His AI Bot Closed $10,000 in Sponsorships | Alex Finn

0:00I think the best way to do it [music] is kind of start small, think of like the three to four main areas [music] of your life that you want AI to help automate and improve. Make like one bot for each. [music] Uh I don't really have a plan. I'm just going to go through everything I do, and I think there'll be some interesting things in there, and so I hope it's

0:23helpful. Um this is my Grok bot. Uh I have a whole team here on the left. What I did was I I broke it down, my team, into basically all the areas of my life, content, my software I build, uh the community I have, uh because I found something really interesting with Grok bot that I think is very unique to it. Um compared to any other AI agent there, which is I find that your different bots kind of naturally interact with and engage with each other based on their like roles and titles and context. And what I mean by that, I'll

1:02show an example here. So, this is my team over on the left-hand side here for one of the apps I'm building, Henry Intelligent Machines, and I kind of have different roles here, uh the chief officer of the app, a developer, a researcher. And what I actually kind of naturally found happens with Grok bot is they'll look at the titles of other bots. That's why I I don't recommend like naming your bots like email specialist, content specialist. Like give them like names, and then make their titles what they are. Like you can see their title next to their name there. What I find kind of naturally happens between your bots is they

1:41actually can see what each other's titles are and interact with each other organically based on their roles. So, example here, I'm in Kelly right now, who kind of leads everything to do with this app I'm building. And Kelly, I I never told Kelly to do this. I never told AI bots how to interact with each other, how to message each other, anything like that. But Kelly, as you can see here, organically just messages Hank, who's the developer for this app, and Build, who's the CTO for everything I do. Build's the only one that has an unoriginal name for their name, but it

2:15says CTO next to it. Didn't tell them interact with each other. Didn't tell them to work with each other. They just organically, as they work, I don't think I've messaged any of them for like days now. It's literally been like a week since I messaged Kelly. She just autonomously works and just chit-chats with like Hank. Hank will organically go to Kelly and be like, "Hey Kelly, is it all right if I do this? I need permission for

2:39this. Are you all right if I do this?" And Kelly be like, "Okay, yeah, you can do that." Or if Kelly needs to get like even more permission, she'll talk to Build, who's the CTO of everything. And so this really interesting, natural, organic relationship starts to form in between the bots based on their roles, who they are. And so everything I do in Grok Bot is like in an effort to build the

3:07synergy between all of my bots, right? And so, you know, they'll She'll go to Build. Build's the CTO. I don't even look what the chats are like. Goes to the CTO, says, "Hey, we have to work on this ticket. We have to work on this ticket. Should I do this? This is blocked. Can you help me out here?" And I never said, "Hey, go to Build, the CTO, if you need help there." She just kind of naturally does it. So I've started like doing things to build that synergy between all of the bots. One of those things is I've taken all my executives, so every team,

3:37content, community, Henry, the tech team, the product research team, they all have an executive, a leader. I actually don't use chief of staff at all. I'm like bearish chief of staffs. I'm more like I like the modular teams with the leader of each team. And I put them all in a channel and every morning they have a stand-up,

3:57all the executives between the teams. So, every morning they have to go and talk about what are their top four priorities that day, where they're like blocked, and then they give each other feedback based on what their priorities are. And one thing that's key here is you want to make sure every team I hope I'm not showing anything private. I did not review any of these chats, so I'm Don't read the chats if there's anything. I think I'm safe. There's no

4:23IP addresses. Okay, we're good. Um One key here though is like if you just have AIs just kind of looping talking to each other with no goal, it's just all slop. But if you make sure each team has a set of goals they're measuring against both long-term and short-term and put that in their memory, the executives, and then you throw them in these stand-ups, they'll actually work towards those goals. Right? The AI, it's slop if they don't have goals, but if you give each team goals, they naturally work towards those goals. So, they'll do the stand-up every morning at 8:45. They'll talk about uh what their current short-term goal is

5:04this week, what's their progress, what's blocking them, and what they plan on doing today to get a step closer to that goal. And then the other executives will jump in and say, "Hey, you know, Kelly here from him, uh here's what I'd recommend based on the context I have." Or, you know, "Frederick here from the content team, this is what I would do based on what I've seen work here." Or, "Dusty from the community team, this And they kind of naturally help each other based on the context in different parts of my business move towards that short short and long-term goal. And like what makes this unique and why I never really done

5:36this with other bots and why, you know, I'm really in love with Grock bot is there's some sort of magic going on behind the scenes where they're all very good at figuring out who's who, what their roles and responsibilities are, what like context they have, and like what they're in charge of, and then reaching out to each other at the right time, right? Based on, okay, I'm going to do this really risky thing, so I'm going to go to the CTO who has all the context around permissioning and things like that. So, everything I do is in an effort to like build the synergy between the bots, make sure they

6:13have the right contacts, the right titles, the right teams, the right roles under them. Now, what's the best way to do that? I don't think you just go home right now and then you just make 40 bots. I think that'll be a a good recipe for confusing yourself and being like overloaded and like, you know, just I I think the best way to do it is kind of start small. Think of like the three to four main areas of your life that you you want AI to help automate and improve. Make like one bot for each, just the chief of each, and then as you go, the more roles that naturally

6:52uh emerge as you're working, you can build the bots for them. And so, like, I'll give you an example here. Um here's my content team. And Barry was kind of the original content guy. He's like, "Hey, I need to know the moment uh a new piece of AI news drops so I can make like a hype YouTube video that gets me a bunch of clicks." I'm like, "Okay, I'll check X every 5 minutes and let you know when

7:16the news drops." And then like, I started getting I get if if you do YouTube and you have more than five subscribers, you get a thousand spam emails a day from companies trying to get you to shill their horrible app, right? And so, I have an inbox with 10,000 emails a day looking to get sponsorships. I'm like, "Well, Barry's checking X every 5 minutes, so maybe I'll make another bot under them, Cindy, who is actually in charge of monitoring the inbox and like finding to see if there's any real sponsorships there. Day one found like $10,000 of sponsorships that Cindy like individually closed herself. Um and so

7:53just organically over time these new roles emerge and you just create the bots as they emerge rather than trying to go home and do it all at once. Um time check, how much time do I got? 5 minutes. Perfect. I can keep going. All right. Um and so yeah, you want to slowly make a section for each area of your life and

8:15then slowly expand. Don't make 100 bots. It's very easy to go on X, see someone go, "Oh, I got 14,000 bots." Then you go and you run and you make 14,000 bots yourself. Like you're This is going to sound weird, but like I think it's critical This is going to be the weirdest thing I

8:30say tonight. I think it's critical you like form some sort of relationship with each one of the bots where it's like I get a I bear with me here. I get like an email from like a sponsor. I don't know if it's real, I don't know if it's fake, I don't know if it's legit. I instantly think, "Oh, I got to go to Cindy to have her review that because she's seen 10,000 emails and like has context around that." If I were to have gone and made just like 40,000 bots off the rip, I wouldn't have that intuition to know which bot kind of has the can handle

9:04each task the right way. And so I think if you expand too quick, you try to do the whole AI agent thing that you see on Twitter with 40,000 roles, I think it's going to be hard to kind of form those relationships so you can intuitively find the right bots. Now that's kind of the core of what I do. I'll go rapid-fire tips left and right here. Uh favorite plugins, X, obviously, because that's just something that Grok bot has that the other bots

9:28don't really have. Um Agent Mail, I'm not sponsored by any of these people. Agent Mail, uh I love Agent Mail because I think the kind of hack to getting your bots in your accounts before Grok Bot, I mean, you could have done this before that, but was to like invite them to your or make your own Gmail account for them and like do or element your own Gmail, which I I thought was just too difficult or like too risky. Agent mail, if you turn that plug-in in, it's like free. They You can just say, "Hey, make an account here and like manage it yourself and then like

10:02invite yourself to my team." And it works so well. Like I have like a community and I went to Dusty and I said, "Dusty, uh make your account in my community and invite yourself to it." Now, Dusty's like a moderator in my community, made it like really easy to do. So, Agent mail, great for inviting these bots to

10:22your teams to do things for you online. Tailscale, if you have multiple computers, uh like I have a bunch of computers at home. I have my own AI lab. I have different In my tech team, I have my uh my bot that manages like my computer, my AI computer, whole bunch of things. Uh I have a bot for each. So, if you have Tailscale, they can go and connect to all your different computers and move around and get things done for you, uh which is awesome. What else I got

10:48here as last-second tips? Um I like going to like the big juicy models. I I won't name competitors, but go like the big ones that burn tokens and go to your chief of staff bot and be like, "Hey, create a markdown file that uh describes our entire Grok Bot system with all the bots, their roles, their responsibilities, and how they work

11:12together. Put it in a markdown file. Take that, go to the big juicy models. Grok 47, hopefully soon will be the big juicy one I use. And then, give it to them and say, "Hey, how can we improve our Grok Bot setup? Break it down, see what bots we can add, what routines we can change, what routines we can add." I found that that doing that kind of grokbot audit uh regularly keeps my team like nice and succinct and like chugging. Um Linear as a second brain I like using Linear as a second brain cuz they all your bots can work off one board with all your tasks on it. And I

11:48don't know, that's it. That's everything. All right, give it up for Alex. [applause]

11:56[applause] We're going to go to questions. Questions? I'll be careful. I'm going to try this. Oh. Thank you. Um how are you keeping your usage under control because it resets every week, especially because you have a CTO bot? Like I assume it all sources to a bigger model to do the implementation and then

12:19comes back with an answer. Yeah, one uh one thing is is uh this is a big reason why you should eliminate the chief of staff bot is when you go and you do this whole delegate thing everyone's telling you to do, which I was I was for that originally and then I saw how fast usage goes. Uh you burns a lot you use a lot of context, burns a lot of tokens. So, try to go just directly to the bot that manage like if you have the right context for each bot like if I want a quick script for YouTube video on new model, I can just

12:51go straight to my YouTube bot and just say, "Hey, make a script on this." I don't have to go to my chief of staff, "Okay, who should I go to? Oh, I might do this. Let me find the right context." So, if you have the right bots, the right context, just go directly to them rather than trying to go the whole orchestrator route,

13:05um you'll save a lot of usage there. Mhm. In the back there. Yeah, do you guys have any issues as the as the threads get long? Like do they start behaving differently or do you have to like reset them at any time? To be honest, I haven't experienced that. Um I I think that if you and I use it a ton like I'm just on it all day. It's always like the first window on my computer. I think if you again split up the bots nicely write concise context not too much not doing everything. You should be

13:41good to have like really long chats. Um Yeah, I I I haven't really experienced that. describe like a couple of the long-term goals? So you mentioned that like you gave them some long-term goals to shoot for. Yes. What what are the types of things and and at what altitude are you giving that at? Yeah, for sure. So example of the chief community officer. I have a online community. I want to reduce churn. I say hey I want you to reduce churn from X to Y over the next month. And every day during the stand up it comes up with a new way to address churn. It'll go in

14:16and like like look at the data in my community cuz it has its own account in there. It'll come up with new strategies. Oh it'll say I'm going to send a message to everyone who hasn't engaged in the last week and then it DMs each user individually. Uh for uh my app I'm building Henry. I want to launch beta in 3 weeks. It'll go and be like okay, I need this this this done by beta. Okay, which of these can I tackle today so that we can get closer to beta launch by then. So those are the type of

14:43goals I put in. All right, last question. Last question. Someone all the way in the back corner.

14:58All right, uh I think you mentioned uh Tailscale. So Yes. I was wondering if uh there are use cases to chain agents together or agent proxy. So for example, like coding agents take a lot of tokens, but I have like other models that I use. So is it possible for a communicator to just like talk to someone other agent which does

15:19the heavy lifting? I've seen I I've seen people do that um where they like go into their Grok's virtual computer and like log into other coding apps out there and be like, "Hey, every time you code, use this now." In an In an effort to like save tokens. I'm not in love with that. I think it's just like really easy for things to convoluted and broken and weird doing it that way. If I have like a heavy coding task and I know my cursor credits are getting a little low that day, I'll just go directly and message another agent. Tailscale for me is purely so that all my networks, all my

15:55agents are networked together so I can go to one bot and say, "Hey, go on this desktop computer and load up this model. Go on this Go on my DGX Spark and load up this model." I can just quickly move around and manage my entire network itself.