← AI Explored
Loading summary
Transcript107 lines
- [00:01]
A
Before we get started, I want to ask you something. Are you tired of feeling like AI is moving faster than you can keep up? Because I know what that feels like and I hear it all the time. Marketers keep telling me they're drowning in all the new tools and updates and features and must try frameworks and. And they got to do all this while they're also doing their job and trying to run their business. And that's why the AI Business Society was created. It's your personal curation team for AI. We tell you what deserves your attention and what you can safely ignore. And inside, you get live trainings from leading AI practitioners and interactive skill building sessions, bite sized learning libraries. When you just have a few minutes and a community of incredible marketers who are just like you, hundreds of them are on the inside and you can join right now and lock in an incredible special sale we've got going on. Visit socialmediaexaminer.com AI give it a shot. You got a 30 day money back guarantee, so you've got nothing to lose. Go to socialmediaexaminer.com AI and let's now jump back into today's show. Welcome to the AI Explored podcast, helping
- [01:08]
B
you put AI to work.
- [01:11]
A
And now, here's your host, Michael Stelzner. Hello, hello, hello. Thank you for joining me for the AI Explored podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzner, and this is the podcast for marketers, creators and business owners who want to know how to put AI to work. We've got a really incredible episode. Let's transition over there right now. Today we'll explore how to build AI agents that can help you operate your business. My special Guest is an AI implementation strategist who helps B2B businesses scale with AI automations. He's the founder of L2Digital. Keith Morgan, welcome back to the show. How you doing today?
- [01:48]
B
I'm wonderful. Thank you for having me back.
- [01:50]
A
Well, super excited to talk to you about this topic today. But before we do, I just want to ask, what is one of the biggest misconceptions you see when it comes to using AI agents and AI automations?
- [02:01]
B
I mean, there's a lot of people out there talking about, like, spinning this agent up to do this and that. In my experience, they take a lot more work to do than just spinning them up. Specifically, if you want them to do the one thing that you do and to kind of take over that part of the work for you, I mean, you got to give it the context you Got to give it the training, you got to work with it to reproduce the outcomes and do it in a way that you're not entirely reliant on. The AI tool who's helping you build it to create the process and the output. It's like it's got to be yours. You got to own it. And that does take some time. It takes some work to massage it into what it needs to be.
- [02:35]
A
Yeah. So I guess what I'm hearing you say is there's plenty of people out there that make it sound super simple. All you got to do is follow these six steps and before you know it, you're going to have this automation that can take over the work of another person. And what I'm hearing you say is it's not, it's not really that simple.
- [02:49]
B
Exactly. Yeah, it's. It does take some work. There are some layers to it. And I will probably get into it today, where as long as you've got a blueprint and you know exactly what you're trying, trying to do and you've got an approach, I think that's the big thing is like, how do you approach what makes it different from when you do it, when someone else does it? If you have that stuff in mind and defined, these things can be very, very powerful when they're built.
- [03:10]
A
I love it. Okay, so let's talk about the upside, the benefits that you or others that you've worked with have had as a result of following this procedure. We're about to talk about. Said another way, when we deploy a really great system, you know, or blueprint, as you talked about, what is the possibilities? What are the benefits that, that we could achieve with AI Agen?
- [03:30]
B
The thing that right out of the gate they can do is they can take smaller tasks that take time and energy and effort out of your day. They can take those on and do a very, very good job with them. And there's a, there's a book I love. It's called by Dan Martell. It's called buy back your time. And in it, he talks about the value of getting something 80% of the way done is 100% freaking awesome. And so that's what these agents can do, is they can take 80% of a task and get it done for you, and then you just layer the 20% on on top of that. And as you build these things, you're going to create these tools that do these specific tasks very well. And then the next level up is you create an agent that can then start using these different tools and smaller task agents to do bigger projects, bigger kind of system wide things. And then kind of the next layer up, which was kind of where the biggest unlocks are, is you create orchestration agents. Their job is to say, okay, this agent over here, go do your thing. And then you go your thing and those things start completing those tasks for you to the point where like at the start of the month, now what I can do is go and say, so I have like a portfolio of clients, hey, go set up all my clients of this type for the month and it'll go through, create the tasks, it'll start some of the projects, it'll start doing all the work, creating the emails, all that kind of work. And all I have to do is start reviewing. And that's all done with one quick prompt at the beginning of the month.
- [04:55]
A
Love it. Okay, so folks, we're going to get into the weeds on how to do this in a little bit. But first I want Keith to share a little bit of his story and journey because it's a fascinating story story that I think will help all of you understand what is possible. So why don't you start where you were when you ended up hiring an assistant and kind of share that journey that you went on.
- [05:12]
B
Yeah, so I'm a company of one. I have a couple freelancers that I work with closely, but for the most part I'm managing an agency by myself, which is client relationships, getting projects completed and approved into clients and publishing and then all the admin work that goes with running a business and paying freelancers and all that kind of stuff. It's a lot. And it was starting to get to really hit ahead end of last year in, into the beginning of this year. And so I decided, you know what, I need a virtual assistant. I need someone who can come in and help me organize my chaos. And so I did a bunch of interviews. I documented everything that I was looking for. Specifically I need help with the email. I need to create like client briefs in the morning, like what are we going to be doing with the client that today? What are the outstanding projects, all that kind of stuff. Consolidate meeting notes, create tasks and project management tool, manage my calendar, invoicing, that kind of thing. So I found someone. She was great. She was experienced, having worked with someone in my exact role and situation. She was funny, she easygoing, she was perfect. And started working with her. And then about a month later I got an email saying that it just wasn't a fit, it wasn't working out, which was hilarious. Because when I received that email, I was writing her the same email, letting her know that I didn't think it was working out. So it was, again, we were on the same page. She was great and it was nothing against her. It was essentially the problem I had was I hadn't given her clear enough directions. I mean, I've been working by myself better part of five years and so I forget about how to effectively manage people. I didn't give her the full controls that she needed to do what she needed to do. I was holding on too tightly and the tasks were too varied. It wasn't the same thing every time. So rather than go out and try to find a replacement for her, what I decided to do was to actually create my own virtual assistant using an AI agent. And it was fortunately, kind of serendipitously, it was around the same time that Claude like 4.6 opus came out. So these AI tools took a level up in terms of their ability to orchestrate these type of process to do these type of agentic type work. So I started small. I picked a task that I didn't want to be doing and it was something that was very important that I always forgot to do, which was the meetings. Like after a meeting I wouldn't sit down and write out my notes and then create the task. I wouldn't do all the follow up work. I would just jump into the next thing and then always forget about the stuff that came out of the meetings. So I started the process there and the first one I built was kind of, it was game changing. It was like, oh my God, this is exactly what I needed. And then I just started extrapolating from there, like what other tasks can I do? And then how do I start to unite this together? And that's when the real unlocks started to happen.
- [07:57]
A
Yeah. Tell me a little bit about like what did this make possible for you? Because I know so many people listening, whether they work for themselves like you do, or own a business like I do with a bunch of employees or work inside of another business or maybe are thinking of starting a business, either they don't have the funds maybe to go out and hire an assistant. Right. Or they would really love to get kind of work that they know they have to do, but they don't enjoy doing off their plate. You've already mentioned Buy back your time by Dan Martell. So what did this actually do for you and how long have you been doing this now? Like six, eight months or something like this?
- [08:30]
B
Well, so the cloth stuff really opened up the doors for a lot of this work. So I had been using AI tools for a while, right. And I had been using ChatGPT, creating custom GPTs. And then I got real into much of last year and even into it before that, using N8N and make to create these AI powered automations, right. That would go out and do these tasks on schedules or on demand on my behalf. It was until the Claude 4.6 opus came out and then 4.7 that it opened up the opportunity to move away from those automated systems that can be a little more error prone and into something that was a lot more easy or a lot easier to manage. Could shift between different AI tools very quickly and kind of build the agentic folder system and context that allows for this kind of stuff to take place.
- [09:22]
A
Yeah. What kind of impact did that have on your time?
- [09:24]
B
60% of the work that I have to do for clients is done now within the first hour of the first day of the month. It's to that level.
- [09:31]
A
Wow.
- [09:31]
B
So, like, I have a portfolio of clients and what would normally take me a couple weeks to get all of it done and created and improved and polished, and all that 60% of that stuff is just handed to me Monday morning of that month. It's a huge unlock for me because now all I'm doing is reviewing, improving. I'm not even starting from scratch and half this stuff and then I'm sending things out. Clients are starting to get things, like, right away and they're starting to like. And by the end, by a couple weeks in, I mean, we're in really, really good shape on all the client deliverables. And now I'm just managing relationships and coordinating different things. I'm not executing on a lot of the deliverables.
- [10:09]
A
You know, this is such an important concept. And I talked to so many entrepreneurs who've been on this show or who I've met in various different settings. And I have this thing called the flywheel effect that I often talk about, which is the hardest part of anything is to get started, I think, or to finish. It's one or the other. It's either getting started or getting finished. But for a lot of us that have this creative desire inside of us to do new things, it's the starting that, that that stops us. And what I love about what you've done here is you have effectively done something that's not just started, but also maintained. And now all you got to do is kind of go in and probably add finishing touches on it. I would imagine that has freed up your creative time and your strategic mind for a lot more stuff. Is that accurate?
- [10:49]
B
Yeah, 100. The other thing that's kind of done for me that I didn't really fully appreciate initially is there's a lot of stuff you do in the moment that you forget about. Or I remember I did that, but I don't remember where I did that. The way this works is it's all consolidated, and all I have to do is go ask the agent, hey, when did we do this? Or when was this project complete? Or how do we do this specific process? Because I didn't document it well the last time. And it's all right there. There's a term floating around, like a lot with second brain, like creating a second brain. That's essentially what I did. And it's like my go to resource for things that I forgot I did or forgo how to do.
- [11:27]
A
Love it. Okay, so, folks, I hope you understand why I asked him to share the story. Because I want to open up everybody's minds to the possibilities of what can be done here. And I want everybody to think about their unique application. You may not be an entrepreneur, you may not be creating agents to specifically run your business, but it doesn't mean you can't learn from what we're about to talk about here to help you in any kind of thing. So where do we actually begin, Keith? Because let's say we want to. Want to, like, learn a lot from what you've learned here. So where do we start?
- [11:55]
B
I have started several different ways trying to approach these agents and the one that consistently produces the best results. And I've done this with other clients, too, just trying to help them build these systems into their own organizations. And if you're going to approach this, it is best to do it starting out with the org chart or accountability chart, if you're an EOS fan. Yeah.
- [12:16]
A
Describe what that is. For people that don't understand that, essentially
- [12:19]
B
what it is, it's a hierarchy of how the business operates. So you. At the top, you've got the CEO or visionary. Reporting to them is the integrator, the president. So the person who's kind of in charge of operations. And then there are three main, like, business functions. There's sales and marketing, there's operations, and there's finance. And then underneath each of those, you've got different departments. So say under sales and marketing, you've got marketing department, then you've got your inside sales and outside, that kind of thing.
- [12:46]
A
Got it.
- [12:47]
B
Well, within Each of those positions within the company, there are roles. So reasons that that job exists for the organization, that role has responsibilities. So what are they in charge of doing or making sure that happens? And then they have tasks. So what are the specific things they need to do on a daily, weekly, monthly, quarterly basis? If you're going to start anywhere, you got to start with the tasks because there are going to be actions that you do regularly that you can very quickly train an agent to do. And they're purpose built. So you're building it to do this one specific task repeatedly, expertly, with the right output. And so the agents can help and do that kind of work very cleanly and consistently. But once you build the first task, then you can create a second task and then third task, agent. And so they're doing all these very specific tasks and then that's when you can start to build the agents over on top of them. So the agent then has these sub agents or tools in that case to do these specific things. And that larger agent has a mandate, it has instructions, it has a goal, it has performance, it's looking to hit. And they can start to utilize those sub agents as it needs to. And then I think the most important caveat to all of this though is put the checks in. Where's the human got to be in the loop, where are they making sure that things are happening the right way, all that kind of stuff.
- [14:16]
A
Okay, I want to ask some, some a bunch of clarifying questions. First of all, thank you for explaining how this EOS thing works or whatever it is. What is it? Entrepreneur.
- [14:26]
B
The Entre Entrepreneurial Operating System is what EOS stands for. But yeah, the org chart or accountability charts, the main thing.
- [14:32]
A
So obviously there's some people that already have org charts like me and there's some people like you who you do everything and you need this kind of org chart. Do you have any advice to people who are already in an existing organization but maybe don't have staff? Would they create kind of a virtual org chart underneath them? If they had the budget, they would hire people to do these kinds of roles. Do they want to think through that and treat the ultimate thing that they're trying to build here, which is a team to support them in their role? I'm just curious how you might suggest that to somebody who's already part of an existing organization.
- [15:03]
B
Absolutely. And the way I looked at it when creating it for my own business was I'm in all of those roles now. So my picture is in each one of those little Role module with all the list of responsibilities and tasks. So look at it that way. Like, look at it as, what am I responsible for overseeing? And then what are the tasks that go underneath that? And if you're just a person in a larger cog or a larger organization, you have one specific focus. There are probably things that you're responsible for in the tasks that you're. You need to be doing. And you can look at it from that hierarchy. So it's like it's just one block and then come down, branch off there you got two or three different responsibilities, and then under those is a list of tasks.
- [15:45]
A
Right.
- [15:46]
B
Look at it from that perspective. If you're an entrepreneur, like I am, and you're wearing a bunch of hats, like, there's a tool I like. It's called 90io. It's an EOS system that allows you to manage your business using that entrepreneur operating system. But I think the free account also includes an accountability chart, so you can create your own visual accountability chart. The other thing you can do is use Claude to say, I need an accountability chart, and here's all the functions of my business and here's what I do, and here's the roles and the tasks. Help me create that and visualize that. I have a picture of it up on my wall, so I can see it. And I remember like, okay, this is the. This is how the organization functions.
- [16:24]
A
Love it. So it's the 90 is spelled out. It's not 90 IO. It's 9 n I n e t y IO.
- [16:31]
B
Yes.
- [16:31]
A
Okay, so let's talk about, like, what you did. So now that. Because we've kind of structurally talked about how we want to create an org chart. Right. And we want to identify who does what. But then you start with the tasks. So do we start with the tasks with the org chart in mind that we're going to ultimately build an agent that oversees the tasks, or do we build the agent that has certain responsibilities? You've built a couple of these that we talked about in preparation. Maybe you could share how you built them just so people can wrap their brain around that question.
- [17:01]
B
Yeah. Think of it as, okay, I've got this one task. I'm going to create an agent to do this one task.
- [17:06]
A
I see.
- [17:07]
B
And when you're creating these things, there's really three main components that you're going to need to do this. The first one would be the AI tool, the obvious one. So I use Claude almost exclusively for this kind of stuff. But OpenAI is great. I'm A Google Workspace person. So Gemini, I use Gemini all the time too, but any of them can work. And the beauty about the way these systems and kind of what I'm going to walk you through here is it's LLM agnostic. You can use any of the AI tools you want, whatever one works better for you, you. So the first one is you got to have the AI tool. That's the first part. The second part is you got to have the user interface. And we'll come back to that in a second. And then the third part is you need the context, you need the memory. And this is kind of the, the most important part. And if you think about it, the best outputs of using ChatGPT or Gemini is when you give it really good context. You give it instructions, you give it a Persona, you give it a background, you give it the data it needs to help make all the decisions. That's essentially what we're trying to do, but in a much more easy to understand and easy to navigate way that the AI tool can go and find what it needs. So, and to do that, I have a folder, think of this on, on your desktop. I have a folder on my desktop called L2OPS. That's all it is. And within that folder there are five or six other folders. So there's a folder called Playbooks Folder called Reference Skills, Templates, Scripts and Clients. In each of those folders are a bunch of markdown files and other tools that the AI can go in and read through and process when it, when it's doing a task that I ask it to do. So for example, the Playbooks folder that has all my SOPs in it, my statement of processes for any specific thing that I need to get done and executed, there is a playbook for that specific task. The reference folder is kind of the overview, overarching. Hey, this is how we, this is the naming convention we use. Here's how the files work and what's included in there. Here's the technology we use, that kind of thing and what it's used for. So all that kind of background context that it needs, the skills, the skills are the really, really valuable ones because they're reusable step by step instructions on how to execute a task. So I have a skill in there that clearly defines how to go into ClickUp, which is our project management tool, and create a task for me, me. And then how that task gets created depends on the context of what it's being asked to do. So what client is it going into? What's the task name, what's the description, all that. But the process of creating the task is the same. And that's what those skills are, they're very, very valuable. The Templates folder includes a bunch of examples of how things need to be output at the end of it. Like here's an example blog post, here's an example email, here's an example this, that and everything else. And they're all named within these folders so that the AI can right off, right by using looking at the name of the file, knows exactly what, what it is and what it's used for. The Scripts folder, that's where a lot of different Python code and tools that we use for connecting to different APIs are. So if I need to connect to HubSpot because I need to know this one specific data and I don't want to give the AI tool full access to everything in HubSpot, I just need to know, pull out this con, these contacts who match this specific criteria, that SCRI is designed to do that one specific thing. And it's also used to process and crunch data and do math and kind of manipulate things the way I want them to be manipulated based on part of the what the task requires. And then for me, because I'm an agency, I have a folder called Tasks and under that I have a folder for each one of my clients. And then within there is all the context they need to know about the client, like the name of the client, the acronym I use, where their folders are in Google Drive, who the main priority contacts are, what they do, that kind of thing. So I have all the context that the agent when doing a task for that one specific client will need. And it's all referenceable right there. I also, that's where I consolidate all meeting notes for those clients. So after I meet with a client, the AI will go grab those meeting notes and save it in that directory. Now the AI agent has access to everything we've talked about over the last three months, which is very, very valuable context, especially when I need help finding different, different things. So it's that folder system, it's that very simple, not overly in depth foldering system. Because what we do with this and what the AI will do as you're asking it to help you build these agents, is it will start to create its own map of all of it. And so when it comes time to ask it to go find that HubSpot API thing and as part of that, then go in and click up and do that, it knows exactly where in those foldering systems to go look and what file to look for. And then it can just grab that information, that context and execute on it and then go on to the next step of the playbook.
- [22:05]
A
Love this. Okay, so real quick, you've created a folder structure, and I love structure. All this is important context for the AI in there. You've got Playbooks, which we're going to talk about in a little bit. You've got reference folder. You've got skills, which is really very important. It teaches the, you know, the skills. We'll talk about those a little bit more. Templates, scripts, which is more the technical kind of side of things. And then you've got tasks under there. You've got clients and meeting notes. Give us a little example of like one task that you have your AI agent do once this sucker is in place, just so people can wrap their head around it a little bit.
- [22:43]
B
Yeah, I think the best one is the meeting notes 1. I think that's my favorite. So I use granola for meeting note taking, largely because I love the way the interface works. I love the way it summarizes the notes because I can write in my own notes as I go and then it'll fill in the context around that. But it has a wonderful integration and allows for a nice integration with like, I use cursor, which we'll get into in here in a little bit.
- [23:05]
A
Granola for those that are curious.
- [23:07]
B
Yeah, so it's a. It's a great tool, but it has a very nice API connection. So what I can do is my agent, like if I'm we're walking through a Playbook, is I'll give it a command. I'll say, hey, go grab meeting notes from the last week for all my clients. And as part of the next step of the process, it knows, okay, here are all the clients. It goes and finds all the acronyms. And then every meeting I have, I. The meeting name leads off with that acronym. And so it'll go through the meeting names and look for the acronyms at the start.
- [23:37]
A
When you say acronym, you mean like, like the. The short version of like SME for social Media examiner or something like that, Right, Exactly.
- [23:43]
B
Like an L2 meeting would have L2 underscore and then what the meeting was about.
- [23:47]
A
Got it. So you use a file naming structure that the AI knows how to read, basically. Yeah.
- [23:53]
B
And that would be part of that reference file, the documents of like, oh, yeah, here's how we name things in granola.
- [23:58]
A
Got it.
- [23:59]
B
And so it'll go look for that. And then the first thing I have in the, in the note itself is what meeting is about. Is it about a project? Is it about strategy? Is it just internal meeting? Whatever it is. So it describes what the meeting is and then the AI or the process, the way it's designed is it goes in, finds that meeting meeting note, aligns it with the right client, pulls the notes themselves, converts that into a, like a text document, and then saves that file in my local directory under client meetings. And then it'll go on to the next one. And so it goes through each of those in turn. I don't have to specifically say go find all meeting notes with like client acronym XYZ and then do this, this and this. It knows to do it all. All I had to do was ask it to go get them. The other thing it does then is once it reviews the notes and identifies any to do's I have from that conversation.
- [24:57]
A
And we'll add that to your ClickUp or something like that.
- [24:59]
B
That, yeah. And then it'll go and grab that ClickUp skill and create the task for me in the, in the month and in that, that, that task list with all the context I need to actually execute on that task.
- [25:11]
A
That's super cool. Hey, before we get back to the show, I want to talk to you about something I've been thinking a lot about lately. You already use AI, and if you're like me, every single day, you're using it. And you're not someone who needs to be convinced that it matters. But here's what I've learned from working with hundreds of marketers inside our AI business society over the last two years. The problem is almost never AI. It's trying to do it all on your own. You're testing tools that you're likely not using again. You're rewriting prompts because they don't sound right. You're probably watching tutorials that give you lots of ideas, but you have no idea what to do with it. And you're working harder with AI, not smarter. And that's exactly why I built the AI Business Society. Every live training is led by an expert practitioner who uses AI in their marketing all the time. And I'm personally there on every single one of these training sessions to make sure that nothing gets too crazy or abstract. You leave with something you can implement that very day. Plus you're surrounded by a community of marketers who are just like you. And they're all there to support each other. And you get to be part of this awesome community. The investment is $497 for the full year. And if within the first 30 days you decide it's not for you and you join, then just ask for a refund. We're not going to ask any questions, but at least you gave it a shot. Head over to SocialMediaExaminer.com AI that's SocialMediaExaminer.com AI and now let's get back to the show. A lot of people are probably wondering, okay, this sounds really fascinating, but let's talk about the cursor and clause connection here, because this is the big mystery, right? Because so far, here's what we know. We're using files on your local computer, right? And in your case, you got a Mac, I would imagine you could do it on a Windows machine, but you've got a dedicated folder where all this data lives, which makes some people feel a little bit more confident because it's on their hard drive. But, like, what's the connect the dots between the technology that is ultimately powering this? Just so people can understand this, Because I'm sure a lot of people are like, okay, how are you pulling this off? Off? You mentioned Cursor and you mentioned Claude, so go ahead and, like, elaborate on that.
- [27:20]
B
Yeah, you know, it's a very good question. That was the missing piece and that was the part I struggled with the most initially when I was trying to learn about how to build these things. So I use Cursor. There's other ones, like VS Code is another one. So Cursor is actually a variation of VS code. But essentially what these tools do, they're the user interface and how you're going to interact with the agents and the files on your system. So it's a downloadable program you put on your computer. Like, I just open up Cursor on my desktop and it has a. I can open it into a. What they call repositories, which is my folder system. So I open up L2 Ops in cursor, and now it has access to all the files along with all the other files it's created to preserve its own memory of what is what and what I've asked to do and who it is and all that kind of stuff.
- [28:09]
A
Stuff.
- [28:10]
B
So, like, for example, there's a Claude file in there. There's a dot cursor file in there with the context of who it is and what it's designed to do.
- [28:18]
A
So talk about why cursor? Because it's cursor.com folks. But what does Cursor do that maybe makes it easier than Claude code and some of these other more technical solutions, codecs, all that kind of stuff.
- [28:29]
B
Yeah. So I think the first and foremost thing, it's not, I'm not operating out of my Mac terminal, which is I think the least inviting user interface of all time. It's essentially like using a web page, but it's a native application. It's got to access the files on my computer. And on the right hand panel there's a little chat window where I can have conversation with the, with the agent. Cursor also has a. It's its own built in AI capabilities, but it also allows you to connect with your own CLAUDE account or its variation of Claude. So you can access all these different models directly from that one interface.
- [29:05]
A
So you could choose Gemini or Claude or ChatGPT. Okay.
- [29:09]
B
Right. So I can choose to use OpenAI's Codex or I can use Claude code, or I get to choose what model I want to use based on the requirements of the task.
- [29:20]
A
Okay.
- [29:21]
B
So there may be a very advanced coding thing that I wanted to get done. I'm not going to trust Cursor's AI. I want CLAUDE code for that and so I get to pick and choose what I want and when to use it.
- [29:31]
A
Okay. And it's a lot easier to use overall. You find that Cursor is way easier than just messing around with cloud code. Yeah, it's like a plugin almost. Right. Cursor is like a front end like interface that makes it a little more user friendly to use these more technical backend tools. Would that be the best way to describe it?
- [29:47]
B
That's exactly. So think of it like a window on your desktop. On the left hand side is a list of all the hierarchy of all my folders. On the right hand side is my chat window that I'm using to talk to the AI tool. And in the middle is the file that I'm working on. And so yeah, it's a connector. It connects the AI tool to the files on my computer and I don't have to upload them into ChatGPT for them to do whatever the heck they want to do with. It's all safe and secure on my desktop.
- [30:15]
A
I've had a lot of people on the show talking about codex cloud code, OpenClaw. One of the big questions a lot of people have is security. Do you find that Cursor is a little bit more safe, if you will, because you could just quit the app? Or does it run all the time? Like, can you Granularly provide access controls. Like folks that have Macs know that sometimes you have to give permission for it to do certain kinds of things. I mean, does it provide a little safer sense of security that's not going to just take over your computer and do crazy stuff?
- [30:44]
B
Yeah, so there, there's a couple elements to it. One, with Cursor, when I open it up in a file directory or repository as they call it, it's restricted to that repository. It's. It can't go outside of that.
- [30:54]
A
Okay.
- [30:55]
B
Again, I'm putting my faith in the company itself and hopefully that's accurate. The other thing that Cursor allows it to do is you can set it up. So depend depending on how comfortable you are are with what you've asked it to do, you can set it up so that it has free reign to just go make decisions and do things. Or you could have it set so before it makes any update to a file on your computer, it has to ask for permission and confirmation, which initially I used that all the time. And then after a while it became this very annoying thing that I had to sit there and watch it and approve it, everything. Eventually you just kind of start to trust it to do the stuff. But you have that safeguard in place which is very, very nice. It's not like an open claw thing where you turn it on and it's, it runs wild. It's. It has access to everything, which is why you have to do the Mac Mini or the dedicated virtual server to launch those things. To keep things safe and separate. It's a, it's. I like it for that, that respect.
- [31:52]
A
Do you have a monthly fee with Cursor and then have to purchase token usage or do you have to set up a API account with CLAUDE in order to be able to make this work work?
- [32:02]
B
So Cursor has a monthly subscription and depending on how aggressive you want to get with the token usage, you pay accordingly. So I think I paid $99 a month for access to it.
- [32:12]
A
Okay.
- [32:12]
B
And it also allows you to directly connect to your own Claude account also. So if I wanted to use my Claude account and I have maybe a subscription there that has a ton more token access that I would have through Claude's or I'm sorry, Cursor's version of it, I can do that and I can run everything directly out of my own cloud account. So it allows for that kind of flexibility. But I found that the 99 version that I use with Cursor, it more than is enough to do what I need to do, and I'm in it every day.
- [32:43]
A
For those of us that might have teams, is there a cloud based function instead of just having it on your hard drive? For example, could it be embedded in a Google Drive or. I don't know what the other options are. But are those options to put those files up in the cloud with Cursor, presumably, or do you not recommend that?
- [32:58]
B
One of the good ways of handling it is through GitHub, because it's a file directory on your computer. You can use Curso or Terminal to upload that directory to GitHub and then you can give your team access to that same GitHub so they can pull it into their own local system and they can use the same context and files and everything that you're using. So essentially you're using one operating system across the organization. There are other ways to do that Cursor online you can access through things like that. Cursor can connect directly to like Google Drive. So if you want to live, have everything live, live on Google Drive. The challenge there is it's a lot more token usage for it to go in, find things in Google Drive, process the Google Docs or Google Sheets and do what it does on that end versus just accessing files directly on your desktop.
- [33:46]
A
Now we develop software, so I have a cursory understanding that GitHub, which is owned by Microsoft, is a massive repository used by software developers all over the world. It's a very safe, granular control. All that kind of stuff is my understanding. And I personally don't have a GitHub account, but I know people on my team do. It sounds like that's really a powerful place if you wanted to do that. And I would imagine eventually, if you wanted to and you had other team members, I would imagine they could also access your cursor. Does Cursor allow you to have multiple users potentially as well? You know what I mean? Like, imagine you had an actual employee and you wanted them to be able to go into your cursor and mess around with what you built. Is that a possibility as well, you think so?
- [34:25]
B
The cursor is just the interface for the most, most part.
- [34:27]
A
Got it. Okay.
- [34:28]
B
Yeah.
- [34:28]
A
All right.
- [34:29]
B
But again, it's one of those two where like, I have it for myself and for a couple small clients who have one or two users. If you're talking about a large organization, I'm sure Cursor would have a lot of that information for you.
- [34:39]
A
Got it. All right, well, thanks folks for letting me go down this rabbit Trail, because I've never done this before, and I think it's really helpful. Hopefully a lot of people got their questions answered. So we're going to bring it back to where we were. We're talking about how to build agents. Okay. We're talking about creating file structures on your hard drive, and we're talking about having specific folders for spec purposes. And I really want to dig in on the Playbooks because this is something that's really, really important. It might perhaps be one of the most important things. So why don't you explain what playbooks are and kind of how they work and let. Let's dig in on that a little bit.
- [35:12]
B
Sure. So the playbooks are essentially SOPs. They're your process documentation, but written in a way that is you're instructing the AI tool on what is needs to do and where it needs to go find specific things. So there's no universal, like, here's the exact way to go build a playbook. It's all going to vary based on the scenario. But what's nice is if you have a couple things in line, which I'll go through here in a second, you can use the AI tool to create the playbook. It'll create everything it needs, and then you just review and approve.
- [35:48]
A
When you say the AI tool, which tool we talking about about here?
- [35:50]
B
Whichever one you prefer to use. If you use Claude or Cursor's version.
- [35:54]
A
Okay. So if you're already using AI to do certain tasks, this is the foundation potentially for a playbook, is what I'm hearing you say. Right?
- [36:01]
B
Exactly. Yeah. So if we're in. Think of it, we're in Cursor.
- [36:04]
A
Right.
- [36:05]
B
It has access to my file directory, and I'm going to prompt it to ask it to create an agent for me to do that meeting scheduling thing, for example.
- [36:13]
A
Okay.
- [36:13]
B
I'm going to give it the instructions, I'm going to give it all the context. And, and the main thing, when you're thinking about. About it from that perspective, it's like when you're going to go build the agent, you have to do a little homework first. There are some prerequisites before you even just jump in and say, okay, let's create an agent to do this. If you go down that road, the AI is going to think, do what it thinks it needs to be versus how you want it to actually be executed. So to make sure that it aligns with your vision of what it needs to do, the first thing you need to do is have a description of your approach, how do you typically go about doing that? Specific task, tasks. And even if it's just a number list of okay, do this, then this, then this, this and this, have that defined for yourself up front. If you have any sort of statement of process about how that thing gets done, make sure that that's included in there as well. Because that's all context it needs to know, okay, this is the order of operations, this is exactly how this needs to flow. The other side of it is, do you have any sort of templates or examples of the work and what it needs to look like on the other end, what tech is required and do you have the connections or the gateways to get access to that technology? For example, do you have an API key to access? We use HubSpot as our CRM. Do you have that tool? Or do you have the MCP or the connector for granola turned on so we can access meeting nodes? Make sure that that stuff's situated. And the reality is if you don't know how to do it, you can also ask the AI as you're going to through this process on how to do that step. Because there's a bunch of stuff I didn't know at the beginning and I the AI walked me through that process and then any sort of additional context and background it needs to be able to execute on that effectively. So here's who we're writing this for, here's the tone and style, here's all this other stuff that goes into it that essentially you're creating. You're dumping your own brain into this document and you're going to take that entire thing and your incursor and say, okay, I want to create an AI agent to do this specific task, high level, here's what this task is. Now here's our approach, here's our process, here's a template of what it needs to look like. Here's the tools we use to do this and the data we need. Help me walk, go step by step, create. To create this agent we're going to use, and I don't think I've mentioned yet, but we use. The framework that we're talking about is called Watt framework. It's WAT framework. So it's workflow, agents and tools tools. And it's a well, well understood framework, especially for the AI tools because it, that context will help it understand and start to understand how you're going to structure your file system to make this stuff work for you. And there's a ton of resources on that on what out there that you can check out? We won't go into too much detail, we've already explained most of it already. But you're going to go in, tell it what framework you're using. Here's our approach, here's our process, here's everything you need to know. Help me create this AI agent and, and specifically the playbook for how this going to execute this thing. And it'll start to go through and do it for you. It'll create the files, it'll create the process, and then you, you essentially say, okay, let's plan this whole thing out. And it'll create a document of how, exactly how it's going to build it out, when it's going to do this, how it's going to do this, so on and so forth.
- [39:28]
A
So cursor in this case is what you're interfacing with. But I would imagine you could do the same thing with Cloud claude. Right, Because I think CLAUDE can output all those files that we're talking about here as well. Right, okay.
- [39:38]
B
Yeah. So within cursor, I may pick CLAUDE to help me build this agent.
- [39:41]
A
Got it.
- [39:42]
B
So I'll select that one instead of the cursor one and we'll just start the same process. Everything is exactly the same, it's just using a different LLM on the back end. But what will happen is it'll create this plan and then you start going through and reading through it like, no, I wanted to do this or this, and then you just highlight it and say, no, we need to change this, this to this, and then it'll update the plan and then you go through and refine, refine, refine. And then once you're done, once it's in a good place, there's a button that says build and you just hit it and it'll start to create the files and it'll start, go through the process and then once it'll. It may take a minute, it may take 10 minutes. It's hard to tell depending on how complex the project is. But you can watch it going through and like the process it's thinking through, we need to do this. Oh no, we have to go also do this. So we should probably update this over here. And I tried to create this connection, it failed because it failed because of this, this and this. So let's go fix that. And like it's going through this whole process and it's amazing to watch, just to watch it work itself through this whole thing. But by the end of it, you've Got this minimum viable agent that you can say, okay, let's do a test run, let's go pull this one meeting note from last week and see what happens. And while the first output won't be exactly what you're looking for, the first output is going to blow your mind in terms of, holy cow. That was so simple. And all I asked it to do was go grab that note or go pull this analytics report, or go create this task or write this sequence of emails and pull this list and scrub these contacts. And it's doing all this work. And all I had to do is ask it to do it because I've given it all the context it needs and the instructions to actually go do the work.
- [41:18]
A
Okay, couple questions here. And this might be a transition to the question that we had previously agreed we were going to ask next. But we've talked about tasks and we've talked about agents. Right? And I think you said each agent does a specific task. So does that mean you have like a bazillion different agents that you're naming? Like, help me understand the difference between a task and an agent. Maybe you can share this LEO example. Maybe that'll connect all the dots in my brain. I'm not 100% sure.
- [41:45]
B
Yeah. So like, like I mentioned before, there's the naming of this. It, it can be very complicated. And, and I don't think there's an agreed upon definition of exactly how this works, but the way I think about it is you've got sub agents. Sub agents are designed to do one specific task. Specific, like, well, and then above that you've got what are would be called more like orchestration agents. So a good example is, so my virtual assistant, which I named Leo, I will, the beginning of the month, I'll go and say, Leo, we need to set up all the, the client tasks and start executing on the work work for all our distributor clients. For this month, LEO as an agent has been programmed with its playbook to understand exactly what I'm asking for. So when Keith asks for this, here's what he's looking for, here's what the next step needs to, what we need to do next. And LEO will then know to go in and say, okay, well we need to do this task. So we're going to spin up this specific sub agent and it's going to do this, this, this, and this. And then we're going to go clean this contact list. So it's going to go this, this, this, and this. So in other words, I've Got a orchestration agent that it can read my commands and then know what to do and what other agents to activate and turn on and turn off based on what the ultimate outcome I'm looking for.
- [43:03]
A
Well, and let's be super clear here, because we're calling it an orchestration agent, but it's not like you go into cursor and say, build me an orchestration agent, right? You're effectively dealing with like a sandbox, right? You can create the heck you want. So what you're really recommending is you create an agent for every little task, right? And it's probably got a title or a name of some sort. And then ideally you create another agent that oversees certain kinds of tasks. Is that correct?
- [43:30]
B
Yes.
- [43:30]
A
And then how does it work on a regular basis? Like how do you trigger it? Does it always have to be Keith going in and saying to do it, or is there a way that you can have it do things on a cycle?
- [43:39]
B
So the, there are the command parts of it, like the manual requirements requests and that kind of thing, but no, there are other ways to handle this stuff. So like for example, if you're using cursor, well, there's cursor automations which allow you to set up these tasks on a schedule. This was always the thing that I loved about like using make and n8n is you can set these things up. So when an email comes in or when at 12 o' clock on, on Monday it goes in and does these specific things. These agents struggled with that. But now there's these tools, these timing, these trigger tools that allow you to, when this thing happens every 30 minutes, turn on and check this, if this, then go and do all of this other stuff so you can start to build out these, these processes that are heavily automated with just the specific set of instructions. And those cursor automations are good. If you're using GitHub, like you're saving all this stuff into the GitHub directory and think of that as just kind of a repository for all your files. So it's like a external drive in many ways. But there are scheduled tasks. You can use Cron Jobs is what they're called. But like you can use them in GitHub and so you can set up GitHub to run these tasks when it's. When it's asked to, or on these schedules.
- [44:53]
A
So how long do you think it's going to take someone to set up their first task and then their second task? Do you get to the point where you get this down To a mad science and it takes less than an hour. I mean like just give people an idea what kind of. Because we said at the very top it's not easy easy, but once it's built, it's probably very powerful. So what's the kind of time commitment?
- [45:11]
B
So the biggest time commit is getting all the context and background in place so it understands one, it's it's job. What is it designed to do? Two, what your organization is, what do you do? Service you provide ultimate goal and outcome. And then once you've got your first task, well, once you have your first agent task in place now you've got almost in many ways a blue blueprint and you've got some skills built already for that specific task. And the second task may also require some of those same skills. So now we're not recreating things over and over. We've got access to the same tools and skills whenever we need them. So the next task may take less time, but how much time it takes is completely dependent on how complex the task is. And so I think going back to, to do it, focus on simplicity first small minor tasks and then you can start to layer these things together and build these things together. So one task leads to another task, to another task. And that's how these agents become very, very powerful. And that's how those bigger, larger playbooks start to really manifest is, okay, I've got, when I have to go in and do a piece of, I have to write a piece of content for client, I may have to go do research first. And then the next thing I have to do is I have to go look at our asset library. What did we write about on this topic already? And then I have to go interview the subject matter expert. I can create an agent to do that first task for me. Go out and do online research and find all this stuff and write this brief for me. That's my first one. And then the second one is, okay, go through the, the asset library, find all the relevant content, summarize what we've written already. So now I've got two things. I've got two agents that go and do the these two things. Well, now I can start to stack them together in a sequence and then the output becomes that much more powerful. Try to break it down to small little tasks and then build up from there. The worst thing you can do when building these things is just to start, okay, I need a content marketing agent. Go. It'll never ever work out the way you want it. And it's the challenging thing with this and kind of a just a red flag here is the deeper you get when it's creating its own context and instructions and all this stuff, the harder it is to unravel. So by starting small and building first, it's a lot easier to control exactly what's happening versus top down. And now you got to go through 100 different files, thousand lines of this and that, to understand exactly why it's doing what it's doing. That becomes a lot harder to manage and scale up.
- [47:41]
A
Folks, I said earlier that we might get back to Skills. We didn't get there today, but I will tell you, we've talked about skills pretty extensively on the last many shows, so if you're new to the podcast, be sure to listen to some of our prior episodes. Keith, this has been solid gold. If people are interested in connecting with you on the socials or maybe wanting to work with you and maybe even experience what it's like to have your agents working with them, where exactly do you want them to go?
- [48:06]
B
It's Keith Mooring on LinkedIn and so that's probably the best social to connect with me on that. And then the website for the company is L2 and so happy to have a conversation, even just kind of brain. So I love talking about this stuff. This is my favorite stuff in the world to do. So yeah, even if you just want to have a chat about it, I'm
- [48:24]
A
more than happy to do so. Thank you so much, Keith, for sharing your wisdom with us today.
- [48:27]
B
Absolutely. Thank you for having me.
- [48:29]
A
Hey, if you missed anything, we took all the notes for you over@social mediaexaminer.com A110 I can't believe we've already done 110 episodes of this show. Be sure to follow this show on your favorite podcasting app. And if you've been a listener for a little while, we'd love a review. And also let your friends know about this show. I'm most active on Facebook, LinkedIn, and X, and do check out our other show, the Social Media Marketing Podcast. This brings us to the end of the AI Explored podcast. I'm your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day and may AI help you become more successful. The AI Explored Podcast is a production
- [49:07]
B
of Social Media Examiner.
- [49:12]
A
Before you go. If you're tired of figuring out AI all on your own and testing tools that don't seem to work the way you want them to work and constantly rewriting scripts and prompts because they don't sound like you. The AI Business Society was built for you. Every week, we cut through all the noise. Every training is led by expert practitioners, and you're surrounded by marketers who are just like you. And right now, you can join for $497, which is our special sale price. And your rate is locked in for life. And you got 30 days to ask for a refund if it's not for you. So what have you got to lose? Check it out@socialmediaexaminer.com AI and I'll catch you on the next episode.