I Just Built A Full Fleet of Revenue Agents Inside My Computer (Full Breakdown)

**Eric Siu** (0:00)
The companies winning with AI right now are not using better tools. They are running a completely different playbook. Most businesses are still experimenting. The ones pulling ahead already have agents doing real work, real systems that do real tasks with credit cards and everything. I have 12 agents running inside my company right now. They handle sales, content, SEO, and recruiting without my team touching it. One of the agents, the finance agent, even saved me 500 grand the first time I used it. The gap between who gets this and who doesn't is opening fast. Here is the exact playbook we're using to win with AI this year. So first, let's talk about the roles that agents actually replaced. What happened when they actually took over? So here's a very simple example. This is an agent that helps us on sales across the board. So literally, what you're looking at on my screen is me asking it to pull up information on a cold emails campaign. It will actually create all the cold email sequences, it'll come up with all the leads, it'll scrub the leads, it'll de-duplicate the leads as well, and it will send it on a sequence, and it constantly iterates over time. It will constantly self-improve. And so this agent we have, that's in our stack, we have a stack called Single Brain, which is all of our revenue agents, right? And that's the company that we have that runs this stuff for clients, right? So within Single Brain, our team can, we'll work with these agents and ask these questions, and then the agent can handle the work. We're talking end to end. So typically what would happen is, sometimes you might need to go back and forth with it, but once you have this set up, you can have it set up as a repeating cron job, or even like a repeated trigger. So if the numbers start to fall down, it can recursively self-improve itself, right? So the good news is, once this is set up, you're going back and forth with your agent over here, it can improve on actually generating meetings for you. So we've actually generated recruiting meetings, we have generated sales calls from this, and these sales calls are legit sales calls, and we're doing this for customers right now. But this is us just working with the agent inside of Slack, and this is our team collaborating with the agents as well. So not only are you becoming more well-resourced with these agents, but your team is becoming more enhanced with these agents as well. Now, not only that, we have an agent known as Flash, which handles content repurposing. So it helps with sales, that's cool, but it also helps with content. This piece of content over here, I talked about how to practically deploy Jack Dorsey's World Intelligence, which is exactly what we do at Single Brain with these revenue agents, 348,000 views on it. And this little diagram over here, this Single Brain piece, all this concept, I'm just explaining how we do it at my company. This is generated using a workflow that we have inside of Gemini. But this entire piece over here, I might have helped edit it for 15 minutes or so, and you should have a human in a loop here, but this is a task that continually repeats. So what happens is we'll look for trending topics on X, and then it will think about all the interesting ways in which I work with my agents, and I'll come up with these topics. And it got 348K views here. Not only that, it got another 150,000, because Gary Chan retweeted one of my other posts from Y Combinator. Not only that, it generated, we're talking about a lead from a multi-billion dollar company, actually two multi-billion dollar companies, is interesting, right? Because you're not just talking about generating views, you're actually talking about generating pipeline, and the agent is helping you do that. So I'm not saying you should just let your agent YOLO and do whatever it wants all the time, but you should let it get you to the point where a human needs a review, and then you're okay to publish it, right? That to me is end to end, and this is what's happening right now. And obviously, you're going to continue to improve these over time as well. We have another one called Oracle. The SEO team, they've actually built their own little app called TiderClaw, and Oracle is thinking in with TiderClaw, and the whole idea here is that Oracle can work with the team inside of Slack, answer whatever questions, and then use it for their SEO workflows. There's trade-offs here. These agents work pretty well. I mean, probably 90, 95% of the time, they're working pretty well, but 5%, 10% of the time, they're broken. You got to reset them, you got to rework them, you got to adjust the configuration. They're not always going to work, which is why you need someone managing this. So that's probably the downside around it, but I think the trade-offs are worth it. What surprised me is once I added these agents into Slack, it changed the game completely, and we'll get to that in a little bit. So let's talk about how we actually train the agents. So the most important thing is when you're setting up these agents, you want to set its own soul.md, so give it a personality, talk about what you like, what you don't like. Memory.md, so for persistence, obviously you want it to remember things as well. And lessons.md, too, you don't want self-correction after mistakes, right? So you don't want it to just, you make a mistake and you keep making the same mistakes over and over. You want lessons.md to remember the mistakes and also help you not just call out the mistakes but cover them, right? I would say that the more time you spend giving it feedback, the better it's going to get. Obviously, I think with the first 30 days or so, you want to spend a lot of time training it, but you shouldn't think like you're done after that. The training never stops. It's almost like employees that you have in your company. You never really stop training them. You want them to continue to get better over time. One additional thing I would add is expert panels that grade output. Okay, so if I'm shipping content, I would put together an expert panel of copywriters or landing page experts or persuasion experts and say, that's how you can get better output. And then you can tweak it. Obviously, you should just take it at face value, but these are a couple of tricks that you're able to do here. And by the way, what I would actually say too is, we have open sourced a lot of our AI marketing skills, and there's about 1700 stars on GitHub just for this repo right now. It's completely free. You can go to singlegrain.com/skills.