox came in with some stuff about the inbox and looking at what types of, you know, tickets are coming in. Friend says inbox is right. We don't need another email person. The whole is about booking.

Fred comes in and says, "Yeah, I already filed those promised calls. They stood on blah blah blah. One cheap thing I can own without a new agent is a weekday check of open emergency call backs." Fred says, "Okay, go ahead and do it." Inbox quotes, reply skill, Fred's call back check, then a dispatch agent, Nate, say the word, and we'll build out an order, blah, blah, blah. So, they have kind of been able to talk based on what they're able to see and based on their area of expertise and help kind of like make decisions and ideulate.

So this is the way that I would think about first of all maybe you you go out here and you create a COO and you create um a CMO and you have them understand high level what is their goal and what are they supposed to be helping you drive towards and then you put them in a conversation where you're constantly saying okay what do we do this week what happened where did we have bottlenecks where should we be building new agents and where should we be looking to have automations and once again outsource the thinking but don't outsource the understanding don't blindly just say yep looks good do it only do these things if you think it will actually provide value and give you more leverage in your day-to-day.

And the reason why that's so important is because there's a difference here between the shared memory of your agents and the specific memory. So, let me just show you that. If I go to friend and I say, "Hey, so I'm just trying to understand what do you and all my other Grockbots know about me. What is the difference between what you specifically know about me and what all Grockbots know?" Because there's, like I said, individual agent memory per agent and then there's like a shared memory.

So friend, because I've talked to him the most, probably knows more about me than inbox and Fred do. So for example, if I go to Fred, hey Fred, can you tell me what specifically you know about me that other agents might not? And then what is like the overall global memory that you know about me and my business and this whole ecosystem? So friend says, yeah, two notebooks. Everyone on your team knows your name, Twin Cities, Chicago Time, keep explanations simple, and that your Google is Nate Herk.

That's the shared stuff. friend knows other stuff like summit is HVAC small team honest no upsell shop inbox as email Fred is clickup inbox and Fred have their own notes for their jobs and they don't automatically know everything I know. So the point I'm trying to make here is when you're building up all these agents and when you have them in a group chat like this they're coming from different places. They have different perspectives.

They don't natively see inside of each people's channels. So, make sure you're orchestrating that your like executives, as we talked about here, that your executives know a lot about you and that the operators only need to know pretty much what they need to know for that specific job. So, Fred only needs to know if I get an email or if I get a message from inbox, I have to do X, Y, and Z in X, Y, and Z order. Fred doesn't necessarily need to know what are our marketing goals for the next quarter or, you know, how much revenue are we making every month.

I'm not going to read this whole thing out, but you can see this is what Fred responded with. All right, so let's go back and talk about some connections for a sec. If I go to the plugins, you can see that there is a lot of stuff here. There's featured things, there is, you know, MCP things, there's payments, there's a lot, but there's also a lot that you can't access. Like for me specifically, I would love to connect my Grockbots to my YouTube channel, which I cannot currently do.

They would just have to do it through native web search. I would also want to connect them to maybe something like our company LinkedIn, which we cannot do right here. So, one solution to that is we can use something called Composeio, which says connect and operate a thousand plus apps right from here. So, what I'm going to do is I'm going to go ahead and click add Composeio. What you're going to do is you're going to go to Composeio and you're going to create an account.

And what that will look like is this. And it's completely free to create, which is awesome. And now inside of Composeio, you have all of these different things you can connect. And right away, you see some of the most highest leverage ones for me are we get to connect YouTube and we can connect LinkedIn. If I go ahead here and search, we can see LinkedIn is active. We can see that we could also connect something like QuickBooks, which is really helpful.

You can connect a lot more through something like Composeio that has all these different servers to connect. And once you've connected it inside of here, it automatically syncs to Grockbot once you tell it to. So, I'm going to authenticate into Composeio. We're going to pull up this account real quick. Awesome. So, we are authorized into Composeio. I come back into here and click on connected. There we go. And now if I go back into friend and say, "Hey, I want you to look at our new plugin called um Composeio and I need you to just see what apps inside of there you can actually access." Now, what's cool about plugins is plugins and skills and the whole like little browser, those are universal across all your agents.

So every single agent that you spin up can use Composio and can use everything that you've connected. Whereas the memory is what differs by agent. So you connect it once, all your agents can touch it, which is great. So you can see inside of Composeio, we are connected to LinkedIn, QuickBooks, YouTube, and Perplexity. So now if I said, okay, whenever you're doing research, I want you to use Perplexity within Composeio.

Whenever you're looking at YouTube stuff, you would go here to YouTube. And now we can also access this other stuff if we wanted to. And then what I would want to do is say, can you just make sure that you add this to the shared memory so that every single agent in the future understands that these four apps are currently authenticated inside of Composeio because I want them in the future if they need to touch it, they need to know where it all lives.

And so this is the point I was trying to make about the connections, but not only these connections, we are now giving it the context to understand where these connections live. And then that helps us power even more capabilities and cadence as you guys can see. So, it's just going to go ahead and update that shared memory so that every future agent understands this. Now, what you've noticed about what we've started to build here is we're building a lot of things like knowledge work.

I like using this kind of stuff for things that are quick tasks, project management, creating resources, approving things, checking things, task based things rather than building production apps or software. So earlier I talked about sort of like what are some things that I might use like cloud code for um or codeex for versus using something like a Hermes or more of like a personal agent or agent army type of deal. Well, I personally would use cloud code for more things like um building like apps or custom software or building websites or games.

If I wanted to be able to do this stuff on the go, I would do more things like I said, you know, triaging the inbox or automatic like research reports and, you know, maybe even like creating marketing assets because we can connect this to different things to create like carousels or LinkedIn post or Instagram reels, all of that type of stuff as well. But really, if I wanted to build websites and apps, I would want to have more of a traditional like coding harness that I can connect to GitHub and have GitHub connect to my Verscell and I can constantly be pulling in these new code bases and things because we have more control over like sessions and context rot and controlling what goes in and controlling what we're looking at.

It just is much better at staying organized on building legit software and websites and things like that compared to something like this which can get messy pretty quick if you're doing like super high-risk work like that because it's even more vibe coding here than it would be if you were opening up cloud code or codecs. But that doesn't mean it couldn't happen. Like I have agents inside of mine that are building me little landing pages that are simple.

Or I have one that's like they're building me these HTML reports and they're putting it on an actual URL so that I can just get my end of week report on my phone and I can view it and it looks cleaner than just looking at like a markdown file or something like that. And what's cool about that is if you find GitHub repos that have skills or projects or other things that you want to play with, you can still do that inside of Grockbot.

So, Hyperframes helps agents write HTML and it renders it into video, which is probably something that a lot of us need for our websites or for marketing assets, but maybe a lot of us don't have like a video editing team or have any experience editing videos. So, let me just show you a quick little hack of what you can do with something like Hyperframes. So, I'm just going to go ahead and I'm going to add a new agent here and I'm going to add a new bot.

And we're just going to create a new bot. This one is going to be specifically just like a video editor. So one it said it's gonna so this one's going to come back and say hey what can I do for you and we are going to just kind of set this thing up. You are going to be my video editor. Your name is going to be slice. What I want you to do is investigate this GitHub repo and pull this stuff in so that you can actually use the hyperframe skills and understand what hyperframes does.

And then you're going to actually write HTML and create a video for us. This video, I want it to be like a 15-second like sizzle reel that we can put on our website as far as like who we are, what we do, and how to work with us. And for the how to work with us call to action, just say book a call. Just put a fake number on there for now. Actually, if you have this real data, which you can get from messaging friend, he's the chief of staff in this organization.

Then reach out to him and see what real data you have to work with and then create us the deliverable. Here is that hyperframes GitHub repo. Cool. So now this was us showing how we actually created a bot ourselves rather than having friend make one for us and it's a very similar process. It's basically just onboarding giving one specific role and hopefully all of this stuff should be set up. You can see that it just named itself slice and we have a description which is you are the video editor for Nate.

You build short HTML hyperframe videos for Summit Home Services starting with website Sizzle Reels. So that's pretty specific. I could basically just get rid of that part and just say yeah you build short videos for Summit Home Services. And then I'm just going to give this guy a quick tag so that we can keep ourselves a little bit more organized. And I might even just move this guy into a new folder or section. And we'll call this one content.

Actually, no, we'll just call this marketing because that's a bit more broad. So now we have a marketing folder that we're starting to build out. And what this might trigger us to do is instead of having friend manage slice, we might want to make like a CMO. And then we could create like a LinkedIn bot and we could create um like a Photoshop bot. And then whenever we want to do stuff in the marketing folder, we have our CMO be sort of like our point of contact to delegate work out that way.

Same way you would do with like finance and accounting and same way you might do with just general operations and things like that too. And you can see already that Slice has been messaging friend to make sure that it's going to put legit factual information about the business inside of our video. All right, so you can see that we had both Friend and Slice message us about this video. It's done. It's 15 seconds. But what friend did is it gave me a link right here which actually goes to ClickUp.

So if I follow this link, it takes us to this completed project. So that just shows you that it was able to understand, okay, we need to log this in ClickUp so that Nate can see that later. And we actually have the deliverable, we have the agent, we have, you know, all of this information about it. And now if I go back into the Grockbot, we actually want to go watch that real quick. So I'll go into Slice. It sent us the video right here.

I'll click play. We have Summit Home Services, Twin Cities Heating, Cooling, and Water Services. So that is real. We fix it. We don't sell you a system you need. So that aligns with what we said. You can notice the background is green as well, which follows our guidelines that I told I think I told inbox that earlier and it must have relayed that to friend, but something just happened here. It couldn't load media. And that happens around the like 12 second mark.

I'm not sure why that is, but you can see that this is a sizzle.mpp4. So like I said, I'm not exactly sure why that happened. Maybe a rendering error. But what you guys might be wondering at this point is where are all of these deliverables being saved? You know, sometimes we're creating things like a Google sheet, in which case we're getting it stored in the cloud in Google. But what about when we're creating these MP4s or when we're creating markdown files or the knowledge files, all the stuff that we've talked about, where does that live?

And that's where if we open up the computer, we can see that they do have something right here called the file manager. And in here, we can see that we have this workspace and we have videos. We have the weekly email report Python scripts. We have a weekly email report, Excel sample, we have hyperframe skills. So what it did is it manages this and it saves things. So the GitHub repo that I gave it earlier, this hyperframes one, it pulled in what it needed from like the cloud files and the skill files and it saved this in its local file directory that every agent can use, which is pretty cool.

And it also saved the video right here. If I go here, the Summit Sizzle, this is where it should have saved the MP4. Um, if I go to the renders, we can see that it lives right in there. That's the MP4 file. So, this is where all this stuff lives. If I actually go ahead and um, open this up. I'm not sure if it's going to let us open this up in anything. I'll try Google Chrome. And there we go. It actually lets us open it up right here.

And now we can see that we get past that little rendering error. Good. So, that's where that loaded up. And once again, all of your agents can access this stuff. So, the same way that if you're on, you know, if you're on Mac, you've got your Finder. If you're on Windows, you've got your uh file explorer, and what we're building towards is a local file directory of all of this context. So, this is what mine looks like.

You know how earlier I talk a lot about this AIOS with our context and our connections and all this stuff? This is what mine looks like. If you guys have been following my channel for a while, you've probably heard me talk about my Herk 2, which has everything about my business, about me. Uh, it can touch everything. It knows the business and our goals better than I do, honestly. And it's just a collection of files and folders.

I've got my Herk Brain. I've got all of these other worlds. I've got um a projects folder. And inside of the projects folder, it's all the projects we work on. As well as in here, one of the projects is called YouTube videos. And this is a bunch of stuff that I do for all my other YouTube videos. So, this is a massive folder with a bunch of files. And the cool thing is I can give that this whole file or sorry, I can give this whole folder to different agents.

You know, actually if I come to my repos in my GitHub, this is my Herk 2 project, which means I could give this link to my agent, which is what I did for my actual Grockbot, and I could say, "Hey, can you just like clone this locally?" So then this whole workspace could look just like my actual Herk 2 that lives on my real desktop right here. And like I said, all the agents can actually use this. So, if I leave out of here and I go to friend and I open up his computer and he has a different Chrome tab pulled up, but if I go to the um file manager, they can all work together here.

But what you need is you probably need to instruct them to have some sort of system in here that they can all share together because remember how we talked about the shared memory and the agent specific memory? If you had more shared context in here, they could all reach it. So, that's one thing that I want to quickly tell friend about. I want you to create a bit of a better folder structure inside of your local file directory since all of the agents can work out of there.

I want you just for now to set up a folder called context and to set up a folder called projects. And whenever we're creating new projects, just store them in a new folder in the projects folder. And right now just put in some base information about me and about our business and just give some context in there. And over time as we need to grow that, that's where you guys should be saving work in this sort of local file directory structure.

So anyways, the reason I wanted to point that out is just so you understood where is all of the stuff going because if you have something making like an HTML every Friday or markdown reports every day, those are being stored somewhere that the agents can touch. And it's important that you at least understand that you can come in here and you can click through yourself. And also that when agents are passing work around, they can basically like give each other the file paths of different reports or different docs.

And that's how they can more effectively communicate and share work and contribute to the same stuff. But it is important to think about the way you want to set that up, the structure you want as far as maybe you just start with a big folder called projects and later you make that more organized. However you and your team will typically work on like a shared drive or a Google Drive or however you want to store data. Just think about either doing it locally or maybe in Google Drive or in notion or maybe all three.

Just be aware of where your data is actually going and where your projects live. You can see that it just started to build that out. And if I open this up, we now have a folder called context and we have some markdown files with gaps and summit and the team and me. And then if I go back, we also have projects and it put in like the sizzle reel and other projects that we build will go in here as well. And it also went into ClickUp and logged that as something that we just did.

So, if I go in here and I show completed tasks, you can see that it set up the shared context and projects folders. So, as I was editing this video, I realized that the Hyperframes video that I showed you guys was just really boring and really bad. And hyperframes actually is pretty cool when you build some skills around it. So, just to show you a quick example that you can actually get some pretty solid motion graphics in here.

This is how I use hyperframes for some YouTube videos to create intros and to also create things like um some open loops. If you see like this full video, I was able to prompt it enough so it understands the style where I want it to animate these flowcharts and it keeps like these like spatial elements. As you can see, it's going to move on to number two in this list right here. You can see that it takes this whole flowchart, puts it back into the box, and then we're going to go into two, three, four, five.

So, um, it's just pretty cool when you really start to prompt it and build some skills around it. It is pretty solid. So, I just wanted to show that off real quick. But then I went ahead and asked Aaron, our CMO, to make another one. And you can see that it actually delegated around with Slice and Studio to create like different images for B-roll and also logos and stuff. And then Slice came back and made the actual video.

So if I come in here real quick to Aaron's screen, I'll just give this a quick play. So, I mean, this is nothing groundbreaking at all, but if you don't have a team where you can afford a video editor right now and you take a couple iterations using the bike method I talked about, you can really get some pretty cool stuff done just in here with a few different agents. Now, one use case that I think is really good to do with that is to have an agent that is always like reading your meetings.

So whether you use Granola or you use um Fathom or Google Meet or you know Fireflies, which is what I'm currently using, having an agent do that. So what I want to show you is an agent that I use myself. We're going to create a new bot. And what we're going to do for this bot is connected to Fireflies. And all we have to do is log in. And what I want to do is at the end of every weekday, because I usually only have calls on weekdays, what I wanted to do is grab the transcripts from those meetings from that day and then put it into a folder locally that is like meeting recordings.

And maybe that's organized by quarter or by month, but I want those transcripts to be stored in there so that all of my agents can see that and have context of like some of the important meetings that I'm going in because that context is really good for your business, for making decisions, for understanding priorities and stuff like that. So, you are going to be my meeting bot. Your name is Fireflies. Basically, what's going to happen is I need to authorize into Fireflies because that software has all of the transcripts from my meetings.

And then at the end of every day, I need you to set up a routine. Let's just say 6 p.m. every single day. Um, weekdays, sorry, 6 p.m. on weekdays where you're going in, you're pulling all the transcripts and you're storing them in our local file directory. So, you should probably create a new folder in there called meetings. And I want you to organize that by month. And then, yeah, you're basically just storing in the transcripts of my meetings every single weekday, putting it in there so that our other agents can always pull from that if they need context on certain calls.

And you are going to go by the name fire. And so, that's really the beauty of it. I have one very specific type of task and automation that I want to set up. So, I just make a new bot. I instruct it what to do. It's going to build a routine. I'm going to authenticate into Fireflies. And you just keep scaling up different capabilities and different cadence things just like that. And I think the toughest part about this stuff a lot of times is figuring out what are the good use cases rather than like how does the tech work.

It's more about what are the use cases that actually help. And I think a really good way to think about that is just by thinking about the triggers. Meaning what happens in your business that triggers you to do something. Whether that's time based or whether that's event based. So are there any things that I have to do on Mondays or Fridays or Wednesdays after I have a meeting? Is there anything I do every time? You know, um when I get this type of email, what do I do?

When you think about the triggers and you understand if X happens, do I always do Y? Then that's a really good opportunity to set up some sort of bot for an automation there. So it set up its name as fire. It's currently unassigned. I'm just going to move this to the general folder. Um it's created the routine right here, weekday fireflies dump. And now I need to add the Fireflies connector. So I said, "Yep, go ahead and authorize that up." So once again, it's just as simple as signing into your account.

And luckily, there was a native plugin for this. Otherwise, we'd have to try to find it with something like Composeio or a different sort of like MCP server aggregator. And now it's just going to go ahead and make sure that that connection actually works. And it told me it was going to grab everything that's already in Firefly. So I'm just going to go ahead and say only grab five just to prove the concept. And you can see that I was able to actually interrupt and steer the direction because otherwise it was probably going to start crawling through who knows how many, maybe like a hundred transcripts.

And I know that was going to take a while. So I was able to interrupt it because I watched it and I saw that it was going down the wrong path. And that's why it's important to be watching them because each chain of thought, it's basically telling you what it needs to do. And that's good because then, like I said, you can steer it in the right way before it starts wasting more of your tokens. Because if it was going to waste 30 minutes and a bunch of our session limit, as you can see, we're at 30% now by just like doing something that you're not even going to use at the end of the day, then you want to stop it before it gets too late.

Awesome. So, that has come through. Let's just go ahead and make sure that it actually created this correctly in our file explorer. We're connecting to the browser. We're going to open this up. We can see we have meetings and now we have August of 2026. And this is where we can view those markdown files if we needed to. And now we have a system set up where this is actually going to do that automatically at the end of every single day at 6 PM on weekdays.

All right, so I think it's time that we need to build a CMO. So I'm just going to create a new bot real quick. This is going to be our chief marketing officer, and it's going to help us create one more bot that's going to sort of like work and report to the CMO, which is going to help us create marketing assets. And I'm talking things like pictures and images and ads and things like that. So let's go ahead and get this spun up.

You are my chief marketing officer. We are going to call you Aaron. E R I N. And your job is to help me pretty much stay on top of things like marketing, right? Like video editing, um at some point creating content at scale, analyzing content, marketing strategy, growth, stuff like that. So, the first thing I need you to do is I need you to help me spin up a designer agent. This designer agent is going to help with really just creating photos and videos for us.

So that's what I want you to do for now. Okay. So as this is getting spun up, you can see that it just created a new agent called studio which I'm going to go ahead and move to the marketing section. Um Grock has something called Grock Imagine and that helps you natively right in here create photos and videos and stuff like that. So what I'm going to do is just say um wait for a specific brief and then I'm going to say go ahead and have studio create an image for the whole marketing team.

So, one for Studio, one for Slice, and one for you, just like profile pictures for our Grockbot account. And we're going to see how these come out. But there's a way to get a lot better models and way more. So, there's a thing called Key AI, which is free to sign up and then you have to put in credits to actually use it because it's an API. But in here, we have access to a ton of different models, video and image, and also like music.

If I go to the API market, you can see that we can get um you know like Google models, we can get Juan, we can get Grock, we can get Gemini, obviously that's a Google model. We can get Opus, we can get VO. So if I go to like uh text to video, you can see that we get all of these like Cance. If I go to text to image, we get GBT image 2, which is my favorite one. It's personally what I use for pretty much everything. Where'd it go?

GBT image 2 right down here. So we have the ability now to connect Grockbot to this so that we have way more flexibility with what models we want to use rather than just being locked into Grock imagines ecosystem. So in order to do this though this is where composio really comes into play because if I go to plugins there's no key AI. There is something like Higsfield which is a very similar type of tool but this one's a little bit more expensive but Higfield is also pretty solid.

I've made some other videos with Higfield before. So, what we're going to do is because we're already connected to Composio. If I go back into Composeio, which is right up here, they have KAI, which is amazing. So, I'm able to connect right here to Composio and Key. I have to do this by getting an API key for key. So, all you have to do is go into KAI. You're going to go to API keys, and you're going to generate a new key right here, which is kind of just like a password.

So, what I would do is I would create a new key. I would just call this one my Grockbot key. I want all models by default. Create that. And then I take this value, copy that, and go ahead and put it in here. Whatever the tab is, where'd it go? This one that's asking for our API key for key connect our account. And now when we go back into Grock, we should be able to have it use Composio to use key. So what I'm going to do is say, okay, so Studio's on that.

What we also just did is I connected um key.ai inside of Composeio, and that gives us access to way better image models. So what I want you to do is send studio another task to do the same thing, three profile pictures, but this time do it twice. So do it with GBT image 2 and then also do it with Nano Banana 2. So we can just compare these different AI models and which ones we actually like the best. And it didn't spell key AI right.

So I'm just going to go ahead and fix that. But it should be able to have found that either way with context. But anyways, we're going to go ahead and shoot that off. Oh, and you can see Studio already changed its profile picture. You can see here studio created this picture and changed it because what happens is inside of here when you edit the avatar you can generate images you can upload or you can use the bot ones and so studio actually proactively made the image and then changed it as well.

So that's just to show not only can you create pictures like that and this is Gro imagine but you can also do things like creating marketing assets. Let's say you were building like you had an e-commerce brand or whatever. You could create different product shots, different marketing assets, different ads, things like that. And now that we have key, we get access to better models because once these other ones come in, I'm assuming you guys will see how the other models in my mind are better.

So it sent studio the same message and now it's doing it with two other image models. So I will let you know when these come back. Okay, this is actually interesting. So what happened is the connection to key.ai AI doesn't actually allow these agents to make an image. So, I'm not sure if it's because the way Composio connects or what. So, we basically have two options. We could use the API documentation in here and we could basically just have key use its CLI to hit the API.

That's one thing. That's one workaround that you could try. But what else you can do because we understand that it's able to use browser use. All I have to do is I can open up the desktop and it's opening it up right now. And I can just go ahead and sign into key.ai AI because this is also a headed interface. So right here you can see that it lets you choose the model. It lets you put in your prompt and then go ahead and generate these assets.

So what I'm going to do is just sign in once to key.ai and then studio this agent should be able to figure out how to navigate here in order to still generate the assets that we're looking for which is going to give us the end result that we want. So a little bit of a um workaround that we had to do but as you can see because of the computer use and there's other ways to work around when you can't find the right connection.

Cool. Now you can see that I am signed into this account. So now we just have to wait for Studio to learn the interface and click around, put in the prompts and get those assets for us. Okay, so that took a little while because Studio had to get in here and it had to learn the interface and it was trying to generate these in parallel by having different tabs open, which was good. I'm glad that it did that, but it still had to learn the interface a little bit.

And so honestly, that could have been a good use case of teaching a task if I said, "Hey, give me the prompts. I'm going to come into here. I'm going to show you how you navigate, where you put the prompts in, how you generate these, how you save them, and then it turns that into a skill. So, that is something I'm going to say real quick is just now that you've navigated the interface of key and you've understood how to generate different types of images with different types of models, it would be very similar if you did videos.

So, I just want you to save that as a skill so that next time we need generations, you're able to just navigate the interface quicker and, you know, just give me the the outputs quicker. So, anyways, good job. Build that skill. So now we're going to have that turned into a skill. Whenever you do something that you might do again, just turn it into a skill. It's ra you'd rather be safe than sorry there. So I got this message back from Aaron.

Aaron is the CMO if you guys remember. Actually, you know what? I'll just go ahead and give her the tag because we don't yet have that. So anyways, we got that message back from Aaron. It says both sets are in and it hasn't yet installed any of them. So these are the three from GBT image 2. You can see that the prompting was consistent in a way where these two women look the same. These two women look similar, but these two men look very different.

Um, but they have similar backgrounds, right? They've got like the the green style with the little like kind of light behind. This one has like these yellow streaks. We have GBT image 2, like I said, and Nano Banana 2. You can see the styling is different. I'm assuming it used the same prompts for all three. Maybe a little bit different, like I said, or maybe these prompts weren't as specific. But anyways, now we can just choose which ones we like best.

And for now, I'm just going to go ahead and actually just say Nano Banana 2 because I think these look a little bit more like they were trying to be realistic, whereas these look a little bit more like I don't know, just like agent profile pictures. So, I'll just go ahead and say Nano Banana 2. It will relay that message down. And then our marketing team, Aaron, Studio, and Slice will all have different profile pictures.

And so, what happens here is it has to message each agent individually to do it. As you can see, Studio changed theirs. And if I go to slice, it says, "Hey, Nate wants this to be your Grockbot profile picture. Please set it now." And look what it did. It gave it the file path. Meaning, if I actually go into Slice, for example, it gave it the exact file path. So, Slice opened up its computer. It went into here, it found the pro the portraits, and then it chose this picture and set this as its own um profile picture.

So that's just another example of how it's saving things locally and the agents are able to pass work around because they have that shared computer. So I wanted