Setting the Stage for Agentic AI: A Practical Framework

I'm grateful for marketers like you. Not the ones waiting for their boss to tell them what to learn, but marketers who actively plan for their future. Because you listen to this podcast, you're already ahead. You're seeking to understand AI instead of waiting to see how this AI thing turns out. But here's what I've learned after a decade of running conferences. Interest doesn't create results, implementation does. That's why we created AI Business World 2026, where you'll master AI skills that make you indispensable, where you'll get your questions answered by experts, and where you'll connect with over a thousand marketers who are implementing AI right now. Years from now, you'll look back at this moment and remember this is when you got ahead. Head to AIbusinessWorld live and secure your competitive advantage.

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 Explorer podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzner. This is the podcast for marketers, creators and business owners who want to know how to put AI to work. Today's episode is spectacular. Agentic AI claims to be a game changer that will transform businesses of all sizes. Lots of AI companies claim to offer agentic AI capabilities, but is that really true? In today's episode of the AI Explored podcast, we'll explore how to prepare for agentic AI and the tools that will get you there. My special guest is an AI expert who helps businesses implement AI. He's the chief data scientist for Trust Insights, a company that provides AI consulting workshops and customized solutions. He's the author of a new book, Almost timeless, the 48 foundation principles of generative AI. Christopher S. Penn, welcome back to the show. How you doing today?

Thank you for having me. Thank you. You forgot to mention Social Media Marketing World and AI World Business World 2026.

That's true. Chris will be there. That's going to be awesome. So, Chris, let's start with my first question. What's one of the biggest misconceptions you see when it comes to agentic AI?

There's a lot.

Because it's a. It's a hot topic. It's a hot topic.

It's a hot topic. And here's the thing. There' lot of people who are using this term, banding about this term, and it means different things based on who you're talking to, which creates the opportunity for Confusion. And that in turn tends to create fertile grounds for snake oil salesmen to come in and say, everything's an AI agent, we've got an AI agent, this is an AI agent. And because there's so much confusion, regular folks who want to get started with this stuff have no idea where to start. For example, this is not a snake oil example, but there's a definition of an agent itself. And this is not AI, just an agent itself. If you go take your phone and you go to someplace like social media marketing world, right, and says get on the WI fi, say get on the WI fi, put in the password, you're on behind the scenes, there's four different agents on your phone that are working to get you connected to the Wi Fi, negotiating your IP address, doing, you know, BGP4 this, that and all these technical things. You don't see any of it. All you see is yes, you're on the WI fi or no, you're not on the WI fi, right? That's what an agent is. It does stuff. Think about where else in the real world we use this term, a travel agent takes care of a lot of the minutia so that you don't worry about it yet obviously you pay for a real estate agent takes care of, you know, either buying or selling your house for you. When people talk about agents of in AI because they've conflated this massive spectrum of everything from here's a simple workflow all the way up to a truly hands off self driving app, right? No one knows what you mean when you say AI agent or agentic AI. And so that to me is the biggest misconception of all is there is no clear definition. It varies based on who you talk to. And as we saw with AI optimization, aeo, Geo, you know, EIEI or whatever the news, because there's so much noise and so many people talking about stuff without having clear definitions, there's a lot of folks who are taking their existing SEO practices, scratching out SEO, putting AI optimization on and adding a zero to their price tag. Well, the same thing is happening in agentic AI, right? People are taking their custom GPTs that were the hot thing two years ago and saying I was an agent, add a zero. And that's sort of that to me the biggest issue that we are having in the space right now, we're going.

To define what you feel like agentic AI really is. But before we go there, I want to talk about the benefits when done well to your standards, right? To your high standard, right? What's the upside? You know, what are the positive upsides when agentic AI is implemented and fulfills its promise?

Here's how to think about AI. When we think about a tool like ChatGPT or we think about models like GPT5, those are engines. Those are engines that are very powerful. And the way we've been using them is sort of like directly pulling on the chains and levers of the engine, right? When you prompt, when you open up ChatGPT and you're typing in your prompt, you are essentially controlling the engine directly. And a lot of people are saying, why can't I use chat GPT to check my email and tell me what emails I need to respond to? Why can't I use it to do this, why can't I use it to do that? We've all come up with our various workarounds, copying and pasting and doing this and that and stuff. Agentic AI means taking the engine, which is a model, and building the rest of the car around it, right? So if you think about what's in the rest of the car, the seats, right? The body, the radio, the seat belts, all this stuff, those are the connections to different systems that you care about. Your CRM, your marketing automation software, your accounting system, your email system. And that gives you the ability to scale because no longer are you the one having to manually type everything in. If you've built an agent that's really proficient, it drives itself right? So you, it says, hey, you show up Monday morning, Mike, here's the five emails that you need to respond to today. You've built all the infrastructure to check your email to understand who the important senders are, to read the context, to say, hey, this email contains money or this email contains a bill, right? And that's where these things really grow up. When for I'll give you a practical example. I do a lot of stuff with coding and it is now so nice to be able to use agentic coding tools, tools like Claude Code, for example, or OpenAI's codecs and say, here's the work plan, here's what I want you to do, think it through, tell me where I've gotten it wrong and go do it and I'll come back in 45 minutes. It's kind of like going from being a power user of AI to being a manager of AI, where now I'm managing a team of these agents that are off going, doing their thing, right? And I can just come back when it's ready. Another example that everyone should be familiar with now is deep research agents, right? So you go to Google's Gemini Deep Research or Perplexity or whoever and you say, hey, I want some research on which conferences should I attend next year. Here's my budget, here's the places I can and can't fly to, etc, and it goes and it does the thing, you don't participate. When it comes back it says, here's the five conferences you should go to. That's the power of agentic because it allows you, the operator, to scale, to do much more than you could in the same way that a manager who hires more competent people can scale their business.

Love it. Absolutely love it. Okay, so what are the very basic options we have with agentic AI? And just so I think I understand, let me just define what I believe I heard you say agentic AI is the ability to have AI autonomously take actions for you. Is that kind of an easy way to define it? Do you want to enhance that a little bit? So maybe just clearly define what the heck it is and then let's talk about like what our options are because I know there's three of them and we're going to drill into all of them.

Yeah. So again, going back to agents in the real world, right? A travel agent, a real estate agent, they go and do stuff. Yes. They ask you for your input. Right. Real estate agent is not just gonna go buy a house for you on your behalf. Hey, by the way, I bought this for you already. You'd be like, I'm sorry, you're fired. There is still that integration with you, but they go off and do a bunch of things. And so a truly agentic system autonomously goes off and does tasks and comes back to you with the results. Right. Just like delegating to a team member, it's, it really is just another form of delegation. So that's what an agent is. I call them self driving apps. Right. It just goes and does this thing and comes back and the thing is hopefully done. Now to clear up that whole snake oil thing, think about how we approach as marketers, product market fit, right? There's three levels of product market fit. There's done by you, done with you and done for you. Done by you is like a recipe. Guess what? You do all the work, you go out and buy ingredients, you go, you follow the recipe. You gotta have a kitchen, you gotta know how to cook. But it's very you doing all the work. It's very low cost because you're doing all the work. And at the end depending on Your level of skill, the tools, the ingredients you might have inedible meal you might not. Done with you is where some of the work has been done, some of it is not. If you think about in the cooking example, this would be meal kits. You talk to, you know, a meal kit provider. A box arrives on your doorstep with ice packs, you unpack it and you stick stuff and you follow the directions and it's, it's like just put this container in the oven for 45 minutes and your meal is done, right? Or you know me, frozen TV dinners at the grocery store. A lot of it's been done. Some have been done for you, some, but you still have to do. And then done for you is when you go to a restaurant, you sit down, you say, I want a steak, and they come out and they bring out this thing. As the product market fit framework goes up, you do less work, but you pay more. There's more work being done by somebody else. That framework works really well with agentic AI. Done by you is chat GPT, right? There's the open it up, there's the blank window and you have to do all the work, prompt it, conversation, upload data, this, that and the other thing.

And let's go through these one at a time. Okay? We'll get to the done with you and done for you. So, okay, just to reiterate what I'm hearing, you say this is a great framework. Done by you is kind of how most of us today, except for you, Chris, of course, are interacting with AI. Done with you is somehow some stuff is kind of fully baked, you know what I mean? Like you, like you mentioned maybe there is also some assisting with humans. Who knows, right? I mean, that's possible. Like maybe, I don't know, maybe not. Because when I think of done with you, I think of like I could hire a chef who could come into my kitchen and teach me how to cook.

Cook.

I don't know, I'm just thinking creatively here. But done for you is the most expensive, right? And that's the one where you just show up, you order from a menu and then 20 minutes later the food comes out. So let's start with the done by you. So we're doing this under the framework of agentic AI. So is done by you offer any kind of agentic anything at this point and if so, help us understand that and if not, help us understand how it is a setup and for the higher levels.

So there are, there are in tools like ChatGPT and Claude and stuff. There are some pre baked agents that are available. The most obvious one is the deep research tool. Right? You say I want a deep research report on this. It's right there in the Chat GPT interface. And you can say, you know, I want this or the Gemini interface or what have you.

So you just have to know how to prompt it basically, right?

Essentially, yes, you have to prompt it. Well, we have a whole framework for that. If you go to Trust, Insights, AI slash casino, it's called the Casino framework stands for context, audience, scope, intent, narrator, outcome and it's free. There's no forms to fill out or anything like that. But if you long as you prompt it well, and scope is the most important of all of them to say, like, you know, I want research about this topic from these sources. The agent will go out and do that for you and you will see because this has been the trend for all of 2025. These AI providers to have these web interfaces, these consumer interfaces are trying to build more and more, you know, baked in agents so that there's less for you to have to do because they recognize you're going to hit a limit right. To your skills. You can't build, the average person can't build a deep research agent on their own.

Well, and not only that, but you're going to cost them a lot of money every time they have to prompt the darn system. Right. So let's just spend a few seconds talking about the importance of asking one question at a time. Because when we were prepping for this, this is something that you wanted to kind of reiterate to everybody when you're just basically working with these systems and having a conversation.

Exactly. So there's three types of prompts that I think are, bottom line, you have to learn how to do for any use of AI, whether you are just, you know, just using Chat GPT day to day or whether you're building agentic systems. Number one, when you are starting out on a new task, no matter what your prompt is, end it with ask me one question at a time until you have enough information to successfully complete the task. Oh, okay, ask me one question at a time until you have enough information to successfully complete the task. Because AI systems are tuned on three basic imperatives. Be harmless like don't tell people how to do very bad things. Be helpful, follow the user's instructions and be truthful if you can be, which is kind of a crapshoot. Helpful is the most important imperative. So if you say to a tool, hey, help me write a small Business strategy for my company, right? And it just goes right into like an over eager intern that's had three cups too many coffees. Like yes, here I go, here's your business strategy, right? This is exactly what you want. And you're like, no, that's super generic. It's not very helpful and it's not tailored to me. If I say write me a small business strategy for my business. Ask me one question at a time until you have enough information to successfully complete the task, it forces the overeager model to say, so what is your business? Who are your customers, right? What do you sell? And it will go through lead you through this process of gathering the information until it thinks it knows. Because what it's doing behind the scenes is it's saying, okay, I know what a small business is, I know what a strategy is. I generally know what pieces should be there. And this user has not given me any of these things. So it will go through that and then you get the information out of you. So that's number one, one of the most important things you can do. If you do Nothing else that will 2x your AI results immediately. Second, at the end of a conversation, say, recap the entire conversation as a set of system instructions for the next time using your prompt engineering knowledge. Recap the conversation as a set of system instructions using your knowledge of prompt engineering for the next time. What that's going to do is consolidate the entire conversation you've had and all those questions and answers. Here's the thing, AI is better at prompting than you are that I am right. Every model knows how to prompt itself. So if we have it recap the conversation as a prompt for the next time, we dramatically cut down the amount of time it takes to not have to do that cold start the next time.

What do we do with that system output?

Well, there's a bunch of things. At the very least, you copy and paste it into a notebook so that you have it for the next time. But you can also, and we'll talk about this with agentic AI, you can also start to refine it to as the one of the building blocks for an agent. And then the third and final prompt engineering BASIC is avoid asking for the answer. Right? Avoid asking for one answer.

Okay.

These tools are probabilistic. They work in probabilities. If I say what's the best performing social media channel for a consulting business, it's going to say it is LinkedIn, right? It's going to come up with a high probability answer. If you say Give me three to seven different options for my consulting business, for social media channels. You're automatically forcing it to widen its, its internal knowledge and come up with different answers based on probabilities. This does two really important things. One, it makes the model expand its field of probability, which tends to generate better answers. And two, it prevents you, the human, from cognitively offloading and offloading decision making skills to a machine that may not reflect you and helps keep you from getting lazy. When it comes back and says, hey, Mike, here's seven different options and the pros and cons of each of your strategy. You're like, well, now I need to actually think about this as a person and figure out what should I be paying attention to.

Well, and what I really like about this, Chris, is I do this a lot when I'm writing. I will say, come up with multiple variations. But then what I will do is I will, after it's done generating like five or six examples, I will say, tell me which is the best and why. And I wait till after it's generated it and what it does. And this is usually inside of a cloud project, but it could be the same within a custom GPT. It'll go and look at those examples up against its original request and all the data that it has and it will usually pick not its first one, but maybe its second or third or fourth one. But I still have to analyze them all and not take it at face value. And I think this is really important, right, because this cognitive offloading thing is something Mark Schaefer and I talked about when I had him on the podcast. I think the episode before yours, we can get really lazy because it sounds really smart and we gotta be careful, don't you agree?

I'm gonna add one more thing to your workflow there. I'm gonna say you should, you should ask it. What have we forgotten?

I like that too. Yep, I do that sometimes.

What have we forgotten? Because there's a field of probability of things. The thing that AI is really good at is encyclopedic macro views of anything because it's seen all the text on the Internet. And so if you're working with your biases and assumptions, and we all have them, right, we may have forgotten some things.

Okay, cool. So we've been focusing on the done by you and we've talked about two things. Well, actually three things. Maybe, I don't know, I lost track. But one question at a time. Ask me one question at a time. And, and then also recap this entire conversation. Into a system instruction following best practices for prompt engineering. Prompt engineering, right. Those are the two big take homes and that's going to help us ultimately with the next thing I would imagine. Right. It's foundational. So let's explore the done with you and let's explore some of the recommended tools that you're messing around with. Let's start with Opal because I know this is something that a lot of my audience is not going to be familiar with.

Well, let's set the table as to what done with you means.

Yeah, do that.

First is the AI equivalent of meal kits where some of the work is done for you and some of it is not. And you have probably heard of some of these things like custom GPTs. You mentioned Claude projects. Google has Gemini Gems. Those are things where there's some of it's pre baked. Right. If you build a custom GPT or a Claude project or a Google gem, you have system instructions that you may have built in the done by you stage. Right. You have background knowledge or what I call knowledge blocks that's pre baked into this thing so that you have a mini app inside your AI tool that you can use for that specific task. So you might have a Mike Stelzner writing voice of here's how to write like me, here's some samples and stuff and that becomes a gemin Gemini. And every time you get a, you know, you were on the road, you dash off a voice memo, you get back to the office, you just put the voice memo into that custom mini app and say make this, you know, clean this up. But it still has to sound like me. Make it sound like me. And it will go ahead and it will build that, you know, based on the existing pre baked instructions and the pre baked examples will come out with the thing. So that's sort of the, the entry level of done with you where you've built these mini apps within these systems.

And by the way, if you don't have these mini apps, there's a, there's a marketplace inside of OpenAI's chat GPT where you can find these things as well. Right? Which is kind of part of the coolness. Like you can find existing ones but go ahead with the however, I can't wait to hear what you have to say there.

However, those mini apps don't really connect too much and so part of what we want to do with agentic AI is connect them to other things. So there's this whole raft now of what are low code or no code AI workflow tools. So there's one called Opal from Google, which unsurprisingly talks entirely to Google's ecosystem. You can use it to connect to YouTube and your Google Drive and your Gmail and stuff like that, and work within the Gemini ecosystem and push data out to like Google Docs and so on and so forth. It doesn't integrate to much else than that, but if you're a Google workspace shop, that's perfectly fine. There's no infrastructure to host. It's. It's kind of a nice workflow system.

Before you go to the others, real quick, can we just talk about Opal for a second? Because a lot of my audience, they are small businesses, they tend to use the Google workspace and they are utilizing docs and sheets and emails and calendars. Just kind of help them understand what the heck this thing can do. Right, because we've gone from like, done by you, right? And that's where you're mostly just prompting these things. Then we were in the. Done with you and we've opened up with the, hey, you've got gems and you've got cloud projects and you've got custom GPTs. But now we're like getting into the sexier stuff, which is the, the connections of all the tools together. Like, can you just explain like, what it is you, you like about Opal and just talk about it for a second? Because you and I both believe that Google is kind of like the sleeping giant, you know, in this world. And I have a sense that they're doing some pretty powerful things. What can you tell us about Opal before we go on to the other ones?

Real quick, Opal, like I said, it is a workflow designer. In fact, we can bring up, since this is a video, we can bring up my screen here, let me go ahead and hit share and go ahead and bring that up so you can see what Opal looks like. Opal is a. These are one of these no code workflow designers. So you can, you can go and say down here, I want to build an app that will search for news articles about AI and construct a short video explaining them. And if I just put that little prompt box down there, it will start to figure out, well, what do I have access to? What is the user trying to do? What are the outputs that the user wants? And we'll construct a workflow that from its options. So the options it has are things like obviously, user input and Google Search.

Right, obviously Google Search.

What it did in just that few seconds is it said, okay, we need an input section. So the user has to tell me what the thing is. I have to do some research with Google Deep Research. So it's using Gemini a deep research. It says I need to summarize and it wrote the prompts for us. And then I have to generate the explanatory video with Google VO to create the output. And so in just one prompt, it created a very straightforward workflow.

There's a moment happening in companies right now. Leadership is discussing 2026 priorities. AI implementation is at the top of the list. And someone asks who on our team can actually lead this? That's when it becomes clear some marketers are prepared. Most are not. The ones who invest in real AI training. The hands on implementation focused kind suddenly become invaluable. Here's what happens when you attend AI Business World 2026. You walk in uncertain, you walk out confident. You arrive overwhelmed by AI tools. You leave knowing exactly which ones to use and how. You come feeling irreplaceable, you return indispensable. You multiply your output. You end guesswork on which AI tools actually work. You create content faster while keeping your authentic voice. And you automate hours of daily work. And you teach others inside the company how to do it. When your company asks who can lead our AI implementation, you'll raise your hand with complete confidence. Head to AIbusinessWorld live and grab your tickets today so you can become your company's AI expert. Okay, that's really cool. We probably won't do a lot more screen sharing, but for everybody that didn't see it, this was literally just all created instantly once he put in a smart prompt. And my guess is you could actually start within Google, Gemini or any other and have it build you an even more sophisticated one to start with. But what's really cool is like this just built the whole thing and then if you wanted to, like, what's the output of this? Is it like a link that you can reuse over and over again? I mean, like, just help people understand how this functionally can now become a reusable thing for you.

So this thing's output will be a VO3 video. So this is a video file and I could, if I specified, I said and upload it to my YouTube channel.

Oh, wow.

Yeah.

Holy cow.

Right? Anything in the Google ecosystem that it can talk to, it can build connectors to. And that's going back to where we started this show. Remember, agentic AI is all about connecting to the other stuff that isn't AI and Opal as a simple Example is one of those things that says, hey, we know you're in the Google ecosystem. If you're willing to stay in that system, here's all the things you can build within it.

And it's free, right? And where do they find it? They just Google search, Opal or something like that.

Opal. Google is the address of it now.

Okay, cool. So the limitations are right now that it only works within the Google ecosystem, but for actually a lot of internal stuff, it could be pretty powerful, especially if you're using sheets and docs and all that kind of fun stuff. Okay, Claude skills is something you wanted to talk about. You want to talk about that a little bit?

So this is a relatively new feature inside of Anthropic's Claude. Claude skills are similar to projects and similar to GPTs, except you call them as a function wherever and whenever you're in a conversation. So we did a whole livestream last week about this recently about this on our channel. But functionally, you use Claude's skill builder and you say, this is a skill I want you to have example, you know, building a skill that says, write like Mike Stelzer.

Okay, all right, There you go.

You would build that skill. You would upload documents. CLAUDE will assemble it based on its own best practices and give you a little file, a zip file. If you've ever installed a WordPress plugin on your website, you know what a cloud skill is because it's basically, it's a plugin. You load that to your Claude or give it to somebody else. So, like, you could, you know, Mike, you could hand your.

I could hand it off to my marketing team, right?

Yeah, exactly. And they can install it in theirs. And then in any conversation at any point, you could use the Claude skill. You could say, you know, right? Like, Mike Stelzer is the little function, the hashtag function, call, and it will load the skill and it will do the thing.

Okay, that's really cool. And that's kind of like a little brain version of just this one little thing, right? So it's. Is it agentic or is it kind of like just something that could be strung together in an agentic kind of way? What's your thoughts on that?

This is still done with you. So you've built the thing, but you are still piloting it.

You were about to talk about OpenAI before we stopped and slowed down and talked about Opal. I'd like you to talk about that. I also want you to talk about N8N.

OpenAI has their own Opal like systems called Agent Builder. It's terrible. It connects to almost nothing and it's made for developers to build agents that can talk to things like MCP servers, model context protocols, which are APIs for AI. And it what they're skilled at though is connecting to lots and lots of different pieces of the OpenAI ecosystem. So it can use the text to speech model, it can use the speech text model, it can do, you know, the, the various GPT models. It's still in its infancy. It just came out about a month ago as the time of this recording. And it, it has a long way to go. But it just for folks who are in that ecosystem, it does exist. Another version of this, OH which is poorly named is Microsoft Copilot Studio Flow, which is a paid add on to the paid version of copilot inside Microsoft 365 and similar to Opal and stuff like that, it has its own drag and drop agent builder. And to no surprise, it connects to pretty much everything in the Microsoft ecosystem. So if you use Azure, if you use their classic stuff, if you use Windows and Office365, Copilot Studio Flow lets you build those same types of drag and drop workflows.

Okay, talk about N8N the cloud version because you have some exciting things to share about that. I know you're very excited about N8N.

Yes. So N8N is probably the king of the hill in this workflow automation space. It's been around for a while. It actually was built in the age before AI, before generative AI, because it's really good at just connecting things together.

So it's like make. It's like zapier almost, right?

Exactly. Zapier, make N8 and are all kind of the same class and family of things. What makes N8 and Interesting is two things. One, it's an open source project, right? So there is the ability for you to run it on your own hardware if you want, which for companies that are very data privacy sensitive is a fantastic option because it's one less third party you have to be worried about handing your data to.

And you probably have a version on one of your computers in your office there.

I'm imagining I run it right on my laptop. I just keep it running all the time because it doesn't consume much in the way of resource. And then the second thing, and this is relatively new, is like Opal, the latest version of NAN now has a prompt window. One of the big obstacles people have faced with NAN is the fact that it is not particularly User friendly. It's very powerful, but it is slightly more technical. The new version does, allows you to say, I want a workflow that does this, this, this and this. And you can prompt it over and over again. It can pull out nodes that it needs and connect them all together. And you don't have to do all the plumbing work to tie the pieces together. That's only in the cloud version. I believe it's only in the paid cloud version because all things generative AI consume resources. But if you've heard about N8 and if you've seen it, we showed it, you know, I just showed it in my talk at Social media Marketing World 2025. Other people have shown it in their talks. Once the average user logs into it and gets to, they're like crap. Because it's complicated, very complex.

Yeah, I love what I'm hearing you say is the cloud version is similar to what you talked about with Opal. You can give it a. There's a feature in it where you can. And my understanding is make.com also has a similar functionality now where you can prompt it and it will kind of build, for lack of better words, the basic template for you and allow you to kind of have a head start. You were going to mention how you're actually using this on your podcast. Share a little bit about how Katie is using this with you to help with clips and stuff.

So one of the things that we do with our podcast is we use combined with Google's Gemini within N8N to take a transcript of the show and identify the smartest things that my CEO said in that show. Because one of the things that is. Is really important to us is that we have more representation of women's voices in the AI space, particularly in leadership roles. So Katie and I record our podcast weekly. We get the diarized transcript from. We use Fireflies for. For that. You know, any service can diarized is fine. We take that transcript.

What do you mean by diarized?

Diarized means you're breaking it out by who's speaking.

Oh, okay. Keep going. I'm listening. Yeah, thank you. I understand.

Now that goes into a folder on my desktop and then there's an NAN workflow that says open the folder, read the PDF, feed the PDF to Gemini, have Gemini clean up the transcript and produce that for use on our blog. Then it takes the timestamps from the transcript says, okay, find the 30 to 60 second section that Katie says. That's the most insightful thing that our audience would care about. We have a copy of our ideal customer profile that it loads, it looks through, it produces that, and then it produces a little command for the command line. On a Mac, there's a command line tool, terminal tool called FFmpeg, which is a free video editor, no interface. Nan writes the command to say, oh, here's the original video. Just cut out this section, here's the start time, here's the end time. Clip this out. And so it produces for me now not just the summary, not just the transcript, not just the part that Katie says, but also it does the work for me and just clips the video so that I don't have to do the video editing. I can take that and load it to my social media schedule. We use agorapulse, and it saves me so much work, all I have to do is drop the transcript in and hit go.

Okay, I'm gonna geek out a little bit here, but is there a reason why. And I'm getting a little techy here. Of course, but you could have Google Flash watch the video and actually find not just the best section of the video, but where her energy is higher, if you really wanted to, and get you a timestamp on it, could you not? I mean, if you really wanted to, instead of just dealing with timestamps and transcripts, it's just probably a little more work to do. But I would imagine you could do that because it can watch and, and understand actually what's being said in the video, right?

It can. However, it is more costly than doing it linearly. The other thing I've done is there is. There's a Python version of this. Now this is super tacky. There's a Python library that can look at audio waveforms and say, when was this person's speech level the most energized by volume and cadence? And you can then amend the timestamps with an intensity level and use that as well. At the end of the day, all the models are going to do the exact same thing, which is convert this all into math. So however you get to the math. I personally try to, and this is a, this is a bias of mine. I try to do everything at things as much as possible that are local and that they are open source and ideally free. Because, you know, we're not a huge company. We don't have Accentures budget. I think we have like Accenture's cream cheese budget as our entire company.

But for people listening who want to do this in the cloud, I would imagine they could just tap into Google because Google has incredible vision models and they could ask it to watch the video and actually identify the best clips in the video, not based on a transcript, but based on actually a couple of variables. Like, for example, how animated was the person, you know what I mean? And how was our energy level? Because you could say something super great, but sound really flat. I'm just getting creative with you, but I love this.

You could do that in Gemini Flash.

Yeah, for sure. Because it's fast. So, okay, we've talked about done by you and we've talked about done with you, and we've gone down a fascinating rabbit hole where we've talked about Opal and Claude skills and N8N. Now we're going to get to, like, what some of us are going to consider the future, which is the done for you. Right. First of all, is this done for you stuff actually here yet? And if so, like, talk about it. Let's explore this a little bit.

Yeah. So done for you is when the system does the thing without you.

Yeah.

Right. And so, yes, it is already here. You know, as we started the beginning of the show, Deep Research agents, they're already here. They're. They're built in your chat GPT, your Gemini, you can say, hey, just go do this thing. Right? And it will go and do the thing. N8N has the agentic flows where you can set it up and, you know, put it on a scheduler and it just does the thing and you never look at it again. Right. The definition of done for you is literally that you don't do anything other than pick up the results. Just like, you order takeout, you order delivery, and just, you know, someone shows up at your house with food, you're like, oh, that was cool. You had to do nothing. Some of the systems that are out there that you're going to build with, because at this point, you are building systems.

Real quick, before we get into this, I just want to clarify something done for you doesn't have to mean done without you completely. I think we should talk about this a little bit because there still should be a potential human in the loop on some of this stuff. Do you agree or disagree? What's your thoughts on this?

For true agentic. No, because you've already solved for that. Done for you is the evolution of all the practices we've talked about. So you absolutely should not go straight to building an agent. Your first step in this process is you figure out if even works. Like so at the done bayou stage in Gemini or Claude, get the prompts to work. If you know the test, get the prompts to work, battle, test them. How are the different ways they could go wrong? Then you graduate that to a GPT or a GEM or an N8N or an Opal, and you start connecting to systems and you still go back and test. This is software development, you know, does it still work? And then when you get to the agentic portion, where it's doing it without you, you know, because you've built it from the foundation up, that everything you're putting into the system already works, this.

Is kind of the equivalent to having someone who works for you that you really trust and you just let them do their job and you kind of stay out of it. Is that really what I'm hearing you say?

Exactly right. This is full delegation to say, I don't have to look over your shoulder. I know you know what you're doing.

Okay, so you're about to unravel some of the tech options we've got. So go for it.

There's so many. All of the big tech companies have foundation services that allow you to build true agents. Microsoft Azure, aws, Google Vertex are all examples of true agent ecosystems. CLAUDE code and so on and so forth. Of all of these, CLAUDE code is probably the most accessible to people because it requires an anthropic account and installing a piece of software to build the agents. But then once it's built, you just kind of have it construct the other pieces that it needs. I'll give you a real simple, real life example of how this works. I built four agents for writing a book, a writing style analyzer, a chapter editor, a voice validator, and a fact validator. That one has web research skills.

What did you use to build these, by the way?

This is all claude. All right. Within side Claude cloud code. Yep.

Okay.

And my friend Becca Boltzma, who speaks many same conferences we speak at, was saying, I would love to write a book on her area of study, which is AI ethics, but I don't have time. And I said, yes, but you post on LinkedIn like three times a day. I said, so why don't we do this? Let's hit Export on your LinkedIn archive. So if you go into LinkedIn's privacy settings, you can actually export your entire LinkedIn history. Oh, we grabbed. Wait, wait, wait.

Tell everybody how to do that. So wait, how do you get to there? Because that's fascinating.

So you go to your LinkedIn, go to your settings, go to your privacy and data, and there's a section called Download archive.

Okay.

And this is True, by the way, of every social network. They are required by law to give this to you. Google calls it takeout.

Okay, so what did you see? You exported all of her writings and then what happened?

All of her posts for the last two years, which comes out as just tabular data. And I put that in a folder inside this Claude code environment. And I said, you're going to use these four agents to read through her content. Build an outline for a book, build a chapter by chapter outline for each chapter, and then write each chapter. Then engage the writing style agent to make sure it's compliant. Engage the fact checking agent to make sure it's, you know, there's nothing factually untrue in here. And engage the sort of the coherence agent to make sure it all makes sense and it flows. And we hit go and walked away. And 92 minutes later, Claude had spit out a book, chapter by chapter, ready for her review. And she obviously is, you know, has to then do the human review part to look at it and read through it. But it did a phenomenal job. Like, I read the first draft, like this would be good enough to go right now. Obviously tune it, you know, as you want. But that's an agent, right? I didn't have to keep reprompting it in the middle of the process. Oh, no, do this, do this. Nope, Agents go to work. Tell me when the book is done.

Wow, that's really cool. Give us some tips on how in the world to actually build an agent inside of cloud code. Do you have to be super techie? I know cloud code, by its very nature sounds technical because it's got the word code in it. What's your thoughts on this?

There is a hurdle in the installation process in that you have to know how to install it and there are directions on the website. They're phenomenally unhelpful.

Okay. It'll run on a computer though, is effectively what you're saying it's not?

Oh, absolutely. It runs. It'll run on pretty much any computer that has a command line or terminal because it's using anthropic cloud to do all the processing. So it's not going to tax your computer at all. So you have to know what the terminal is and where to find it. You then have to be able to make sure, however your system works, that you install Node js, which is a programming environment, and then Node does the installation of Claude code. So those are sort of the three steps to get through to get this installed. However, once that's Done, then it will use your anthropic. Subscription is paid only, there's no free version, and at that point you are ready to start working in the environment.

Is there anything you can do in the cloud to prepare you for this? Do they have a Claude code cloud version so you don't have to?

They do.

Oh, talk to us about that.

It just came out this week.

Oh, what does that make possible? Talk to us a little bit about that.

So it's almost identical to the command line version, except that it works on the web. In the cloud, however, there's a separate blocking obstacle for folks signing up with. It is easy, you just sign up right inside the environment, but then you have to provide it with a GitHub repo to start working in. Now, a GitHub repo, for those who are not familiar, is a cloud storage code base. You sign up for a free account on GitHub.com you set up your first repository and then you can give that to cloud. And remember, just because it's in a coding environment doesn't mean it has to be code. Right. You can set up a GitHub repo for anything. So you could have. In fact, I'm doing a test right now. This is kind of funny. I'm doing a test right now of cloud code on the web to set up a repo to write a trashy romance novel from beginning to end with a multiple sub agents, just to see if we can make it do this thing. But it has to have a repo, you know, so the repo's literally named, you know, trashy romance novel. It will have agents for character development, world building, etc, that's the. And it will engage all those agents to try and build a 50,000 word trashy romance novel.

You referred to, you have multiple agents inside of Claude code. So I'm assuming. And whether it's in the cloud or on your local computer, these agents are engineered by their very nature to communicate with each other, presumably.

Right, exactly right. And they know to call each other. Okay, so if you were saying, hey, you know, you do some writing and it knows that because you've declared it, in the instance there's a writing style analyzer, it will automatically invoke it.

Folks that don't use Claude, it's exceptionally good for writing. I feel like it is the best platform that I, as someone who's a professional writer, I've written a lot. I feel like Claude is just ridiculously good. You mentioned outside of Claude, Google and Amazon and Microsoft all have their own Things, I would imagine these are even incrementally more complex. Is that fair to state that?

Or they're substantially more complex. So Google Vertex is, for example, Google's AI agent ecosystem. It is incredibly powerful. Like you can build customer service agents that can interact with customers and things. You can build crazy automations and it is about as user friendly as a porcupine.

Are these restricted to only working on Google ecosystems or do they work with anything?

No. Once you go into Vertex, you can access hundreds of different models and services. You can actually access anthropics models right from inside Google. It's a true developer platform. So it contains a ton of power and also requires a Google Cloud account which is separate from Google Workspace. And it is intended for developers. The documentation is written for developers, is not written for business users.

Where do you see this all going? Do you feel like there's going to be somebody who's going to develop, for lack of better words, middleware that's going to interface with cloud code and Google Vertex and Microsoft Azure and Amazon aws? Do you think this is coming? Do you think it's already here? I just love to kind of know where this is going because I think people are actually getting excited because like they can start doing things now that they never thought they could do before potentially here, right?

We're already there in some respects. So for example, there's a service called Open Router that connects to like 400 different models and you set up your preferences and things and then it will automatically decide what and where to send your AI requests based on your priorities. If you say like, lowest cost is my priority, it will route things automatically. We see OpenAI doing that even in their own system. Right. GPT5 is five different models, but it's disguised as one and it tries to do its own routing. Where the value is going to be for a lot of people is as you see systems like N8N and Opal mature to be even easier, ever easier to use. Right. So the ability to talk to it and say, I want you to build me this. Oh, I didn't mean that. I meant this. That to me is where you're going to see, you know, substantial value. One of the big risks right now in AI to business is the fact that software is even more of a commodity. Right. If you know what you're doing with a tool like Claude code, as some companies have, you can say, hey, you know what, write me a new CRM, I don't like the one I have. Right. And you can sit down and specify your requirements stuff. And yeah, it will take a lot of effort to do that, but it is possible to do that, whereas previously it was not. And so the defensive areas that you have as a business to lock in value for your customers are going to be around your people and your processes and how you just and your data, the data you have that other AI companies and models don't have. That's like literally the only emote that you will have.

Christopher Penn we have just scratched the surface of that brain of yours. If people want to connect with you and potentially do business with you, where do you want to send them? And if they also want to connect with you on the socials, what's your preferred platform?

So Trust Insights AI is where you can find everything business related. You can also find my personal blog, Christopher S. Penn.com@cspenn.com. on most social platforms, reluctantly, I spend the most time on LinkedIn because that's where my audience is and I probably create the most content on YouTube and I argue with strangers on the Internet the most on threads.

Chris, thanks again for coming on today.

Thank you for having me.

Hey, if you missed anything, we took all the notes for you over@socialmediaexaminer.com A81. Be sure to follow this show on your favorite podcasting app. And if you've been a listener for a while, while we would love a review on whatever listening platform you're on. And do share this with your friends. You can tag me on Facebook, LinkedIn and or X. And do check out our other shows, the Social Media Marketing Podcast hosted by yours truly and the Social Media Marketing talk show. This brings us to the end of the AI Explored Podcast. I am your host Michael Stelzner. I will 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.

What if you could get year round AI training? That's exactly what's waiting for you with our AI Business Society. To learn more, visit socialmediaexaminer.com AI.