How I Run a Marketing Agency With 6 AI Agents

0:00Does your experience with AI sound a little something like this? You've been prompting for 20 minutes. The output is polished, confident, and completely useless. That's what happens when AI doesn't know anything about your business. HubSpot's AI works with your actual customer data, every interaction and every signal your team has generated over time. So instead of prompting and hoping, you ask once and ship it. Check out HubSpot.com, the agentic customer platform for growing businesses. I'm going to bare my soul to you all tonight.

0:27Eric Siu:We're going to talk about implementing AI in your organization. And we're going to talk about making one brain that's going to help you all move a lot faster. And I think maybe a good place for me to start first is more so on this slide first. So just a quick show of hands. How many of you use ChatGPT slash Claude? Okay. Now keep your hand up if you use, let's say you're using Claude code. Okay, keep your hand up. Wow, hands are going up. Okay, how about Codex, Cloud Code? And then how about OpenClock? Keep your hand up. Okay, and then Hermes Agent. Okay, oh wow, my hand went up. Okay, fascinating, great.

1:07Eric Siu:So I'll start on this slide first. For my company at least, the way I think about it is, everyone's talking about AI transformation, change management right now, all these corporate words, right? And people are saying there's a lot of resistance. It's really hard to get people to adapt. And I remember I was talking to my dad about this because he was a programmer in the 90s. And I was using the internet when I was like eight years old or so. And it occurred to me that I was like, this AI thing is no different than kind of the internet transformation when that was coming out. I was like, Dad, how come you didn't use the internet really?

1:40Eric Siu:Like, when did you, how long did it take your companies to adopt email? He said, a long time. I'm like, how long is a long time? A very long time. Five years, right? Right. Which is crazy to me, because now that you have the power of infinite intelligence in your hands, you can't afford to wait five years if you're operating a business or you're inside of a business. Right. Because if, let's say, this gentleman up here is compounding at 10x and he does it for 12 months, he's a magnet. He's already way too far ahead. Right. Versus if I don't adopt it at all. And so the rubric that we use at my company is really four levels.

2:17Eric Siu:So you have unacceptable, which is obvious. You're not using any AI at all. Then you have capable, which is where maybe you're using AI. Maybe you're using Cloud, you're using ChatGP, or maybe you're using it to search. Okay, maybe you believe in it a little bit. And then you have adoptive. It should say adoptive, but there's an A there. It should say adoptive, which is where you're using, you're building end-to-end workflows, right? Which we'll talk about in a second. And then you have transformative, where you're building closed loops, which we'll also talk about in a second too. And people are talking about end-to-end workflows, they're talking about closed loops.

2:46Eric Siu:We'll get there in a moment. But all of this ties back into like how you do this change management thing, right? And so this combines with the whole concept of having a single brain. And, you know, as recently as last week, the founder of Shopify wrote a post just saying, oh, we have released this thing into Slack called River. And everyone's like, oh my God, this is amazing, right? River, anybody can talk to River. You know, we're doing all our, we're sending all our pull requests. We're upgrading our product. And then people are just talking to River and everyone's learning in public, okay? That's one example.

3:18Eric Siu:Maybe a month or two ago, Jack Dorsey, founder of Twitter, now called X, he runs Block right now. And he's, you know, they cut 40 % of staff, but he talked about this concept of this world brain. Okay. So we just call ours single brain because I have single grain with a G. So I bought the domain single brain. So that's why I call it single brain. And we'll get into it. So let me back up a second here. I'm just going to call it a couple other things. So this slide shows that only 9 % of enterprises are actually scaling with AI right now. The other 91 % are not really touching this as much as they should, right?

3:57Eric Siu:That sounds a little weird, but anyway, you get the point. So almost nobody's shipping right now, and this whole concept of having a single brain is going to help get you there, okay? And I'll show you some actual examples. I don't want that just be all theory, and I'll maybe, time permitting, maybe I'll give you some practical tips on how you can do it, because we've been doing it for ourselves for a couple months now, we've been doing it for our clients. The moment they drink this, it's like game over. They're just like, whoa. I think one of the, I'll share one of the quotes. One of the clients was like, when I don't have this or it's not working, it feels like I'm drinking soup with a fork, right?

4:30Eric Siu:So that doesn't quite work out well, right? So execs see it, right? Execs see the benefit of this. Those of you that are running a company right now, you might be seeing the benefits of this, but maybe your team might not be seeing the benefit of this, right? So 72 % of execs can see the benefits, but 10 % do not. But the gap isn't necessarily the technology because clearly it's working for some people and for some it's not, but it's the implementation, right? And this will help bridge the gap here. So here's where we get into the actual bits of single brain right here. So you can see these little animals here.

5:02Eric Siu:My great designer, we have a raccoon, we have a bear, we have a fox, a cat. Nobody's talking to each other here. You have Salesforce icon, you have a HubSpot icon, you have Google Analytics, you have Slack, you might have Shopify, you might have Google Search Console. The problem is when none of your tools talk to each other, when none of your data nodes talk to each other, you can't compound. And we all love compound interest, right? It's the eighth wonder of the world. And so if your tools don't talk, your knowledge stalls, change management, there's that word again, change management, right?

5:32Eric Siu:And then what ends up happening is pilots don't ship and everyone gets frustrated. And then what ends up happening is you start having this AI theater in your company. We've certainly been affected by that in the past. where it's like, if Eric wants this, we're going to do this, right? Which is stupid because it doesn't end up benefiting anybody. It doesn't benefit your customers. It doesn't benefit you. And it ends up being this AI like theater that makes it look like you're doing something because of motion, but you're not really doing anything. So the way we think about it is one brain, okay?

5:59Eric Siu:So you can connect your GitHub, your Slack. Everything ties into this one brain, okay? You can call it the world brain, unified intelligence, single brain, whatever you want to call it. That's the concept here, all right? And the org chart then starts to shift. When you start to connect your tools into Slack or Microsoft Teams, here's the thing. You don't need to worry about change management because your team is already working inside of Teams and they're working inside of Slack. So if that's happening already, they're used to talking to humans. These agents that you have that are living inside your Slack that are connected to all your tools, guess what's happening now?

6:33Eric Siu:You don't need change management because they're already starting to eat this stuff up, right? So the way this org chart works, and don't worry, we'll share the slides with everyone afterwards, those of you taking pictures as well. Single brain is at the top. That's the main brain that you have. Around it, you have like, because my company does marketing, right? So we have SEO lead, you have sales, you have paid media, you have ops. Those are, these are people asking questions to single brain up top. And then right below it, okay, so single brain is an agent, right? It's an ambient agent that lives inside of your chat tools.

7:05Eric Siu:and then you have a fleet commander, like an agent commander that basically will manage all of your other agents below and then you have sub-agents below that. So we named ours after like DC characters, Oracle, Flash, Arrow, Cyborg. My designer just decided to make them like these mammals, I guess. But it works. So the way it works is you have a human up top, you got the main brain, you have a fleet commander and agent fleet and then you might have sub-agents. That is how this works and I think this might be interesting right now. I think in the next 12 months or so, almost everybody's going to be doing it.

7:36Eric Siu:So you might as well start doing this right now. Because I don't think it's that particular. I think it's very simple to implement. Is it a pain in the ass? Yes, but it's still simple and it's worth it. And I'll share some numbers with you later as well. So for us, at least, we have six agents. And you want to think about it this way. You want to have, when you're designing your company around AI, you are designing for specialist agents. So you might have agents that are good at SEO. You might have some that are really good at conversion rate optimization, some that are really good at email marketing.

8:05Eric Siu:You want them to focus in these areas, right? And when you set these up, you want to make sure that they're set up in isolated instances so they don't bleed over to each other. And I'm not going to get too technical in this talk because I have limited time. But for this one, just for example sake, Alfred is kind of the chief of staff. Hermes, or in our case, we have an Hermes logo in our Slack, so we would call it Hermes. That one is more so like the brain that will check Alfred because you need these agents to check you. other sometimes they become unreliable um so there's a lot of work to do around um security stability and reliability we like to call that ssr okay so a lot of work to be done there and then we have like an analytics agent called oracle we have flash that handles social content and just an example with flash it writes a lot of my twitter posts now and some of these twitter posts are getting like 300 000 views 500 000 views or so and we're getting amazing we're driving amazing pipeline from it and we're having amazing conversations um with with people that i never would have imagined that I'd be talking to.

9:01Eric Siu:And that's all being driven by just the work I'm doing with these agents and me going to the agent and saying, hey, what's some cool stuff that we did this week that would be a good X long form article. You can do that for LinkedIn as well. You can do that for whatever social channels. There's tons of people doing this for TikTok right now. You want to talk about TikTok shops, all these things, trying to apply this to e-commerce too. The sky is the limit. Once you start working like this, everything changes. It's not just you love ChatGPT anymore. It's totally different now, right? And I think some of you that raised your hand with OpenClaw and Hermes, you already know what I'm talking about.

9:32Eric Siu:But you as an individual using this is not enough. Because you're actually depriving your teammates of being able to grow with you as well. And you want to bring everyone along, ideally. That's what I learned. And I was like, oh man, every day it's like so cool. But I'm like, but I feel like if my gap is getting, if I'm growing like this and my team's growing like this, that gap's going to get wider and I can't let that happen, right? And you shouldn't let that happen to yourself and also to your teams. So that's what I'd say. That's what kind of this agent fleet looks like. And then what happens afterwards is, as you start to learn this, like I just mentioned, you start to give every person on your team a fleet.

10:10Eric Siu:So this paid me to lead. They might have Arrow, which is good for social media, Flash. They might have the Oracle one, which is good for analytics. And they will start to customize it for themselves. And I mean, look at your desk at work. Look at your home, look at your car. Don't you like to customize and decorate it? Everybody likes customizing their own, right? I think some of you who are old enough, I certainly remember this, Tamagotchis, right? Pretty cool, right? Yeah, right. So once you start to give everyone the ability to customize it, people feel more bought in. Every single person that we've given an individual agent to, we've been testing this, they're like, wow, I feel like I'm reliant on this now.

10:48Eric Siu:I feel like I can't live without this, right? And so again, this is something that everybody can build on their own, right? The reason why, again, I call that OpenClaw and Hermes, these types of agents, is these types of agents allow you to do this type of stuff and unlock the ability for you to have this unified brain, this world brain that a lot of people are talking about. And it seems scary when they're talking about it, but once you start to implement it, it's really not that bad, okay? So again, you have an ambient brain for your company, Then you have a fleet for yourself and then you make one for your company.

11:21Eric Siu:And then you start to scale a lot faster. So in terms of, maybe there's three levels right now to all this, right? You have do-it-yourself, okay? You have done for you and then done with you, okay? So on the very left side here, you probably can't see it as well. Do-it-yourself, maybe you're using ChatGPT, maybe you're using Claude, maybe you're using Claude Cowork and you're just going back and forth with it, right? done for you is when you're working with these specialist agents, okay, where you have these end-to-end workflows. And let me explain what that means. So an end-to-end workflow is when you have a human input.

11:57Eric Siu:That's the first step. I'll go over here. Human input first, then the AI does thinking in the middle, and then the human reviews the output, right? And you think about a lot of the workflows that we're doing at work, like these things are all just workflows, right? So then what happens is you build a bunch of these workflows together, and then what happens is you you start to create these specialist agents that can do a bunch of things around SEO or conversion rate optimization, paid media, ad creative, which we have like one that spins up a ton of ad creatives. And I'll show you an example in a second.

12:28Eric Siu:But once you're able to do that, then you start to create loops, okay? So the way I look at the org chart of the future is you have specialist agents and then you have loops. And a loop is basically a bunch of different tasks that you might insert agents into. And then you just have people, a human at the top managing these loops. So the future is really agents, loops, right? And workflows. And if you have an org chart that compounds that way, you're just gonna grow a lot faster. And the thing is, anybody on my team that is using this stuff right now, work when I ask them and I'm just like, hey, don't BS me.

13:03Eric Siu:Are you having fun, right? I'm like, more fun than I've ever had. How hard are you working right now? Harder than I've ever worked, okay? The only thing I'm concerned about right now with these AI pill people on my team is actually burnout. That's the only thing I'm concerned about right now. Now, on the flip side of that, the people who aren't AI-pilled right now, it's like, oh my God, it's like trying to drag someone over. And it's, you can't, you can bring a horse to water, but you can't make them drink, right? So that's the challenge that I have as a business owner right now. I wanna bring everyone along, but not necessarily everyone's gonna wanna come along and you're gonna have some tough decisions that you might have to make, right?

13:38Eric Siu:And, you know, as noble as it might be that you wanna bring everyone along, you just can't. But I will say the ones that are super AI-pilled right now, it's such a pleasure. It's such a joy to work with these people. I look forward to coming into work every day. I don't dread it anymore. And you can apply this to e-commerce. You can apply this to any form of marketing. You can apply this to services. List goes on. So the way we see it is one human plus five agents is a 10X team member, okay? Maybe even 50X, maybe even 100X or so. But we're just looking at leverage ultimately because when you have the ability to call on this ambient brain, whatever you want, you're actually compressing time.

14:19Eric Siu:So next slide, I'm going to show you what I mean by that. But just think about it. If you're with an e-commerce company right now, you have to, maybe a year ago, two years ago, if you wanted to get a data pool from your data scientist or your data analyst, what is that? Three days, four days, a week? A week's pretty effing crazy to me, right? Because a week is 2 % of the year, right? Two weeks, 4 % a year, are you kidding me, right? So like, and they don't want to do the data pool for you. Like you don't want to ask for the data pool. Like it's a pain in the butt, right? So you might as well just ask the thing instead of waiting for a week, you get it in seconds and you compound it on all the things that you need from a context standpoint, you're just able to move a lot faster, right?

14:57Eric Siu:So I'll show you an example here on the screen. So this is inside of Slack. This is me asking for data, okay, to Oracle. And maybe you can't read the text, but you can get the slides afterwards. And here's like, oh, here's how Meta's performing right now. Okay, cool. Here are the top spending campaigns. And then Christy comes in. Hey, I want to refresh this meta ad creative over here. What should we do? Okay, Picasso comes in. That's another agent that we have that does ad creatives. And then it does more data over here. And then after that, it's starting to strategize with my team. It's starting to say, okay, here's what's happening right now.

15:32Eric Siu:But the most interesting part is when you're able to combine the agents. So boom, Picasso spun up a bunch of different variations based on what was already performing well, right? So you can just say, hey, look at my meta ad account. I want you to double down on creatives that are working. I want you to spin up 100, 200 different variations. I want you to kill the campaigns that aren't working right now. And then let's move on to the next thing, right? And the crazy thing is now, those of you that raised your hand with OpenClaw and Hermes, you know this already, but you can literally, now you have the slash goal command that can help you do a lot of stuff overnight, right?

16:05Eric Siu:It can help you finish tasks and you can go a lot further in terms of just getting things done. And you can plug this in. So there's so many, any of you that have played video games growing up, there's so many different upgrades you can add every single day. And that's what it feels like. At least for me, it feels like I'm playing a video game. And I hope it feels that way for you too once you start to do this. All right. Some examples here. You can simply ask it, oh, what's our ROAS this week? Okay, and you can see the performance over here. Then you can go deeper. Okay, what's our ROAS? which creatives, again, which creatives should we double down on?

16:37Eric Siu:What are our competitors doing right now? Okay, what does their ad library look like? What are the longest standing ads that they had running for a long time? Approximately, what are they spending right now? What are some other ads that are working in maybe adjacent industries right now that we should consider? Okay, let's spin up 500 variations, right? That is all possible now. It wasn't possible before, but again, when you're able to compress, let's say, 22 weeks of work into 22 seconds, the sky's the limit, right? So that's why earlier with David backstage, we're just talking about, I was showing him these AI stocks.

17:06Eric Siu:I'm not gonna tell you, I don't wanna give you financial advice, but man, some of these are just parabolic. It was, I looked at one, it's like up 5 ,000 % or something in the last year or so, and these are all part of the AI stack. But if we're all trying to buy this infinite intelligence that we've never had before, I think the build-up, it's gonna go even crazier, right? And so anyway, all that to say, if you can do this for your business right now, what's gonna happen in a year? Scale it to 200 variants, scale 500 variants, whatever it is exactly. And here's just an example of if you put a prompt in campaign out, hey, I want more landing pages over here.

17:39Eric Siu:I want nurture emails spot up. And so with one of these single brain agents, we it literally, it will put together your cold email infrastructure, okay, or you can put together your nurture sequence. So what it did for us was it literally bought 100 email domains. It came up with a naming vernacular as well. It also put together the sequences and it recursively self-improves, meaning that it constantly is testing itself over time, right? And so that is a closed loop, right? We want to think about once you start to use this, you want to think about how many closed loops can you build? Because a year ago, two years ago, three years ago, us humans were working in open loops.

18:19Eric Siu:Oh, Bob. I know your name's not Bob, but Bob. How's that status report going right now? Mike, how are the paid ads going over there? Anna, creatives, how's that going right now? How annoying is that? Nobody wants to do that, right? Nobody wants to check in with each other. By the way, I'm in the client services business. I even talk to my clients too. When I show them this stuff, I'm like, let's be honest. You don't wanna talk to me. I don't wanna talk to you. If we just do good work, who cares, right? And they're like, exactly. And they're like, right? Literally, I did that. So I'm like, I don't usually get on the client calls, but I've been hopping on them recently.

18:53Eric Siu:And they're all like, yeah. And that might like, you also have to help your team understand this too, right? Like, um, because they're so like, they're, they tend to overwhelm themselves. But if you start to get them to see the light in terms of how to work this way, my God, works would be way more fun. By the way, this is an example. When, once you have an ambient agent going, um, I was like, oh, you know, man, our, our costs have been kind of going up a little bit. Can you make me a CFO agent? And so it made me a CFO agent and looked at all my costs. And it's like, yeah, you're being really stupid with your spend here in X, Y, and Z.

19:28Eric Siu:Within three days, so he found me 500 grand in savings. I cut all those costs, 500 grand saved. And some people are like, oh, Eric, how do you justify the$12 ,000 a month you're spending on AI tokens, right? Well, this is how, because this is the ROI that you get, right? By the way, I'll give you a fun tip. You should probably write this one down. How many of you are spending a lot of money on tokens at your company right now? like over 10 grand a month? Okay, 50 grand a month, 100 grand a month, anybody? Okay, well, it's probably gonna become that at some point, right? It's because we're paying for this intelligence.

20:02Eric Siu:It's unreasonable for us to not pay for this, right? So my token costs, and maybe we'll talk about this in that marking school recording that we do in a little bit with Neil, but my cost got up to$12 ,000 with Anthropic Spend, right? And I'm like, man, like I'm having fun and sure I can argue that I'm getting ROI, but that's still crazy to me. So if you want to drive your token cost to zero, maybe not zero, but a lot cheaper, all you have to do is because right now I'm paying for Claude. So I paid$200 a month. And then I started buying more and more open AI accounts because ChatGPT 5.5 is actually more for right now to me is more superior than Opus 4.7, which is Claude's best model, right?

20:44Eric Siu:And all that happens now is for 200 bucks a month, they're very generous. You can eke out like five grand 10 grand 15 grand in tokens and so once i run out on the first one guess what i switched to the other one 200 a month and i'm just gonna keep doing this 200 a month until everyone starts abusing this and then the party's over so maybe maybe i shouldn't have told you guys this but i figure someone's gonna do it so i might as well just tell you but you're welcome you have gotten an roi on from this conference already um so so anyway um agents are the new buyers right you have optimize accordingly this is kind of going into if you're using your ambient agents you have to think about how you can optimize like how many of you have software companies anybody or work at software companies okay not many okay well so i look at it this way you have uh you have d2c right we have b2b you now have b2a which is business to agent because everyone's spinning up all these agents now and you can see the way i'm treating my agents it's almost like my chief of staff my assistant and I have multiple ones, right?

21:47Eric Siu:I just want them to do the maybe the buying actions for me. I don't want to do the research. And so, you know, Stripe recently just came out with the ability for agents to purchase Shopify has their I think it's called UCP. So everything is kind of going into Hey, how do we make it? How do we enable agents to purchase more? Cloudflare, for example, you can actually buy Cloudflare infrastructure. And so I think about this from an SEO background, Oh my god, I can spin up so many different landing pages. It's so many different subdomains. I can buy all these things and just tell my agent to do it.

22:16So when you think

22:18Eric Siu:about from an SEO standpoint, you're going to want your agents to optimize for that. And here's one thing I'll say. If you have a software company, for all these agents are crawling all the API documentation, they're crawling all the MCP documentation, right? Yesterday, and by the way, you guys can reach out to me for this. I have a contact at Shopify that texted me yesterday. He's like, you know, I lead the Shopify API slash MCP. I want to talk to a lot of e-commerce companies, right? So if that's you, feel free to hit me on Twitter or Instagram, and I'm happy to put you in touch with him. But that's literally what they're doing.