Autonomous AI: How to Get AI to Manage Itself

Hi, I'm Kea Kelly. I teach TikTok marketing and I'm here at Social Media Marketing World. So I have loved being a speaker at Social Media Marketing World the last five years. My favorite conference because not only do you get to speak or whatever, but you get to learn so much and be connected with all your peers in the social media space. So one of the things I love about Social Media Marketing World is everybody here is intentionally friendly. Like everyone, it doesn't matter how many followers you got, whether you are new or you've been around for a while, everybody is connecting with everyone. Like everybody is just normal and like people want to connect, they want to talk to you, they want to just hang and it just feels like a really safe space for everybody, whether you're new on social media or you've been around for a while. This year for me has been like my absolute favorite because not only could I learn all about Social Media, but I was able to come into space and learn about AI and how I can apply it to my business. So if you're thinking about coming to Social Media Marketing World, you definitely should. Get tickets at socialmediamarkeningworld.info And welcome to the AI Explored Podcast, helping you put AI to work. And now, here's your host, Michael Stelzner. Hello, hello, hello. Thank you so much 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. Today, we'll explore how to get AI to manage itself autonomously. My special guest is an AI Operations Strategist, who helps teams use AI to improve how they operate. Her group coaching program is AI Operator Boot Camp, where she helps people deploy AI. She's a returning speaker at AI Business World, part of Social Media Marketing World, 2027. Rachel Woods, welcome back to the show. How you doing today? Thank you. I am so excited for this conversation. It's always so fun to get to Neur.in Hangout. Sweetness. All right, well, let's begin with the biggest misconception you see when it comes to Autonomous AI, AI agents, all this kind of fun stuff. What do people not understand? What's one of the maybe lies we believe? I think let's just start with, when you think of Autonomous AI, you probably think of. Agents. Right? Agents is like the big thing these days. What's really interesting about agents is that everybody thinks that the hard part is building them, setting up. Maybe that's on your bucket list of like, "I need to learn how to build agents." But the reality is, and I hate to break this to you, even if you go through the effort of learning to build agents, doesn't necessarily mean that you're going to unlock Autonomous AI for yourself. And why not? Well, it's because to actually have AI be Autonomous, it needs to be something that you can actually trust that it's going to do work the way that you want it done, so that you can have it running in the background. You don't have to babysit it. And unless you figure out how to unlock that you're actually trusting it, then you're never going to get to where it's actually Autonomous. You're just going to have something going around running off, doing random things, and then you have to go and undo that work. And I know a lot of us have run into this, right? It's the. And I love using the human analogy, even if you have a person that you've asked to go do something, and what they run off, and they go do it wrong, because you didn't necessarily clarify your expectations. The same thing happens with AI agents if you don't really figure out how to unlock that trust. Love it. We're going to talk about this today, and what is waiting for everybody on the other side when it comes to these Autonomous AI systems, if they pay attention to what we're talking about today? Said another way, if everybody does what you advocate, what's on the other side waiting for them? So I'll warn you that parts of today's conversation may not come across as really sexy, because we're going to talk about process and operations, and those typically aren't super sexy things. But frankly, what is sexy is the outcome, right? Which is that the way that I run my life and my work these days is I have AI, truly autonomously running for me in the background, or I don't have to jump in and fix the work. The way that I do that is by getting really clear about how I want it to do the work and setting it up and testing it before I hand it off. We'll talk about how to do that. But nowadays, on average in a single day, I'm getting AI to do about eight hours of work autonomously for me in the background work that I don't have to redo. And I literally just go on like my morning walk, and I'm checking in on my morning routine of all the work that my AI is autonomously doing for me. So we're going to talk through like literally exactly how you can start going down that path and setting that up for yourself. Well, and just as we were about to record, tell them what you told me about what your agent reminded you this morning. So one of the questions that my guest was like, oh, I know we did the prep. Did you get a chance to kind of think more about what you want to talk about today? And frankly, I had forgotten until this morning, my AI reminded me of, hey, you have an interview today and you should go through and make sure that you're prepared with the outline. And so that was the topic of my morning walk. The way that I have my morning routine set up is my AI is going around and making sure that I'm doing the work that I need to be doing. So yeah, that's how my morning started. I love it. I love it. I am so excited about this because I am not here yet. And I'm guessing most of the people listening to this podcast are not yet to the point where they have just trusted AI to go do things on their behalf. Everyone is listening. It's probably a lot like me like I have to initiate these things and I have to check all these things. And I'm really, really excited about the possibility that this is actually a reality. So as we get into this, I want to start with this question. Is there any foundational stuff that we need to talk about before we get into kind of the technical and how to do this? So one of the first things is that if you're trying to get AI to do more for you and you've been disappointed with the results, it's not that AI can't do that work. So this is a really common thing that I hear from people is they're like, "Oh, like it must just not be good at blah, blah, blah, blah, blah." So I have not personally met any single type of work. And we've worked across all different industries, different roles, all this stuff. But I couldn't, what we call playbook, which is about getting really specific about what you want the AI to do inside that work and setting it up so that it does it. I have not met a single type of work that this couldn't be applied to. It's just that you might not have that skill set yet. And so I want you to kind of open up your mind to this being a new tool in your tool belt that you can add for you to actually be able to get more out of these tools, rather than feeling like that's something that's not possible. So that's one thing. It's just making sure that you really commit to getting specific about the work that you want the AI to do. Love it. Okay. Talk to us a little bit about this trust thing. And kind of when we were prepping, you said, "You don't need a baby set this thing as either." So like, talk to me a little bit about those two things. Just because I think it's hard for a lot of us to get to the point where we feel like we can't babysit AI, you know? Yeah. I mean, so let me ask you this. Like, you know, when's the last time that you handed off work to somebody on your team and they didn't do necessarily exactly what you want? And you probably felt frustrated. 100%. 100%. Yeah. Yeah. So one of the favorite quotes that I've now lived by is that any time that you're frustrated, the root cause of that frustration is just a misalignment of expectations. And so if you are feeling frustrated by the output that you're getting out of AI, to where again, you're having to go back and babysit and fix stuff, it's just because you haven't set clear enough expectations up front. And so I like to ask and think about like, if you were to really sit down and think about how you would ideally teach, let's just even say a person, how to do work the way that you wanted to where you knew, as long as they followed what you said that you would be able to trust and rely on that they have that work covered and you wouldn't have to babysit. Like, I'm sure that you and your mind have an idea of the clarity and level that you need to get to, right? 100% and I have people that work for me that I trust and I do not babysit, right? And it took a long time to get there. Yeah. A lot of defining expectations, right? Correct. And to fast track that, you said it took you a long time to get there, right? But to fast track that, what do we do? We sit down and we get really clear about our expectations. AKA documenting work, documenting processes, and outlining really step by step of what we want done. And so that same exact mindset applies to AI. If you're in a place where you can't trust it to go off and do work correctly, then you need to really boil down your work into the step by step process or what we call a playbook so that then you have the AI following that playbook exactly. And you know that as long as it follows exactly, you're going to get the result that you want, which means then you can trust it. You don't have to jump back in and babysit. Okay. So there's two things that I'm hearing you say, which I think I just want to, you know, talk through just a little bit. So number one, if you want AI to autonomously do work for you, you're going to need to get to the level of trusting it to do the work. And the only way you're going to trust it to do the work is if you properly train it on how to do the work and it does it consistently over and over again, just like if you were to do this with an employee and eventually you let go of it and you just trust it to do it. And the key thing of no babysitting is a big deal because when we, especially those of us that have teams working for us and by the way, for those of you that do not have teams working for you or even anyone working for you, this is important. Anything that you use, you have to ultimately trust it, right? To be able to do the work that you want it to do. And ultimately if you can get to the point where you don't have to watch what it does, then it becomes autonomous and when it becomes autonomous, you become free. And this is the key thing. The moment you become free, you become unchained from your desk, right? And you're able to go on these walks like Rachel does in the morning for a while. Who knows how long, and you're able to just listen to podcasts or think about higher level strategic work. And this is really the promise of what I'm hearing from you, but to get there, we're going to need to develop something that we're about to talk about next. Do I have this right? 100% and I would even go a step further to say that like, what's the promise of AI, right? It's that we can get manual tedious work off of our plate. If you don't figure out this next level, you're artificially capping what you can even get out of these tools. Like they're designed to do this. And so if you're not yet at this level, then you're not getting the full value out of AI. I love it. Okay. So you've already mentioned the concept of a playbook, which is really where we start. So why don't you describe in a little bit more detail what the heck it is and why it matters. And then we can maybe give some examples and tips and stuff like that. Yeah. So let's actually even just, I'm an example person, so I like to talk from examples. Yeah. So the way that I run my mornings is in my morning routine playbook, which actually what that playbook does is it collects the outputs of a lot of other playbooks and triggers other playbooks, which we can get into a little bit more. We call that stacking, but let me just talk about one of those playbooks, which is my email inbox triage, which is like, man, getting to inbox zero, right, especially if you're not somebody who already has a system or process to do that. Like, Mike, I don't know if you're an inbox zero, but it takes a lot of work to be inbox zero, right? But if you can play book it, right, if you can set it up so that AI is doing it autonomously for you in the background, now I'm an inbox zero every single day and actually it runs every single hour. So I'm just always an inbox zero. In order to get there, though, I didn't just say, hey, AI, I want you to help me get to inbox zero. And yours is connection to my email, go. That would be bananas because I would never trust it, right? I would have to go through and I would have to make sure that it's tagging things the right way. I have to give it tons of feedback. I would probably be contradicting myself a lot because I hadn't really thought through how to teach the AI properly, how to manage my inbox. And so that's where you have to start. You have to start by really thinking about how do I structure this work so that AI could follow it? And that's what a playbook is. It's really a step-by-step process to playbook something. Most of the time, we don't actually have this structure figured out yet. So the hard part is figuring out the structure. And to do that, the way that we start is with what we call a playbook outline. So you don't have to just start like you're writing a book and you just are starting from chapter one and free writing and said you can really think about an outline, your playbook or outline the process that you want AI to follow. And then you can fill in all the details that it needs later. To do those outlines, what that looks like is you need to find just four things. First, you need to define the trigger or what kicks off the playbook. What's going to start the playbook? So in my email inbox playbook that I'm talking about, I already mentioned, it's actually running every hour on a schedule. That's a type of trigger. You could also have the playbook be triggered from an event or something that happens. It could be you messaging the AI or a new calendar invite showing up on your calendar, something like that. But you need to figure out when is this work even going to start? When do I want the AI to turn on and do this work? And once you have that defined, you need to think about what are the inputs to the playbook. And I like to define this as what do you need to give the AI so the AI can do its job? And so for my email inbox playbook, what it does is the input is all of the new emails. That's what it's going to do the work on. Then the third part of a playbook outline is just defining the steps of the process. So I really sat down and thought about if I were to teach somebody, ideally to go through my inbox and follow kind of my way of making decisions, my way of tagging things. What are those steps? And that became the steps of the playbook. And then the fourth thing is the output of what is done look like, what is the purpose of running this playbook, which, again, for email inbox is that the entire inbox is triage and handled. I have the responses drafted. I have it escalate things to me that it can't figure out so that I can look at those later. But whatever the output of your playbook is, you need to define that. And so once you do this, and those same four components, for any type of work, that's actually all you need to do to really figure out how do I get now AI to be something that can be doing this work for me instead of me having to do it myself by hand. Okay. So I heard that there's four components to these playbooks. Number one is the trigger, which is what's the time or the event that kicks it off. Number two is the input, which is the information, the data in this case, who is email in the bot, new emails in the inbox. And then the steps are the whatever it's going to do with that information. And then the output is if it's going to, for example, create a report or ping you on your phone or whatever, right? That's the other part of it. Okay. So what's the hardest part of this process for most people? Is it the steps? It's actually oftentimes defining what are the bookends of the process. So one of the common things that we see is like, and this is really common, especially if you're delegating something to a person, the same trap happens of being like, okay, I actually want AI to run my entire morning and then thinking about your morning routine as the playbook, a better approach is to chunk it up into smaller pieces where you can really get one part of the process or one part of the work working and then stack these things together. I'll give actually another example that I think especially in the marketing world is really common is when you think about a marketing campaign, you have a lot of different deliverables that you're building out for that campaign, right? You have the briefs at the beginning, then you have maybe, maybe has multiple channels involved in that campaign. So you have the emails, if the social media posts, you have the landing pages, you have the follow up lead magnets, you have all these different deliverables. And so a common trap is to be like, okay, let's build like a marketing campaign playbook. And now if you've done that, you've not bitten off a huge chunk of work that you're not going to get value out of AI doing correctly until you get that whole thing working. Put differently, you're not going to be able to trust it to go off and do something autonomously until you get that whole thing working. And so it's a lot better is to think about it in smaller chunks where you have the AI help with individual pieces. So in this case, it would be individual deliverables inside the campaign and you get each of those chunks working. And then you can have a playbook overall that runs all of those together. You know, we're kind of, I guess maybe jumping ahead a little bit, but that's when we talk about like AI managing itself. That's also how you do it. You do this stacking process, which I think we can get into a little bit more. But I wanted to connect that for people, yeah. I just want to add some insight here. So I have a AI editor that edits my new daily docu series called Press Start the Road to 3000 and in the beginning, it was taking longer videos and finding the story arc and putting a script together and showing me how it's going to cut the video together. Then I got to the point where I added captioning, right, baked in captioning. So that got a little trickier, right, because now it had to go out there and make sure the captioning was aligned with my lips and all that kind of fun stuff. And it got more complicated because we needed to send it out to Gemini to do multimodal audio checks on it. Then I got to the point where I started adding title sections to each of the sections of the video. So it had to come up with like words that said what was coming next. And then I got to the point where I decided to add music to it. Okay. Now music was a whole new ball of wax. So I had to buy the licenses to the music and train it on how to do the music layering over the entire thing. And then I got to the point where I had to do thumbnails. And then I got to the point where I had to do scheduling. So you can see what I did and I have this now ridiculously complicated skill, if you will, that's got all these little subskills inside of it that can do all these things. And if I just decided to make the perfect editor with all these things, it would have been a nightmare. So I think the lesson that you're teaching here is start small, start with the core thing, and then build on top of it. Is that correct? Totally. This is how you actually unlock this for yourself, right? Because I think one thing that, you know, at least I can relate to is it's easy to listen to all these different ways to use AI and these different approaches. But then turn around and look at your work and be like, man, my work looks nothing like that. How do I actually apply this? And so I want to invite anybody who's listening, if you're like, man, I want to get to this. It's really as simple as picking one type of work that you feel like you could get to the place where it's clear enough that you could hand it off and AI could do it. And just to find these four things, that is all you need to do to start. And then like we're talking about, you can build on top of it. And that's how you get to where, you know, I'm describing, you know, how my morning runs of this. It's not overnight. I just became this person that has this crazy morning routine, you build it brick by brick. Perfect. Okay. So what's the next step? So another side of this, we're talking about how to take like big chunks of work, like entire roles or entire huge deliverables and break those down into smaller pieces, which then should be individual playbooks. But another side of this is whenever you first create a playbook, so whenever you first create a process that you want AI to do, it's really tempting to have kind of an all or nothing mindset where either you set it up and you're like, great, I'm always going to be the one reviewing this and you kind of get stuck there. Or you set it up and you're like, man, this doesn't work without me reviewing it at all. And so then you just give up on the process. What I want to encourage you to think about is again, the cycle of iteration, even for the same type of task. And the way that we do that is we have a process we call the craft cycle where we apply these same five steps for any playbook that we build. and it helps us get really clear on what we're trying to do and actually end up, I like to say, is like crafting ourselves out of the loop and the process. But we don't do it overnight. And so what that looks like is the first time that you build a playbook, you can go through and do the steps, which just really quickly are first actually getting a clear picture of what you're trying to do. You have to start from knowing what you're. Well, let's not reveal all of them. Let's go through them one at a time. Okay, so start with that one. So the first is getting a clear picture of what you want AI to do and really documenting, again, the full ideal process that you are trying to hand off. So starting from that outline, but then you're gonna have to get more specific in some of those steps with more nuanced instructions, especially if your work is nuanced. But that's the first step is getting clear. Well, does this presume we've already got all these little playbooks made? Is that kind of what are we presuming that this juncture we've created a whole bunch of playbooks that kind of make up the process or no? So this would be even from your very first playbook, you can use these five steps. Okay. And then what you're gonna do is you can actually cycle through this and use the same five steps again to improve the same playbook and get yourself out of the process once you've done that. Okay, so first, you have to have a clear picture of what exactly the output that we're looking for or what is the clear picture here elaborate on that a little bit more. A clear picture of what is the work that you want the AI to do. Okay. So it's really about writing the playbook in its most ideal form. Okay. Then once you've done that, the R and craft stands for realistic design. So the first time that you're building a playbook, this is where you might want to be realistic with yourself of which things are gonna be easy for you to get the AI to do and which things are gonna be a little harder. And you can keep yourself in the loop or keep yourself doing actions in that realistic design phase. And the reason you do this is because you want to optimize for getting the playbook done and getting it working at some level. So then you can learn and improve on it and do another cycle where you get yourself out of it. So being realistic is the second step. And that's kind of like another word for that is like scoping it down. I think a good example of this for me at least was this video editing stuff. I did give AI access to Gemini to be able to effectively listen for accidentally cutting words in the wrong spot and stuff like that. But I knew that I still needed a human to watch it every time because there was just some stuff that just needed a human eye on it. So I just decided to get it close enough with this. And then me as the human would always go in on a pass and say, hey, at this mark, you need to cut this kind of thing. That's probably what we're talking about. Something along those lines, right? Totally. And that's where like trust the process of keep yourself in the loop for at least the first version. And another reason to do this is I say this is really common as well is like, if you run into a little bit of a row block, it's easy just to band in the whole thing, right? And feel like it's like not worth figuring out. But AI is getting better all the time. And so something that wasn't realistic for AI to do in the past may very well be realistic for it to do now. If not, I'd like to say in six months, you never know, right? And so you can get value out of having AI help with the rest of the process in the meantime. Love it. Okay, good. So realistic design, what I'm hearing you say is don't try to have it do everything. Identify the easy things that it can do. And maybe the things that you know are going to be hard and you take the hard stuff on and let AI do the easy stuff. And eventually over time start removing some of those hard things off your plate. Is that I hear that right? And think about it as like something that you run in a cycle. So again, you're starting from scratch. Like you haven't had AI do this type of work yet. I'm going to get a clear picture of what is everything that you want AI to do. But then the first version of this, let's be realistic and let ourselves kind of scope things down. Because then the next letter of craft is a, which stands for AI. That's when you actually set it up and you start using it. And using it is where you really are able to start making sure that it's able to do what you expect. And that moves into, and I don't want to jump ahead too fast. But the most important letter out of the whole craft cycle is F was changed for the feedback loop. So when we're using it, well, these two are kind of intertwined. So go ahead and just explain how they work together. You can see I'm, I get so excited about the craft cycle. I'm like, this is what I do all day every day. So I'm like, yeah, this will change how you solve problems with AI and set up processes if you literally just follow these five steps. It, it makes it so simple. So after you set it up, the F is feedback loop, which is where you start running it. And what you need to do is you need to just start noticing, where is the AI doing the work correctly? And where is it doing it incorrectly? And you actually want to look for both of those things. Most people just pay attention to what the AI is doing incorrectly and try to fix those things. But if you see things that it's doing well, you should actually just like an employee reinforce your AI. Oh, okay. Explain why that reinforcement learning thing is so important. Like explain that a little bit for a non tech techy people. What I mean when I say this is you're going to run your playbooks with AI and you're going to see like, oh, wow, it actually included a summary at the front of this document, even though I didn't ask for a summary, I like that. But now if you don't tell it to do that next time, it's not guaranteed to do that again. Because AI is this, you know, great to think it's always trying to guess what we want unless we tell it. So what you should do is you should go back into your playbook and say include a summary at the beginning. Or you can tell the AI, hey, I love that update the playbook so that you know to include the summary next time. But you want to make sure that you're actually reinforcing the good behaviors as well as the bad. So same thing, including the summary and you're like, oh my god, total waste of space. Never include a summary ever again. Great. That should go back into your playbook. And that's why the feedback loop is so important because that's how you actually start to build up the trust and specificity that you can have in the AI is actually doing the work correctly. Hi, my name is Rita Milliken. This is my third year at social media marketing world. And there is a very distinct reason why I keep coming back. And that's because of the amazing speakers that are here. Everyone talks about networking and the connections. And honestly, it's fantastic. But the speakers, they know what they're talking about. And even if you think you know everything about AI, I promise you don't. You're going to learn something new here, just like I did. If you keep telling my clients, I will never stop learning. And that's because I come to social media marketing world. Get tickets at socialmedymarkeningworld.info. Okay. Fellow humans, we don't even tell sometimes our people what they do good. It lists. If you're like me, I'm like, I'm like, I have like a metal detector that finds what's wrong. And I just say here, this is wrong. Fix this. But I think this is so important that we actually, especially with AI, because we're moving towards wanting to give it. So we trusted to do the work, right? Letting it know, hey, this is amazing. In these areas, I love this. Make sure we always do this. And in these areas, we're almost there. I think we need to enhance them. In these areas, I never want to do it again. I mean, that's kind of what I'm here and you say. Is that correct? Totally. And then that's also how you're no longer, I can just like crossing your fingers and hoping that the AI is going to do what you want. It's like literally documented in its playbook. All it has to do is follow the instructions that you've given it. Okay. You've already kind of hinted a little bit that Ask AI to update itself. Like let's talk about that a little bit for some people that don't even realize it can do that. So explain a little bit about how that actually works. So there's a little bit of nuance here that it does depend on the tools that you're running your AI playbooks in. One of the things we haven't talked about yet at all is tools. So taking a step back even this playbooking concept that we're talking about, which again, a playbook isn't like a type of AI tool. It's a method. It's a way of using AI for it to follow processes like we're talking about. You can apply this method in any AI tool that can follow instructions. So I've playbooks running, as we speak, in chat GPT, I've playbooks running in cloud, I've playbooks, some playbooks running in Grockbot. A lot of the students in our program are, you know, in Microsoft co-pilot, they have playbooks that they're running in there. Gemini, any tool, air table, any tool that can follow instructions, which is all of them, you can run playbooks in them. Now, some of these tools have different places where you can save these instructions. One of the most common and I'd say easiest ways, if you're just getting started with this, is to use something like cloud where you're saving your playbooks as skills. And you may already have skills for other stuff, right? Like, kind of saved instructions. This can just go right alongside them and these are now your step-by-step playbooks saved in those skills. And what's great about cloud skills is cloud can actually edit itself. Yes, that's huge, right? And so then as you're running at these playbooks, you can tell it to make edits on your behalf. And I will say, my one warning, you know, there's always like the disclaimer in the warning. This is really important. If you're going to have AI edit itself, you either want to make some constraints on what it can change, right? So we have a framework we use. It's like major and minor feedback, like minor changes go for it, but major stuff, like let's talk about it, you know? Or you just need to make sure you're actually like looking at and reviewing because otherwise you're going to have your AI just like goes off the rails unless again, you have a kind of a playbook for how to update the playbook, which is a lot of what we do. But it can literally edit itself, yeah. Yeah, give a little tip folks. I use cloud code and I use cloud, regular cloud, and co-work, all of cloud, I'm a cloud user. I have gotten to the point where I have asked it to make a backup before it updates its skill. This is really important because sometimes it will overwrite something that's really, really important. So I just now in any of these skills, which are also now known as playbooks, the automatic rule is that before you make a change, make a copy of it. And then also it prompts me to download if I want to a copy locally on my computer. And this is really important because if you lose your skill, you're kind of in trouble. Do you have something like this as well. The best way is actually to not write your playbooks in your tools. Oh, elaborate. So what we do is we call it own the playbook rent attack. Oh, you should actually be writing your playbooks and storing them and saving them in Google Documents and Google Drive or a notion or in some other document storage that's not your AI tool. Okay. Because then you have it as an independent asset. It's just like, you know, like sometimes you get worried about like the people on your team that might have their stuff in only some like personal drives. Like that's what we're letting our AI to their computer hard drive. Yeah. Yeah. What's your preferred tool here? Because what I like about this is this opens up the possibility for you to switch between clawed and chat GPT. For example, code x versus code if for whatever reason you run out of tokens and they could use the same playbook and presumably do the same work, right? Exactly. So internally as a team, we do use a mix of Google Drive and we use another tool called GitHub. Now they'll say we're in the more advanced like our whole business runs on playbooks. That's why we're in GitHub. But Google Drive is really easy because then you can just in clawed tell it where to go find certain playbooks and then things are always kept up to date. Well, and for folks that don't know GitHub is owned by Microsoft. It has automatic versioning, right? So you can go back if you accidentally overwrite something. We're using GitHub for our software product and I'm using it for some stuff. But I'm actually just using clawed skills mostly, which I think live on my hard drive because of cloud code, you know, which is cool because I got a Mac backups on my computer going to my different sources and stuff like this. So this is a worthy conversation. So we've gone down a little rabbit trail. We're in your process called craft at the top. We talked about clear picture. What do you want to do? Realistic design, start with something relatively easy and optimize it for getting it done. And then you can always improve it later, which we explain in the feedback loop. AIFI is actually setting up and using it. And the feedback loop is the F in craft. And this is where you are actually basically giving it positive feedback in the areas that it's excelling with new things and giving it constructive feedback or criticism on the areas that it's making mistakes. And then we went down this rabbit hole of T for tools, which is not the last letter in craft, although it could be you could do tools slash team if you wanted to. But in the tools, little rabbit trail, we went down. The best thing you can do is use skills and the good news is every platform uses skills. Even Gemini has announced that they're going to be retiring gems and they're going to skills. So everybody is using skills. It's the obvious easy first choice I would imagine to use because AI just knows how to update skills. You don't have to do anything special. But if you want to go more advanced, Rachel's saying you can go to a Google Drive. But if you have super high levels of security, you might not be able to go to a Google Drive. Like I used advanced protection on myself and it's not going to allow the Google Drive to be connected in. So in this case, something like a GitHub repo or GitHub repository where this could live would be beneficial. And another benefit to this is you can have different team members on different accounts without having a team account access these skills. Is that correct, Rachel? That's right. And then you can also have multiple people on your team contributing to the playbooks and being able to collaborate on making the best possible processes that you trust that AI is going to do a good job. Yeah. And we could get on a big old rabbit hole on what if they overwrite we're not going to do that. But let's talk about teams because like that's the next that's the last part of the craft process. Yes. So the team craft stands for team roll out. And this is the most overlooked. But this is where you actually get the benefit and the ROI of whatever you set up, right? Which is that if you don't end up formalizing how you roll it out, then it's going to be something that sits on the shelf. And so there's a lot in here. A lot of like kind of change management and different things. But one of my favorite tips is to really be thoughtful about habit stacking. Are you going to have it stacking? Only because you told me. So it's one of those like I think that the classic one is if you want to like get really good at pushups, the way you should do is not just be like, I'm going to do pushups. And instead, you should be like, every time I brush my teeth, then I'm going to do pushups. Oh, okay. That way you never forget to do pushups. And so apply the same thing to using your playbooks is, you know, you need to make sure that it's clear like when in the broad scheme of things, you're going to actually use this. And then also make sure that anybody is using it is trained on how to start it, how to review the work, what the expectations are, all that stuff. But once you've done this whole cycle of craft, and these can go really fast. Once you get into it, the key is, and when we're talking about, again, zooming out to actually getting to autonomous AI, because you can run this process just for setting up AIs that you're still using every day, is that you can run a second cycle on the same playbook. Once you've started using it and you understand what it's doing well and not doing well. And that's how you actually can craft yourself out of having to be the one reviewing. So wait, wait, just to be clear. So we're moving on to the key to the whole thing, right? Which is to now, how do we get ourselves out of it completely, right? Is that where we're going with this? Okay, love it. And so the way to think about this is, again, in your first version of setting up the work, you probably had to plug yourself back in to that playbook, right? Because you need to make sure that the work is good. You're checking in or whatever. But as you're running it, you're going to understand what are the things you're looking for? What are the things you're checking? What are the mistakes does it making? And that's going to help you be smarter about the gaps in your playbook that you need to be more specific on. And once you close those gaps, that's how you get AI to be able to do stuff autonomously without you having to check. Is you're able to remove yourself from the process? And so once you have that clarity, start off with another clear picture, and then you can just run that cycle again of removing yourself from the process, changing the playbook a little bit, removing yourself from steps based on what you've learned, the AI needs in order to be successful. Love it. Okay. So the next thing that we're really going to talk about is the thing that I'm super excited about, which is this concept of a middle manager. And this is not a human middle manager. This is an AI middle manager. This is going to be like the big unlock for a lot of people. So for anybody who's already managing on a team or has managed in the past, I will say you may not realize that you already have this skill set. You just need to apply it in this new way. And the skill set specifically is being able to break down work in a way that then you can delegate to other people and then manage the outcomes and be able to be sitting over on top of the system instead of the one in the weeds doing all the work. If you haven't managed before, great. This is the skill set to develop, right? It's to figure out how to become like the manager of the of the system instead of being the one in the weeds doing all the work. And then as you start doing this, so what happens when you're a manager for the first time is you are able to get other people to be individual contributors underneath you. But you also learn like your process of management, right? You figure out, okay, we got it to do morning huddles. And we're going to make sure that the tasks are all on track. I'm going to make sure that Jimmy knows to blah, blah, blah, blah. I'm going to check in on these things. You end up creating your playbook of management. The same exact thing happens whenever you start getting to that level with your systems is if once you're managing your systems for a bit, you can then playbook yourself out of that level and go to the next level. And you can continuously do that. And that is a very clear and direct path to be able to have the equivalent of like a whole role or a whole team even of work being done underneath you, not by humans, but by AI's. So first of all, what I'm here and you say is, hey, we've got at this point hopefully a whole bunch of AI playbooks or skills or whatever we're going to call them doing all these different tasks for us. But we're still managing over the top of it, making sure that everything is being done. What if we were to create a AI that effectively took that task from us and effectively managed all these other AI's and the tasks that they're doing? Now one of the first questions that comes into my mind is how in the world do these AI's communicate with each other? Because that's the big question. Like how are they actually capable of doing that? Because that's I don't even understand. So it does depend on the tools you're using. I'll tell you how we do it and how I have it set up is you may have heard of subagents. Not really. So subagents is something that Claude has support for, where one agent can actually go spawn off like subordinate agents that can go run playbooks and report back. And so again, we're getting into like the more advanced zone here. But if you have playbooks for those subagents to follow, those subagents are going to go do work the way that you expect autonomously. And then come back to your main agent, your main agent can manage them from there. That's one way. Another way that we do it that can be a little bit easier, especially if you have an or already using a task management tool inside your company, is to actually just like a manager would assign work out to team members and let's say Asana or click up or something like that. Yeah, we use Asana. Yeah. So just like a manager would assign work out to people in Asana, you can have a manager agent assign work to other agents in Asana that you have agents watching your task management tool. Okay. So those agents are triggered by certain activities that happen inside of Asana. Exactly. And then another agent is effectively going in and following up to make sure tasks are completed. Is that what I'm here and you say? Exactly. And it's not just doing random work. It's all following playbooks you have to find. So what's the preferred method here? task management tools because it's really easy for us to kind of see what's going on in the beginning or is it more like agents managing agents within like behind the scenes and you don't even know what's going on. It's one of those that it depends on your use case. So if you're in a team, right, being able to have that audit trail and knowing what the AI did and didn't do and get wrong, you know, all that stuff. So in our company and the systems that we're running across the team, we're using the Asana task management tool method. But for my own personal morning routine, like, that's overkill. I know enough about all of the systems that I have set up that I really just want to talk to my main, my todaying agent, that recently started calling her Pepper. I do, thank you, reach a new level when you do have like a name for your. That sounds like a dog or a cat name. I like that, though. That's super cool. Because you Pepper her with questions, that one. Well, one of my. Maybe this is another side tip that's so fun. I actually just made a tick talk about this morning, which is that you really know you've reached a level with AI when you start making up your own vocabulary. Oh, okay. Because when you have specific terms for stuff, that makes the AI less confused and less guessing of what you're wanting. So one of my favorite things right now is my feedback playbooks that I have where I'm kind of giving overall feedback and observations of the business and the system and stuff. If I call it feedback, it'd be like so confused because I use that word like all the time. So I call it nitpicking. Okay. This is my nitpicking playbook. I highly recommend building your own vocabulary for how you talk to and manage your AI as well. Folks, Rachel Woods is the real deal, which is why I've invited her back on the show and it's why I've invited her to AI Business World. And she is incredible. And these systems that she's built has allowed her to do things that are just beyond my wildest imagination. So Rachel, outside of people meeting you in person at AI Business World and social media marketing world, tell people where they can discover more about you. And if they're interested in possibly joining something you've got going on, like where do you want to send them? Yeah. I mean, first of all, would love to see you in person. It's such a fun time. I'm really active on LinkedIn or Instagram. So message me, tell me what your thoughts were. And I will say we do run a free class pretty often about the playbooking method going way more in detail. And so if that's something of interest, just message me class and I can get you access to that. Awesome. Rachel Woods, thank you so much for sharing your wisdom with us today. Thank you for having me. Always so fun. If you missed anything, we took all the notes for you over at socialmediaxameter.com/a129. Be sure to follow this show on your favorite podcasting app. And if you've been a long time listener, we would love a review. And do check out my other show, the social media marketing podcast. This brings us to the end of the AI Explored podcast. I'm your host, Michael Stelzer. 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 of social media examiner. One more thing we just announced the first speakers for AI Business World 2027 dedicated AI sessions, personally recruited speakers, pitch free training, designed for implementation, and virtual tickets cover both AI Business World and social media marketing world. See the lineup at AIbusinessworld.live.