Run AI Agents Like an RTS, Not a Chatbot

Chris DuBois0:00

You run agents like a RTS game, not a chatbot. Why does that matter?

Bryan McAnulty0:57

Yeah. So you probably saw me mention this on my my ex account, the idea of yeah, thinking about how do you operate with these agents as like a real time strategy game rather than the chatbot. It's actually interesting. I think since I posted that, that was maybe like a year or so ago. The way that I work with them has changed because I think there's a way to really like burn out quickly in using these things because it's just they're doing so much so fast, but a lot of people I think the the overarching idea is that a lot of people are thinking too small with how they're operating with these tools, both in terms of like how much they could have being done for them at once, and also like the scope and the the breadth of like how big the tasks are that you're giving to these agents, and one of the reasons is is because they're they're getting better so quickly at like the the length of the tasks that they can accomplish. And so, like a year ago, you were able to say like, "Hey, I'm going to have this thing run for like 20 to 30 minutes. It can do a lot for me in that time. But then it would kind of like get stuck or give up or like that was the most I would do. And then you'd have to go in there and clean it up and prompt it some more to work on those things. Now we're having these agents that are working for all these hours or even sometimes days towards a specific goal. And so, yeah, originally I was thinking, okay, well, how can we? How can I? Is the way that I'm going to become more productive than what I saw other people doing at the time was like you're staring at a a chat window while your your coding agent's working and you're not doing anything. That's not leveraging AI, right? That's just not working. And so for me as as a founder, I'm I'm not trying to like just replace my job and have something do the work for me. I I want to have a greater output, right? And so my thought was like, okay, I want to have multiple of these agents working, and I'm going to bounce between what's the next plan or next thing that I can have the next agent start working on. Now that is the the path to a burnout in in a sense that I talked with so many developers at the beginning of this year. Everybody was feeling that, like, wow! Like, I'm getting so much done, but also I'm just like so fatigued by like the the end of the afternoon, even because like my brain has to process all these different ideas we're going through. And so today, I believe that the way to work with these agents is kind of there's like three different workflows that you should have one is the the kind of background like recurring automation loops. Two is the the big like massive scope tasks that it's going to spend many many hours on, and then three is like the the kind of back and forth that really needs like your taste and ideas, and that quicker feedback loop for you to be more involved in, and I think having some things, like some tasks and things, in each of these buckets is the way you have to think of it. And if you want, I can describe like what what's an example for for each of those.

Chris DuBois3:55

Yeah, let's uh, yeah, actually, let's start there. Yeah, so

Bryan McAnulty3:59

a recurring task. The kind of things that agents are really good at are like these these tedious things. These things that don't require like the human ideas. I think the the ideas and everything that the the the way that you do things it has to come from you still, not the agent. But the the things where you have to do research or create a report or the these kind of things are like tedious, recurring things that can be done by the agents. And so, an easy example of like something that we do is like we have a weekly report that the agent shows to me of just like, hey, here's what got done in the last week in this piece of software that we're building. Another one is like, okay, from the things that we released in the last week, go and check it for check those things for security issues, and like those are things where an an agent can be positively contributing to it without you really needing to to have to do anything, and so this is where now we're starting to see this kind of like these AI like factories of like your agents are are constantly working in your business and you have not only your your human capital of your. But also this token capital that you're leveraging in some way. Second part is these these bigger tasks that, like I'm talking about, we had in the last week our agent agent platform Latch Loop. We had an agent go through and create all these competitor comparison pages about our software and all of our competitors, and we said like, "Hey, your your goal is you got to make 100 of these. We want each one to be like well researched and thought out. It actually ended up doing about 150, and it found these thing, these tools that like either we never thought of or we thought like the the competitor is a little bit smaller. It's not really worth the human time to go and like do all that research on it, but it spent the time and worked for like over seven hours of like researching each one using the format that we talked about of building out each page, and it did all that work. And like it would have been like months of human work for somebody to do that. And now what we put into that initially like was a relatively like comprehensive prompt because we wanted to just be able to hand this off, so we had to upfront communicate like, hey, here's what's important, here's why we're doing this, here's the way that we want you to do it, and and kind of give our input on the ideas, but then like the tedious research is what it could do, and then the third bucket is like the the stuff where you need that that bigger feedback that quicker feedback loop so that's like building a feature in the software maybe apart from like the the larger technical stuff that that the foundations can be laid out by the agent when you're trying to actually design well like how would a human want to interact with this that's still where it comes back to that the the human like taste and ideas are really important,

Chris DuBois6:42

right? Okay, so many different things that we can get into. For I want to touch on that comparison page idea because so there's like multi multiple pieces here that I think we're enabled to do that that we couldn't do before and that we might not have wanted to do before, where, yeah, the human cost, right, of having someone go actually do this work-it's like it's not worth it. Like, let's write it off. There's also the from a positioning standpoint. I don't necessarily want to create a comparison page for every single competitor if, like, the people that I'm targeting aren't thinking about these competitors because now I'm just giving them options. I'm inviting comparison, but because we live in an age of AI, someone can go to AI and just if if they mention that other tool and say what are the alternatives because it has the context of this person, it's now going to serve you up as an option, and so like it doesn't necessarily like the SEO ramifications that it could have of like introing other options now kind of go away, and we get to benefit from having this. And it's like it's crazy. Yeah. Yeah. Well, I completely

Bryan McAnulty7:53

agree with that that approach. That a year or two ago, I would have thought like, do I want to let my like somebody who doesn't know about these competitors, but now it's exactly that: that you want to design content for what does the AI need to know. So, like the content, I think people should be publishing in their marketing is not like anything that's in the AI training data. That like just forget about it. You want to be publishing the data that it needs to know. So if if it's going to go and research something for somebody, you want to make sure that AI has all the information of like why it should compare you and how you're better in which ways,

Chris DuBois8:26

right? And all the AI needs is like there's like as it's doing its initial research. If it even gets an inkling of like these competitors through you, then it's going to search. Okay, well, what are the alternatives and and all this? Like it's going to pull in your information because you're the only one writing about it, anyways. It's awesome. The one of the other, I can't remember who who was talking about this, but they had essentially created a an agent that would look at all of their content, and anytime the metrics went below a certain point for you know how many visits you're getting to this this blog or whatever, it would then run an update on that blog, and it would you know would find it would look at the top articles that it's a it's competing against for the keyword, figure out what do we need to add, and then it would just process that. A human looks at it, approves, and now all of their content is continuously getting approved, so they don't get that like drop off that you'd normally see with SEO, yeah, yeah. But huge use case for for agents, right? Yep. How do you? What's the right way to frame this? One okay. One of the piece of advice that I had heard was think about agents as like a workflow artist, almost like if you know what the path is already, it's like you can get the agent to do that. Now we started introing some other like challenges, goals, like things like that that the agent can actually think through as it's going through this. And so it's not just setting the workflow. It's like, hey, you also know what my rocks and my my targets are for you know q3, and so it can take those into account. As is building it, I guess. How are you thinking about like setting a an agent versus just like a strict workflow of do X, you know, Y, Z?

Bryan McAnulty10:10

Yeah, I think like last year with the the models that we had at the time, the the best way to get benefits from these agents was more like constrained workflows of like, hey, you're going to do this, and then some other agent like you're, you have to do specifically this. But now the the agents, the harnesses, and the platforms, everything's a lot more capable. But I think you're exactly right. Like if you know exactly what has to be done, and you can define that well, and communicate that clearly, then the agent can actually help you to do the work. One of the ways that that I kind of express this is like all we're really all doing is like communicating intent to direct attention. And so, like if the agents, sometimes we'll see the agent doesn't quite get something. Like it'll it'll do some of the work, but like something's left off. And so what we have to do is just be able to design ways that we're communicating it to to let it know of like where does its attention need to be directed, and that basically like these these features where the agents have the the goals that they can work towards and things like that, we're just essentially telling the agent like, hey, we need to spend attention on these items as well. But yeah, the way I'm thinking about this overall, because I think you made a really good point. I saw somebody said a quote that was similar: "Is that you can outsource your work, but you can't outsource your understanding. So you have to be able to understand what you're asking these things to do, and and and what you have to potentially where you need to review things and where you don't. But also, the great thing is you can talk with the AI and have it help you understand if you if you don't already. But then, when you can ask it something clearly for it to actually do the work, the work part can very much be done by these agents nowadays. And I think even for for the bits that can't, you should begin thinking in a way as if assuming it's going to be coming very soon. I jokingly but not jokingly said to my team that I'm seriously considering to say that like nobody is allowed to do work anymore. That what you have to do is actually coming up with the ideas and then directing these agents, where the actual work part is being done by the agents. But this is a very different way of working for not so much a founder, but definitely for the average employee, and so I think there there's going to be have to be a lot of training. There's going to have to be this big shift in people being able to stand understand not only the capabilities of these agents, but like how am I going to work in a different way to come up with these things. And I wish I I had the resources to be able to say like, can somebody go do all this research for me, or something like that, so you can work on these kind of bigger picture ideas and goals.

Chris DuBois12:44

Yeah, I know you said you were half joking about it, but I love that as a like a thought experiment to just like

Bryan McAnulty12:51

yeah yeah

Chris DuBois12:51

for someone to be able to sit down at their desk and say like okay well I have to do all of these things every day, and because I have to do it every day, that means there is a like defined input that needs to go in here, and there's a defined output. So I should be able to instruct someone, someone else, how to do this right. The same way you would train a new team member, but getting yeah, getting the team who's like the specialists in whatever task to be able to think that way and just keep improving it. Like you're the 5x force multiplier that AI could give you actually becomes like a six or 7x, and now your entire team is just operating so much better.

Bryan McAnulty13:28

Yeah. Well, so actually, I want to talk about two two kind of ways to branch off this. Let me know what you think. Is that? Yeah. One is how teams have to think about their data and like what what what what is important and what should they own in this age of AI? And then two is like how teams think about like what what becomes valuable in the future. So what I would say is if we imagine a year or so into the future, and everybody's got these magic AI agents that we press a button that does all of our work. What makes your business any different from your competitors? I would say that it all comes down to how you do things. The like your processes are what is valuable, and like that is something I believe that you should be owning. Just like you own the SOPs that you've you worked on to to instruct your employees, right? There, like I think any like AI agent platform or tool that you're using, you should make sure that you're retaining ownership of that information. This is not even about like if the the platforms can train on your data, but it's that like you have to actually have access to that data to be able to use in whatever way that you want. An example of this that now now I'm not sure how much of this is intentional. I'm not sure how much of this like may may change in the future, but there is this product that Anthropic recently released, Claude Tag, and it's like a Slack bot. You can mention like at Claude in your Slack, and on one hand, it sounds like this is a. Way for employees to understand very easily how they can interact with these agents because you just at mention it just like you would a teammate. But the problem is that while the agents can do some of the same work as humans, the way that they work is different, and so often you want to be able to have some kind of paper trail of like, okay, why did the agent end up at this result? Like, what happened in between, and with a tool like Claude Tag, I don't think you can actually see that. But what's worse is that the process is learned by this agent over time of like how it's going to perform the work that you're asking it to do. That is a business asset like that. That's how you do things, right?

Chris DuBois15:37

Right.

Bryan McAnulty15:37

And so you should own that. But with Claude Tag, as far as I understand, you can't even see that. Forget about owning it; you can't even see

Chris DuBois15:45

it. And

Bryan McAnulty15:45

so, I think you should be using an agent tool where you have ownership of that, and you can go choose use any other agent harness or platform or whatever with that data. Because really, if the agents do the work, then like that's everything of like the way that you approach things.

Chris DuBois16:00

Right. I agree entirely. The my model for positioning, like an agency, is that there there's 11 different levers we can pull on, and we got our positioning levers, which are you know this is how we're we want to show up, like who's our market, who all these things. But the structural differentiators, like how do we deliver the work? What are the things we refuse to do, like all our methodology, our IP, like those things are the critical component that most agencies overlook. And when you stack those with like some market, like your market focus, and maybe your economic model is also different. When you put all these together, it's like it's actually really hard for someone to copy what you're doing exactly. Yeah, to exactly what you're saying, if someone can come in and copy my entire delivery model with agents, now they don't even have to know my model. They just have to let their AI do the same thing I'm doing. It's like I just lost a huge competitive advantage to be able to do this. Now, yeah, we blend in with everybody else.

Bryan McAnulty16:56

Or if you, or if it's just that, like, let's say that the vendor, like Anthropic or whoever it is is not going to like take that and use it somewhere else. But even if they just own that and you can't get access to it, now you're you're stuck with that vendor. Like that vendor is your your business, you know, and you can't go hire a different team or something because it's all stuck in there. And I I would I feel like that's a hostile environment for for a business owner to put themselves in, but but what what I would say for like agencies out there, have you heard of earlier this year? People were talking about the idea of software as a service has to become service as a software.

Chris DuBois17:35

Yep.

Bryan McAnulty17:36

So my thought is that agencies also have to become service as a software. Just everything is basically services and software. Meaning that before we had the these tools and these things that we were trying to do to hopefully deliver some kind of outcome to the user, in the case or the customer. So, like in the case of software, nobody wakes up out of bed in the morning and says, "I can't wait to use this software. But like that was the best that we could do at like providing some kind of forms that if you press the right buttons, hopefully you accomplish your goal. Now with AI, the AI can just deliver the outcome to you directly. And so I would say if you're an agency owner and you're thinking about like, okay, how are we gonna like what is the path forward and how we utilize these tools? I'm really optimistic about it, but I think the wrong way to think about it is, oh, this is going to help save us a little bit of time or save us a little bit of money, and it's just like the the checkbox. The way that you should be thinking about it is, how are we going to use this to create a like 5x, 10x, 100x outcome for our customers versus what we could before? And if that's the way you're thinking about it, that's how you're going to win versus your competitors who are like, oh, let's try to save a little bit of money by doing this AI thing,

Chris DuBois18:45

right? Yeah, the I hadn't heard the service as a software really anywhere until I I took on a challenge of saying, could I build an agency entirely ran by AI? Because the only reason I'm not running an agency myself because I don't want to deal with clients. Sorry, any agency owner listening understands that. But the I was

Bryan McAnulty19:07

originally a small agency before becoming a software company. Reason I became a software company was the same.

Chris DuBois19:13

Exactly, and when I was running an agency, that was always one of the thoughts. It was like, oh, I could go be a tech founder, like, and feel like they got it easier, but I built an SEO agency, like air quotes for anyone listening, that would essentially do all the content, do their Google listings, all that stuff by just someone subscribes to it, and it's now it's trained on myself. I had a buddy who specializes in these areas, and we brought in a couple clients who were all all paying, and they were getting the service. It was getting results. Just didn't. It was kind of a distraction from what I'm normally doing. But it was like the test was proven out that this can work. There can be a fully AI agency that can come in, and like I know there's one right now that's a I can't remember his name. They got a bunch of funding. And so now that they're showing up on every YouTube ad that crosses my way, but they're they're making even bigger promises than I'm going to manage your SEO for you. They're you know trying to do everything, but yeah, to to go back to what you were saying, like even if the competition doesn't have access to this, if you if you also don't have access to this. It means you've lost an edge. So yeah, that's a

Bryan McAnulty20:25

yeah. Well, I I think like if we look back at like January, February, like everybody's talked about open claw and all that, and I think some people got interested in like discovered like okay, this is what an AI agent potentially

Chris DuBois20:39

is. Yep.

Bryan McAnulty20:39

If you had a bad experience with things back then, I would encourage you to to try again now because back then I would probably agree with you that like trying to babysit those things and like and fight it to try to get it to work. I rather just do the work than than be stuck fighting with the thing. But now the the models have gotten better, the the harnesses, the tools, everything has gotten better. That now, like, I think people will be mind blown when, like, just think bigger than whatever you're thinking and ask the AI agent to do it. And I think you'll be surprised on in what you get. And when you see that, I think that's like what flips the switch to realize, like, okay, we we just can't work the way that we did anymore because how like the amount of output that you can potentially get from these different agents, it unlocks this whole new possibility of what you could actually be building.

Chris DuBois21:30

Right. So let's one of the biggest challenges for getting into this and like thinking about agents and stuff is that most most agencies specifically don't know what they should be automating first. What do you have a model, something for like? All right, let's start here and then build build from there.

Bryan McAnulty21:49

Yeah, I think agencies are in a great position. I think like any like operations heavy businesses are in a great position to be able to leverage AI well. So I would say like all of the and like you can start small on like one piece here and there of how you're going to use these things, but like all the like the tedious like research tasks and things like that, like getting AI involved in that, and like there's like if you were like a a branding agency or something like that, and you would say like okay, well our like very deep understanding of a client and how we're going to translate that into coming up with a brand for them, like that's very very important for us. But undoubtedly, there's going to be a good amount of research that could be done by AI. The final the final nuance in in your understanding of how to make that turn into a brand that can be done by the people, but the everything leading up to that can be done potentially with the help of AI. So, like some examples, because I think there's also been a lot has changed in terms of like the technicalities of like what does it mean to get set up with these agents, especially if like the last thing you tried was something like Open Claw or something like that a while back. Nowadays, you have tools like the ChatGPT work was just announced like last Thursday with GPT 5.6. I'm really excited about that because they kind of redid the ChatGPT app. They put in Codex, the coding agent, is now built into that as like like ChatGPT Codex, and then there's ChatGPT Work, which is like the general knowledge worker agent, so I think probably agencies will be using more of that. But I'm excited about this because my agent platform, Lash Loop, that's available at lashloop.com, and we are also like general knowledge worker agents and coding agents. And the thing that we've been doing for like over a year has been new chat is not the way that you start off. It's new task, and like our interface is much more like a project management system, where it's like here's the the task document, and then here's the activity on the side of the humans and the agents involved in it. And so OpenAI is helping kind of pave the way and like educate the market for us. But then besides LatchLoop and ChatGPT, there's also Claude Claude Coworking Claude Code. I would say like these are the main tools, and they are now making it a lot easier to get started. So if you have not touched these things at all, and you're wondering like, okay, what do I do as an agency to build an agent without having to get a developer involved or like spend all this time on setting something up? A quick exercise that I would say is like inside a tool like Latch Loop, you can go and say, okay, I want to create a new project. It's going to be a general agent project. If you have like a repository or something that you want connected to it, you can do that. But you can just start say, okay, create a new one for me. And then what I would do is I would start it off by giving it a task and saying like, hey, you're going to be a marketing agent for our team. This is our website. You're connected to our project management system. Here's where you can find some of our SOPs and stuff. Go and learn about us, and and and learn about what you're going to do. And so it kind of kind of can onboard itself into building this knowledge about your business, and then after it's done that.

Bryan McAnulty24:59

Now it has some kind of baseline, and you go and like give it its first task and say like, "Hey, look up our recent SEO data and figure out maybe some areas that we could improve, or like go and like figure out like the competitors that we have and like help draft out these competitor comparison pages, things like that. So it's it's much easier to to get started and see something happening than it was like six months ago. Where it's like, okay, you got to set up all these folders and commands on your computer and all that,

Chris DuBois25:27

right? So, all right, I want to explore an idea with you. You you might have thoughts. If not, we can we can move on. But

Bryan McAnulty25:37

sure, the

Chris DuBois25:38

same way that we would look at like hiring people, right? Is there value in looking at hiring a specific agent? So basically, compared to like saying, "Hey, I've I've given it some guidelines to be able to accomplish these tasks, versus I have created like a senior marketing specialist, and they are trained on SEO and maybe social marketing, and like that's it. Like we have a tight constraint, so that it just gets better at these skills. Is that? Do you feel like that's something worth exploring for for companies where it's like this agent has a very tight parameter, so it gets really good at this versus having to spend the token cost of letting it like think through. Okay, well, what do I need to do next? What are what skills do I need to incorporate here? Like, what's what's the best play from your perspective?

Bryan McAnulty26:27

Yeah, it's a great question. I think it it's like this whole new world, and it's changing so fast. And so it remains to be seen, like if these very specialized vertical agents will become so good in the way that if it's not like directly tied to what your business does, then it makes sense to be using them plus your internal stuff. I mean, I I envision in some way a situation like that because I have a software as a service company, Heitz Platform. It's probably what we're we're most known for. It's an online course creation community building software. Basically, we help entrepreneurs and creators build these online knowledge businesses. Inside that, we have Heights AI agent for creators to use. But we're releasing like the Heights AI three version of that this year. And I've been thinking like if everybody's going to also have their own kind of personal agents or their own general agents, like how, like how does that interface? Like, where are they going to use our like specialized vertical agent, and where are they going to use their own? And I do think that for certain things, like if you have a business, not only are your processes valuable, but also like the data that you have. And so, like for us as a service provider, we can take all this like knowledge and all this data that we know from helping now more than 10,000 people on our platform of like how can we help and serve this other person, and so like they're trying to accomplish the goal. We can now provide that service in like a more specialized way, but at the same time, if you are like an SEO agency, and then someone says, "Hey, we built like the SEO bot or whatever. Use ours. I would never do that because, like, that's that's where you're supposed to specialize. And if you're doing that, you can't win against your competitor, right? So, if it's like if you're you're some kind of other company that you know next to nothing about SEO, then yet probably like go and and do that SEO agent instead of trying to to figure out your own. But yeah, if it's like core to your business, then that's where you want to have your own agents,

Chris DuBois28:29

right? It would make sense. It's kind of like don't outsource your main thing, right? Like that's been general business advice for forever. No, it's just the same. Don't let AI be that outsourcing partner. So I want to go back to something I thought of very early in our conversation. So you're going through this, but have you ever heard of the DIKW pyramid? There's probably a name for this, but it's essentially for how we process data into turning it into wisdom. But so DIKW is data, information, knowledge, wisdom, and as you're, in order to move it into like it's a pyramid, right? So wisdom's the smallest section, but the in order to get data to turn into information, we need the context, and so like we've seen that with AI, like context has become king. It's like I need to give it all of this so that it knows what I care about. To move something from information into knowledge, it then needs meaning, and so this is where humans have to start providing a little more thought. It's like, well, why do we care about this, right? But then for knowledge into wisdom, it's like we need insights, and so experience is really what drives this. Like we have pattern recognition, and even there, AI is starting to get get good at this. Do you? How do you recommend like applying AI to move something from essentially data to wisdom as fast you can, or is that even like the game? Should I just be thinking about like how do I get it to knowledge, and then I worry about the wisdom piece?

Bryan McAnulty30:00

Yeah, it's a great question. I think that again, like the ideas, the taste that has to come from the person because like the AI like doesn't understand what a human actually cares about and why, and like the the nuance and how you would do something. But what the what has changed with these AI tools compared to like a year ago, is now the the AI agents don't just need to be given this big thing you like copy and paste to say like okay here's all the the context of my business, but now you can give them access to all these tools and then they are good enough to go out on their own and figure out the things that they need to do whether that be in like your codebase whether it be on like researching on the web, going through your project management system, or like a combination of all your tools, and so I think they can build up that context very well when they're given access to the right tools. So yeah, it comes down to like kit, like do you have the agents that have access to the right context? And I think for many like smaller businesses, the issue might be that like is that context stuck in all your employees' heads and your head and not written down somewhere. And so, if you if you don't have like clear SOPs and like things like that defined somewhere, it's probably good to make a better practice of like getting those things documented in your company. But yeah, then in a way that AI can kind of onboard itself to get to some certain level where it can start contributing. But again, yeah, the ideas should come from you. Like people talk about a GPT 5.6 just came out. We have the Club Fable came out recently. Yes, I think that Fable is better at ideas and creativity than 5.6, but is it worth like more than two times the cost if you're using like billions and billions of tokens a month? Like I don't know. For me, that the answer is no, and for me, the answer is like yes, it's more creative. But I'm more creative than it still. So like the idea should come from me anyway. Yeah,

Chris DuBois32:03

yeah. Something I did with as soon as Fable came out was rather rather than using it as like the new model I was going to use for everything was go through all of my current skills and everything else I'm using and improve them so that with like Opus I can get a better a better result now, and it like tore up all of the skills and like found you know added gold standards for everything and like I felt like it was a pretty useful exercise to now get more value from Opus. But on the the context thing of like you know knowing the SOPs and like pulling things out of your head, I think, especially for a lot of small businesses, take like a CRM. Most of a lot of the data in there is just wrong, outdated, incorrect, and that's why a lot of it's really hard for small businesses to adopt a CRM and actually stick with it. Because as soon as the end user knows that the data isn't like true, it's not the most up to date stuff. They just stop using it, right? They they don't care. Is there a way to like, or I guess do you build like a failsafe into this with your agents as to like, hey, make sure that the the data you're looking at is realistic, it's good, it's you know it's up to date. Empower users to provide you with the right data so that we have you can have more context for everything.

Bryan McAnulty33:22

Yeah, yeah, that's a great point. I have a good analogy that may have helped people think about the way that they should approach this because I do think that's a potential pitfall. Like the agents have no good concept of time, so if there's a date there that says that there's date and like they understand that like they're looking at stuff that that's years old, they may be able to take the initiative and say like, I gotta go go do some research to first verify this. But like, yeah, there's lots of ways where things can go wrong. What I would say to people is like, we imagined that the sci-fi scenario of like how AI works is that it's this thing that's always thinking in the background and processing, but then ChatGPT came out, and we realized, like, okay, you send the message, it sends a message in response, and it's kind of done. But now, because we have these AI agents, people, I think, don't understand, like, okay, well, now is it that sci-fi thing where it's always like thinking and processing? And I think the analogy to explain the way that they work technically is actually that the models have been trained to like basically do their best at whatever you ask them to do in the short time window that they have, and when you have an agent, it's the equivalent of that ChatGPT responding with that one message, but you're just taking a software program to say, "Hey, go and run again after you respond with that one message based on like the tool that you used, or the the other thing that happened, and so what's really happening is you have this agent that's like coming to life, dying, coming to life, dying over and over and over, and it's technically a new instance of that agent where you're just saying like, hey, here's all the memories. Like we're gonna shove all these memories into you. You're doing this thing. Good luck, and so like each time you can actually think of it as like a new message or a new agent in each message and step that it takes. And when you think of it that way, and think of it as this thing is trying to do its best to accomplish something for me, it's easy to see how like if you are somehow put in that situation, if somebody says like, "Hey, we just brought you to life a minute ago. Here's this is supposed to be your memory's good luck. You're gonna kind of just do with that what you can and say like here, like here you go. And so that's why when we have situations where like the AI doesn't quite follow through with everything we asked it to do, or like assumed it would do, or maybe the AI like gets something wrong or makes something up, it's because it didn't have the right context or access to the right context in order to do this, and so you have to be aware of that. While the agents can do some of the same work as humans, the way that they work is different.

Chris DuBois35:49

Right. So, one more AI focus question before we start to wind down here. The I've heard the argument against having an orchestrator agent of having the human as the orchestrator, so that there's always a human in the loop managing essentially a team of agents. What are your thoughts on on that versus being able to give all of your instructions to one orchestrator agent that's managing it for you, providing all the reports back and things like that?

Bryan McAnulty36:18

Yeah, that's interesting. I'm experimenting with like a orchestrator agent kind of idea, but it's not something we have yet in our in our product inside LatchLoop. Right now, what we're doing is like every agent is its own project, so it might be a coding agent, might be a general knowledge agent, and you can be in one of those projects and say, "Hey, actually, okay, I need you to go create a task that is going to be run somewhere else, and then they'll give you a little approval card, and you send it, and then it gives that to another agent to work on. But the orchestrator, I think you're talking about, is like you just send a message, and then it decides like, hey, this goes to this or that, or I'm gonna, I'm gonna, as the orchestrator agent, decide how to delegate it, right? So yeah, I think that can be useful in some cases, but I think like the I think of it more as like an assistant than like a commander or something. Like you're the you're the commander who's explaining like this is what I need to do and why, and it's kind of just saving you the couple clicks of going into that spot and copying pasting. So like you should be the one in charge in that scenario, I guess.

Chris DuBois37:24

Yeah, makes sense. Okay, I got two more questions for you here. With the first being, what book do you recommend every agency owner should read?

Bryan McAnulty37:33

So, yeah, there's a a book that I read a real long time ago. It was recommended to me by my dad actually, and he's in. He was in sales, and and he said I should read this. It was called, I think, it is an interesting title. It's called "Winning Through Intimidation" by Robert J. Ringer, and the the premise is that I think it was like he was a real estate agent, and he kept kept getting ripped off every single time, and he figured out like, well, I have to do I have to do better. I have to do better, and he talks about not how to be mean and rude to intimidate people, but like how you can set yourself up in a way to to like not have people mess with you, and not so much even about like legal stuff, but I think in like at one point in the book he went from like the the solo like kind of agent to he would fly into clients on a private jet with a team of like 17 assistants, and they all show up there at the client, and so he didn't want anybody to to mess with them and show that like he he really meant business. And I think when you make it clear in your agency of like this is the way that you expect people to work with you, then you you risk a lot less of scope creep being pushed around and and all that kind of thing,

Chris DuBois38:45

right? This is completely like random for this, but Alexander the Great right used to do like he would just train his his soldiers on drill and ceremony, and so there were battles where they would show up, and because they would all like move their spears at the same time and like pound their shield together, but they were all doing it like at the at the same moment, like perfect cadence. That was enough for the enemy to say we're out. Like just yeah, yeah. These

Bryan McAnulty39:13

guys are serious, yeah.

Chris DuBois39:14

And so it is interesting. Just like if you got your your shit together, you could actually show up and and just through intimidation, air quotes again for listeners. Yeah, run that. It's free, and it's a book that hasn't been recommended yet. So points for that because this is always going towards my to be read list. And so, last question is: Where can people find you?

Bryan McAnulty39:36

You can find me. I have my own podcast called The Creator's Adventure. It's we talk with some agency owners, but it's more so of like online knowledge businesses, people building like coaching courses, communities. If you're interested in trying out Latch Loop, our agent platform, that's at latchloop.com. Right now, we're giving free GPT 5.6 credits to people who want to sign up, and. If you are interested in having us kind of implement agents in your organization for you, email build@lashloop.com. I

Chris DuBois40:11

do want to before just closing the episode. I do want to hit two things sure for the listeners in regards to what you're doing. One, latch loop, awesome! Like go check it out because like this is really the future of where most agencies are going to have to go anyways. So go check it out on creating like courses and everything. The I think I told you this when we first chatted. Like one of the things we had done as an agency was we could collect clients through standard marketing. We could catalyze them through like growing accounts, but then we could also create clients by finding people who aren't necessarily ready to buy from us now, and then teaching them the things that they need in order to grow into the company that could hire us. I think this is even like Alex Hermosie's entire play for why he does content for companies that he wouldn't acquire anyways. Yep, more reason for them to also go check out that platform to be able to create some work. I think

Bryan McAnulty41:05

this is like a very like kind of untapped thing for agencies to be doing, but like they they really should be doing this.

Chris DuBois41:13

Yeah,

Bryan McAnulty41:13

and like if if anything at all, it's it's even great positioning for you against clients that come in and say like, wow! Like these are the people teaching about all this. In addition to being able to serve

Chris DuBois41:24

it, 100% And the the trust that you're going to build by helping someone get their company to a certain point, like you're you're the obvious choice for exactly.

Bryan McAnulty41:33

And to be to be clear, since we didn't talk about Heights platform, just to let everybody know what we're talking about, this is under your own brand, your own domain. So like people don't know you're using our platform. It's not like a marketplace. It's it's all under your own domain.

Chris DuBois41:45

Yeah, awesome. We'll get the links for both in the show notes. Brian, thanks for joining. This was a fun conversation. Yeah, I

Bryan McAnulty41:52

really enjoyed it. Thanks, Chris.