So What? How to Prompt Agentic AI

Well, hey everyone, happy Thursday. Welcome to So What, the Marketing Analytics and Insights Live Show. I'm joined by both Chris and John through the miracle of space and time. Chris just got off a plane. John has internet, and we're all here.

So it's perfect. Uh this week, we are talking about how to prompt a gentic AI. So everyone's talk about AI, everyone's talking about AI agents, building your own, you know, army of robots, all that stuff is great. Not if you don't know how to talk to them and tell them exactly what it is you need them to do. So this week, and Chris, I know you ran an experiment over the weekend, which I was just looking at, and unsurprisingly, I have questions.

But before we get into how to prompt a gentic AI, do you want to sort of set the stage for what it is we're even looking at? Sure. So and I apologize, I've had like two hours of sleep entirely. So if this comes out as garbled delirium, just let me let us know in the comments. Um it's what John and I are here for.

We're just the comic relief, and we'll just keep the show moving when Chris just kind of keels over. We'll be like, hey, so to catch the news a day. Um if we think of an AI model as an engine, you know, GPT 35.3, the new brand new GLM5 that came out yesterday, right in the middle of my workshop. I'm like, I was like, oh yeah, and GLM 4.7 is the latest model. Uh never mind.

Gotta change that slide. Like literally, it dropped in the middle of the workshop, which is hilarious. Uh Claude Opus 4.6. These are models. Models are the engines of AI.

And the way that we've been using them for the last two or three years is inside a uh a model harness, an interface like ChatGPT uh or Google Gemini or Claude, uh, which it by the way, uh funny enough, uh folks who are native uh French speakers pronounce it a totally different. They kept they call it Claude, and I and during the workshop, I was like, who? Oh, Claude. Yeah. Um they provided these web interfaces with a lot of back end stuff, like you know, safety rules and things, and that's how we come to use AI.

Uh, we've come to use it to you know, we you open up chat GPT, you talk to it, it does stuff, and then you copy paste and you go on about your day. Okay. Agentic AI is all about saying, well, you've got this great engine. What about the rest of the car? Maybe I don't want to be the copy paste monkey.

Maybe uh a lot of people have been talking about how there's no ROI of AI, depending on which study you read. And a good part of that is because, yeah, if you are just in vanilla chat GPT and you're copying and pasting everything in and out of it, of course you're not gonna get any like massive savings because you're you're still the bottleneck. Agentic AI is when you say, okay, let's take the engine, build the rest of the car around it and connect it to things. So let's connect it to our Google workspace so it can read my Gmail, my calendar, let's connect it to Asana, our task management system, let's connect it to HubSpot R CRM, and let's give it a different interface where now it has more autonomy. You could say, pull the last 20 opportunities from HubSpot and uh tell me how well they fit our ideal customer profile.

And as long as you provided the the pieces, the model harness that you're using, like a Claude Code or a Claude Cowork or Google Anti-Gravity, could say, you know, I'll go and do that and I'll come right back and tell you what I found. So that's the difference between the AI that we've been using from 2023 to 2025 and the AI today of 2026. Okay. Um so when we first started talking about generative AI, we were giving uh examples of you know best practices for prompting. We have things like the repel framework, which you can get at TrustInsights.ai/slash repel.

We have things like the race framework, role action, context, execution, the basics. And obviously, as the technology becomes more sophisticated, the way we need to think about interacting with it becomes more sophisticated. So, you know, the whole the point of today's episode is how to prompt agentic AI. So, do the frameworks that we educate people on, such as race and repel and pair and you know, pick an acronym, are those still applicable or have things changed? Like what do we need to know today?

So Andre Carpathi, the co-founder of OpenAI, said in 2023 the hottest programming language was English, right? In 2026, the hottest programming language is project management skills. You are giving it a project, you know, a PRD, a requirements document, something along those lines that it can autonomously pick up, run with, and go and do something with. So for example, uh I'm going to use this from the from this week's uh uh workshop that I just got back from doing. Uh this is the world-famous gopherhole museum in Torrington, Alberta, um, which is a museum of 45 dioramas of gophers in common Alberta life.

Uh uh it's it's this cute little place. It's about an hour north of Calgary, and I think it's literally like a one-person show. Uh, and they're only open during the summer. I was gonna say, I'm not gonna lie, I would totally be down to go visit that. That's a good one.

Oh, totally. Totally. And I said, This is part of the talk was to say, you know, static content, like a standard web page, is very vulnerable to AI overviews and to AI replacement and search because you don't need to go to their website anymore, which means that the website owner loses the ability to try and convert a visitor because you don't get the visitor anymore. So I said, what would it look like if we had an agentic system like Claude Code take a shot at making something interactive that AI can't summarize that AI can't take away from us because it would be something kind of fun instead of just a static page. Not that this there's anything wrong with the static page.

And so I had to sit down and think how would I how would I do that in an agentic way? And the single best framework for doing that today, uh, for prompting agentic AI, huge surprise is the 5P framework. And here's why. Because the 5P framework contains all the ingredients that an AI agent would need to do the task. So, what is the purpose?

What do you want to do? We've got a static website that AI can take traffic away. We need something interactive and dynamic that can't be summarized. Uh, and the purpose of all that is to keep the traffic we get to our website, and also because this is a tourist attraction that has a season, they need to somehow raise revenue in the off season when they're not open because you know the bills don't stop when the season stops. So that's the purpose.

The people involved would be people like me who would be using agentic AI and directing, and by the way, this is a really critical part. Today's agentic tools, uh, this just came out a week ago, I think, in Claude Code. You can now create virtual teams of people inside there. So you would say, I need a UI designer, a UX designer, a gamification expert, I need uh an SEO expert, I need a project manager, and you would specify in your project plan based on the five piece. Here's the virtual people who belong on this project.

The process is how you want to approach this to say, let's first gather the team, do some brainstorming. What are some ways that we could come up with that would be interactive that would give people a reason to come to this website, maybe even give a reason to donate to the website? That would be kind of fun to drive revenue in the offseason and preserve their traffic. The platform is something that you would likely have the agents partially designed. So you have some conversation, just like you would in requirements gathering.

Say, well, it's got to be something lightweight. It can't be this company clearly does not have the resources for a massive server farm. So it's got to be something lightweight client side, super easy, super non technical. So uh in for IT IT agent, what do you think? You know, what give us other stuff some choices?

And then the performances can you actually build the thing? And and and when you're done building, does it work? Right? Does it meet those requirements? So I sat down, and this was I was doing this on the plane ride to uh Alberta on Sunday, and I wrote it out at like the five Ps, and I gave it to Claude Code.

We went back and forth and had the conversation for about 15 minutes, and then I said, go off you go. And it spun up the agent teams. And what came up uh at the end of it all, uh, I'm gonna bring uh share my screen here. Uh this it we're gonna go to trust insights.ai slash interactives slash Taya slash gopher. Uh this is hosted on our website, Katie.

Oh boy. You know I love surprises, Chris. Yes. Um, but it consumes no resources because we put it on all the things on the class side. You can choose your diorama that you want, then pick the gopher that you would like to add to the diorama, give it some cute outfits like uh Royal Mounty's uniform here.

Let's make that a little bit bigger, and then uh maybe give it a prop like a broom. I don't know why you'd be doing that to play billiards. Uh but you'll notice at the bottom here is a button that says you can order a print that would tie into the gelato API, which is a print on demand shop. And they allow a company like the the Gopher Hole Museum to essentially be an affiliate. So if someone wanted to make their little diorama, the kids made this cute little diorama and they hit order print, the museum would get like $2 off the $8 print.

And then there's also the way to visit the museum itself, get the gift shop, do the donate and stuff like that. And all of this came together in about 45 minutes. However, it needed the 5P framework as an agentic prompt to pull it off because when the the two bookends are the most important. If you specify the purpose and you specify the performance and tell an agentic system here's what success looks like, it will figure out most of the rest. So I love this, obviously, for obvious reasons, but no, I love it because it really does show the flexibility of the 5P framework.

And I was actually having a conversation with Kelsey, our account manager, the other day. And one of the things she mentioned was, what if I don't know all of the Ps of the 5P framework? And I said, that's okay. Because one of the things that this framework is meant to do is to highlight where you don't have enough information so that you can go get it before you start building things and you know, racking up bills and pressing buttons and getting your CEO chasing you going, why do I suddenly have a $5,000 bill for something that you don't even have an output for? And so I really appreciate that you feel like the 5P framework is the right framework here because to your point, like you you knew a lot of information going in.

You knew what you were after, but you probably iterated with Claude Code and said, I have a lot of information. I maybe don't have all of the information. These are the questions I want to be able to answer. Maybe I don't know who all the people are. This is what I'm thinking of doing.

Who are the additional people in my neighborhood? This is what I think I want to do. What is the process? Like you can use this as an anchor to make sure you have all of that information before you start building something. So I haven't defined my performance.

Can you help me figure out what my measure of success is? I know what I want to build. That's my purpose. Like, is that clear? Do I need to refine it?

I'm I'm very jazzed about this. And I don't know that that sort of dates me, but I am like jazzed is the word that I think is the most appropriate jazzed. My you know, shock and delight. Um even if I have a lot of information, the 5P framework is still important because I don't ever have all of the information. And what I find is happening more and more as I'm doing more advanced projects is the number of blind spots I have is increasing.

So I'll give you a very simple example. Um, I wanted to make a new kind of video game, a very simple game, kind of like Wordle. I had this idea, and I was like, Oh, yeah, that'd be cool to create something, and I have this idea that I'm working on. And I was like, well, you know, suggests uh here's what I think we might want to do. What don't I know?

What plat what technologies in the platform in particular are available? And it said, Oh, yeah, there's one technology called Pocket Base. Uh, Pocket Base takes a SQLite database and wraps it inside of uh uh its own harness to to have an API and all this other really cool stuff. I'm like, where the hell has this been for like the last year that I've been making things with SQLite databases? I looked at this and go, this is fantastic.

Um in another case for a uh some client uh uh um project we were working on. Uh I did a V1 of the project, and at a certain point I was like, this is just not going well. So I said, let's do a V2. And as part of doing the V2, when we rebooted it, I said, as part of the requirements document assume that everything I wrote in the requirements document for the V1 is wrong when it comes to platform. What are the best choices now?

What's what's in market now? And it's it, and one of the things to say was, dude, you you picked the wrong database. This is not the database. You just use this one instead because it also supports vector search. So if you want to eventually be able to use generative AI to understand thematically what's in the database, Postgres SQL with PG Vector will do that.

Whereas the your old choice, you will never support that, and and it will also continue to be a costly nightmare. And so even if I think I have the answers, part of the 5P framework for me, when it would when we're prompting agentic AI, is to say, what are my blind spots? What don't I know? And each of the five P's, ask me questions. What portion of the purpose did I not think through?

So in the one game example I was uh talking about, I said, Oh, what about this for monetization? And said, do not do that. That is a that is a that will ruin your game. If you want to monetize it, do it this way instead. I'm like, oh.

And so now the purpose of the performance changed based on the conversation I had with uh you know with one of the agents. Like, oh, that's that makes for a much better experience. And and you know, the at one point the agent said, here is how you can build in like sponsorships in a way that's not like crappy for people who play the game. So if you want to plan for radical success down the road, you don't have to re-engineer everything, you know, a V2, you can plan up front. One of the keys of agentic prompting is you need to spend about three times as much planning as you do anything else.

Because if you get the plan right and it's solid, you can hand that to the agent and it just goes and does it. And because the plan is specific and detailed and has code snippets in it and has all this and that, it just knocks it out of the park. If your plan sucks, you're gonna get a steaming pile of crap. It's almost like someone has been saying this forever, even before uh generative AI, you know, up-ended all of our lives, it's almost like someone has been preaching to an empty room that if you just got your planning straight, you could execute things so much faster. So I am very much relishing the validation of things that I've been saying for years, and I'm taking full credit for being the one who has been yelling to no one.

But hopefully now people are listening because here's the thing. When we were talking about it in terms of like software development or marketing, planning execution, you know, everyone sort of felt like, yeah, I know enough about it that I can just kind of wing it. Now with generative AI and now agentic AI, the stakes were a lot higher, especially where there's a very tangible cost associated with using these tools. Like there's a cost associated with using, you know, marketing operations tools, CRMs, that kind of stuff, but not in the same way that they are with agentic AI tools. And so, you know, and we've talked about how those costs work on previous live streams, which you can get at our playlist at trustinsights.ai slash YouTube, go to the SOAP playlist, all the good stuff is there.

Now you have to be really careful, especially when this whole what is the ROI of AI conversation comes into play. You can't be spending a crap ton of money and getting no results. And that's what happens when you don't do your planning. And I am just, you know, I am loving this. This makes me so happy because I'm like, yeah, I know.

I know if you have a good plan, you get good results, duh. But in all seriousness, I'm really grateful that the 5P framework is the foundation. It doesn't matter what the technology is, it's always going to be there for you to fall back on. Right. And what I would suggest is that within your agent system, no matter what system you you're doing it in, you actually build it into the system.

So for example, in my version of Claude Cowork and Claude Code, um, I actually created a plugin uh for it, the Trust Inside 5P plugin. And inside is Co-CEO Katie to help with it, and then two skills the 5P framework skill and then the fact check skill. And both of those are things that co-CEO Katie uses a lot. So when I say, let's do this thing, co-CEO K is will you like, okay, I'm gonna give this a read through. I'm we're gonna, and then we're gonna force you to think through all parts of the five Ps.

And so you can have that discussion, or if you really want to go kind of crazy, um, you could create an agentic version of yourself to talk to the agent version of Katie, and the two of them can debate and you can come back later and see how things turned out. Um, I do this. Well, it I mean, that doesn't really work for me because I am co-CEO Katie. So like then it would sort of be this weird, like just as a side note, when I talk to the co CEO, I'm talking to myself, and I already find that very odd because I'm like, God damn it, I would say that. That is exactly how I feel about this thing.

So I'm just sort of I got stuck in this like weird visual of like trying to picture like Katie talking to co-CEO Katie and them just like coming to a standstill because they're both stubborn of like, no, I'm saying that, no, I'm saying that. No, I'm saying that. It's like, okay, can we move on? No, because that's what I'm saying. Sorry.

Anyway, that was an aside. Um, so let's see how you would put this to use. Uh, I'm gonna specify a folder and we're gonna choose a folder, and just for today, we'll go with uh let's see, where's my data diaries? So I'm gonna use my data diaries project, uh, which is what I use to edit the data diaries every week in the Trust Incise Newsletter. And we're gonna say, uh, we're gonna choose the five-piece uh skill here.

I can see we have five P framework, and we're gonna drop in my recipe, which is the general prompt I use to edit this thing, and then we want to now do it agentically. So we're gonna say using the co-CEO and the 5P framework. Help help me reformat and rewrite this agentic prompt into the full 5P framework. And to give you a sense of what's in there, it basically says this is an editing task using the writer agent, the editor agent, and the writing style analysis skills, review the rough transcript in the input folder and transform it into a polished newsletter piece in the output folder. You'll uh clean up speech dissonances, bad grammar, reading the transcript to understand the logical flow, then you use the voice of the customer agent with the ICP provided um to ensure that the it aligns with the voice of the customer, what the customer would find helpful and so on and so forth.

Dispatch parallel agents and and stuff. So that's uh a very simplistic prompt. We can improve it, and we should improve it, and we're gonna improve it with the 5P framework. So let's go ahead and make this just a bit bigger. Thank you.

You can see I'm like, what whenever I start. If you see my face down here, it means I can't read the screen. Me and John are like, uh, we're gonna clean our glasses. It says, what I see in the current recipe, it's a rough conversational prompt that makes us goals, tools, processes, and quality standards into a single run-on paragraph, which is completely right. Uh, the intent is solid, transform a rough transcript into polished uh newsletter piece, but the structure makes it hard for AI orchestrated parse priority sequence dependencies and know what done looks like.

Let's now apply the 5P framework with co-CEO rigor to do this. You're working hard, or your virtual version of use is working very hard. Well, you know, and it's I actually I I appreciate that it's taking a minute because as simplistic as the 5P framework is, you really need to think through what goes into each of the pieces, and it's an it really is an opportunity for you to think through do I have enough information. I like to think of it as like the full 360 of business requirements, uh, because like you need all of the pieces in order to successfully move forward. You can move forward without any of the pieces, but to your point earlier, you're not gonna get the best result.

And so, based on the prompt that you gave it, I'm assuming that the co CEO is gonna point out that you have some gaps. Oh, it all it I definitely point out a bunch of the different gaps, and what it's going ahead now in doing is it's just going to go and fix it. Uh it's to say, you know what? Because the original prompt, the intent is pretty clear because it's such a narrow task, like take this transcript and make it better, um, and and then validate it against the voice of the customer. That's not something that you know, this is not a big stri uh strategic.

This is not a complex piece of code. This is essentially a a just an agentic prompt to get a tool to go off and do something uh and do it intelligently. Um for larger stuff, yeah, you'll go through rounds, particularly if you're using clawed code, uh, and you have the uh you have the superpowers plugin installed, which we talked about a couple weeks ago on the live stream. Uh that will start to invoke with the brainstorming tools. It will start asking you questions and say, like, what about this?

What about this? All right. What what the what the original was a single paragraph, what the five E version delivers purpose, people, process, platform performance. Let's take a look and see what it came up with. So the data diaries newsletter production recipe purpose.

The question we're answering how to consistently transform a rough voice transcript into a polished educational newsletter piece that meets the trust insights brand standard and resonates with our ICP. The problem we're solving, raw transcripts contains speech disposities, unstructured logic, inconsistent tone, and unverified claims left unprocessed. They do not meet the quality bar our audience expects. What success looks like finished data diaries column that educates our customers on the topic at hand, adheres to our writing style, aligns with our ICP, and contains no unverified claims. The internal stakeholder, here are the AI agents that it is now documented, and then it has the external one.

And this contains the character cards that I made from our ICPs. Um, so these are the different these these are yet yet more named personalities that uh I use every week, Katie. Yeah, you know, it's funny that you say that because in you know, offline you had given me uh a piece of research you had done, and my first question was like, who are these people? I don't this report is prepared by this person that I've never heard before. Who is it?

I'm not paying them. They're not on the payroll. Get out of here. Yeah, they're all synthetic, they're all eight, they're all uh characters in uh the character the trading card game we have built around this. All right, process, intake and comprehension, read the transcripts, read the style guide, read the ICPs, uh and things, ask the content under clarifying questions, do not assume ask, clean up the speech disfluencies, map the logical structure, the what so what now what framework, which we uh you which we love to do.

Uh the writing agent uh goes off and does those things, then the editor agent goes off and does its things. The voice of the customer comes and offers their perspective, the fact check skill runs to validate uh claims made in the piece, and then you get the final output and the parallel execution guidelines. Here's how it has the different agents have to coordinate together. This is a critical part of a gentic prompting. You have to specify the order of operations, otherwise, agentic tools will either do one of two things.

One, they'll run sequentially, which means you save no time because you're waiting for one agent to the next to the next, right? Yes, it will do autonomously, but it's slow, it's inefficient, or it'll launch all the agents at once, you'll have race conditions because they'll be trying, you know, each agent will say, hi, I did my part, and you're like, No, but this part depends on that part, and they both did it at the same time. So they they basically kind of hallucinate each other's parts. So part of good agentic prompting is thinking clearly through what is the order of operations that the different agents need to play. Who does what first?

Some things can be done in parallel. Here it says phases four and five can run in parallel if the editor's output is available. But phase six runs after four and five. So you can't run uh phase six agent uh without the previous ones. Here's the platform project plan with dependencies.

Who knew? Exactly. Crazy. Here's the platform, the tools required, uh, the platform dependencies, and then how we measure success. Did it comply with the writing style?

Is it would all five ICPs find it valuable? Is it factually accurate? Uh is it complete? Does it have structural integrity? Does uh asking questions, agent utilization, revision traceability, the definition of done, when is the age when are the agent teams done, and then this first principle, which is something that is I have embedded in my version of the uh the data diaries.

So this now is ready for prime time. So this goes uh in the data diaries folder instead of me. Um let's open this folder up here. Yeah, so you can see it has gone through, and that one paragraph recipe is now a full-fledged recipe that I can now every week say, here's the draft, go, and we know exactly what's going to happen. And when I come back, it'll be done.

I I will I because it the 5P framework creates such rigid, in a good way, rigid definitions of what success and done is this allows an agent to iteratively go, are we done? Let me check. Nope. Are we done? Let me check.

Nope. Are we done? Yes, we're done. Devil's advocate question. So this, like I love this, and because I, as someone who is very familiar with the data diaries, I love this kind of extra detail to it because sometimes it kind of feels like a little bit of a rambling, what just happened?

And so I love this, but help convince someone who isn't me. Okay, let's convince John. What is the benefit of doing this versus just taking your transcript, which you know, you are an expert in the stuff you're an expert in, and you create the data diaries on those things. Therefore, it's really just a matter of cleaning up the transcript so that it's a comprehensive blog post for lack of better term. Why can't you just say, okay, copy editor, human or virtual, clean up my post so it's more coherent.

What is the benefit to doing all of these extra steps versus just having a copy editor clean it up? So twofold. Um thing is that there's extra steps in here that the humans just typically don't assign. So, like the fact check, go and Google every claim that's made in here, do web search on it and validate that it's actually true. Um your average copy editor is not gonna do that.

And if our goal is things like accuracy and truthfulness, I yeah, we should be fact checking. Even if I am an expert in something or I know a lot about something, I can and certainly am still wrong a fair amount of the time. So I want to make sure that we're that you were doing this well. Um and two, I have biases when I write, when I when I talk and things like that, I have a point of view, and that's not a bad thing. But the Trust Insights newsletter is not my personal publication, it's the company's.

A copy editor's job is to say, wow, dude, you have you have like the world's longest run on paragraphs here. Um so by aligning this with our voice of the customer, we refine it to say, you know, out of this huge ramble, what did Chris say that's actually valuable to our customer? That's fair. Um, we have a really interesting comment that I would love for you to address. Um, you know, and so this part again sort of goes into the what is the benefit.

So um we have this uh viewer who said what you just defined is going to use a lot of tokens, it went from being a $50 project to a $150 project. Like what do you say to that? You got it backwards. Uh, in an agentic framework, in a gentic AI, if you give it a crap prompt like the original one, it's gonna spend a lot of time trying to infer what you meant. And you could see this in the reasoning traces inside Claude Code in particular, you burn a lot of tokens for the model trying to figure out what it's supposed to be doing.

Because Opus 4.6, for example, does a lot of that, does a lot of reasoning and judgment. If you have a good detailed project plan up front that clearly specifies here's what success is, you can do two things. One, Opus is going to have a lot fewer debates with itself about like what did the view, what did the person mean? As opposed to saying, I I completely understand the plan now, I'm just going to execute it. Um, two, if the plan is really good and well thought out and and it contains the full five Ps, you can switch down to a lighter model like Sana and save yourself a ton of money.

Because if you plan with the big model like Opus, um, because you can see you when I was doing this in Claude Cowork, uh, I have Claude Corwork set on Opus 4.6 for any kind of planning, plan big act small, which is from last year's book, almost timeless, available for sale. Uh plan big act small. Use the smartest model for the plan, but then when it comes to the typing part, you can use a lighter weight model like Sonnet, uh Claude Sun or uh Gemini Flash or GPT 5.3 medium to actually do the copy pasting and rewriting as long as the plan is good. So, yes, you could you'll either way you're gonna use a lot of tokens, but you the the crappy, poorly defined prompt is gonna make the model spin a lot more, and you will get less repeatable results. You will get less reliable results because if the if I went with that original prompt and I was not specific about what kind of outcomes I wanted, I might get three very different formatted columns from the same piece of text.

Whereas if I say in the detailed project plan, this is the format I expect. No listicles, no, you know, 28 point bullets, it's prose, it's articles. I want it to be like this. If I do that, I get a repeatable result. And but by the way, that cuts down token usage because the model doesn't have to sit there and go, huh, how should I do this?

What does the user want? Instead goes, I'm gonna follow the plan. And so that is a great segue into uh Brian's question. Hey Brian, uh, do you use Claude to create the agentic project plan and framework and then take that into Claude Code, or do you have the conversation and make the plan directly in Claude Code? So, me personally, I do everything in Claude Code, like I live essentially in Claude Code.

For the average user who is not technical, doesn't want to fire up a terminal command line. I would suggest you should do the urgentic plan in regular Claude and then do execute it in Claude Cowork. And that's how I typically work is so I'm not savvy in Claude Code. I understand enough of it, but I have become best friends with Claude Cowork, um, and sort of making the plan, having here's everything I need to execute. And unsurprisingly, though, my brain just works in the 5P framework, so I don't even need to really think like, is this how I'm gonna structure it?

Um, I guess that's the benefit of being the creator of it. But to your point, Chris, you can create the plan and then give it to agentic, and I think that is the smart way to do it. So you don't need to use agentic to create the plan, create it to in the idea of giving it to, and that can become one of the P's, the platform be like, and this will be run through this platform, so that when you're creating the plan, the plan is keeping in mind, okay, this is for specifically Claude co-worker. This is for specifically Claude Code. John, you have been a nonstop chatterbox this entire uh episode.

So I'm curious to get your thoughts. Yeah, well, it's still, you know, I don't know. There's not a huge dividing line between straight prompts and calling agents, you know. And then I know it's different for other tools. So that's really what I'm still trying to get my head around is you know, at which point are agents being invoked, and really how is that fundamentally different from just straight prompting?

So that's a good question. Here's how it works, at least in the agentic systems like Google Anti-Gravity, uh Claude Code, etc. Um, an agent is essentially another copy of the AI, right? So when I say I want you to spin up an agent team of a UX UI UX designer, an SEO expert, uh, a marketing expert and a project manager, and you kick that off inside like Claude Code, you will see little like little employees just start popping up on the command line. They each get their own context window uh with their own working memory, and they work with each other.

There's they they also get a little mailbox that they can trade messages back and forth to each other, kind of like you know, the middle school class. But they're essentially five more copies of Claude, and each one is tasked with a specific task. They do their tasks, they coordinate with each other, and then they return back to the main window with their results. Think of it. So if you think of like when you're in just Gemini or Chat GPT, that is the main, I would call that, you know, just the AI.

When you start using agents, that becomes the orchestration agent or the orchestration window where you are you are the project manager overseeing the project, and the employee the little employees come in and go and they they they come in and say, Hey, I did the thing. What do you got for me next? And Claude is the orchestra saying, okay, now you you got to go do this. Now you got to go do this. And that and so it's delegating.

You're like the the factory floor owner, Claude is like the factory floor manager, the agents are like little you know employees on the factory floor. Whereas if you're just typing the pro to into to regular AI, you're basically just talking with one employee as opposed to a team. And then how about as far as having this stuff running 24-7? I mean, pretty much the use cases we're talking about, you're sitting there hamming around, you you know, you wait for stuff to give you results that go forward. But do any of the platforms have anything where you would leave these running and have them going, or is that not part of the where we're sure you can actually have um depending on this the the how you want to do it, you can tell Claude, I want you to in you know independently run certain tasks with a scheduler, and you just go off and do them and you know, just let me know from time to time how it's how it's been working.

Um it requires a bit of knowledge about uh shell scripting and scheduling on whatever kind of computer you're on, but you can ask Claude how to do that, and then you can say, Yeah, I want you to. So, for example, if I I have um a folder on my desktop called Asana that I get my daily briefing from. If I wanted to, I could say to Claude every weekday, run a script that builds me my daily briefing. Uh, and I could I could have it ready for me. I personally like to ask for it because I also will typically give instruct extra instructions like you know, uh oh, do this and by the way, this.

And I remember I forgot about this that was on my that I was it was in Slack and I forgot to put it on my list. And so I I give that feedback. That's my personal working style, but there is a way that you could say, yep, just give me my briefing, and every morning, 8 a.m. ding your briefing's ready. Like a really good example, I think, John, for you is so you basically live inside of our HubSpot instance.

That is the software that you are that basically like that's your world. I could see sort of creating some sort of uh script that every morning to Chris's point, here's my daily briefing of here's all the things, here's all the people I need to follow up with, here's who's ghosting me. Like basically, like it can deliver you your to do list with the details and additional information because you can do that in HubSpot. So it's like, why would you do that with a gentic? The addition is things that we've talked about on uh previous episodes, which again you can get at trust insights.ai slash YouTube, go to the so what playlist.

Um things like build me my daily next best action uh report based on who's on the list today, or you know, which of the people, how do I prioritize all of these based on who fits with our ICP the best? And I think it's bringing in that additional information where agentic is more helpful versus just working within those systems like a CRM system that doesn't have that additional context. And so I think for someone like you, if that was a road you want to go down, that would be the use case that you would probably find the most valuable. Yep. Think about what would you hire a small agency to do for you?

That's a use case for a gentic as opposed to a single contractor. Right. So if you were to say, I've got the marketing over coffee website, I want more corporate sponsors. And I, you know, I want corporate sponsors. I want sponsor sponsoring, you know, mid-six figures uh for the show.

You might say, okay, well, how are we gonna do this? Right. I um and that's where you would start putting together that 5P plan. Say our purpose is I want to figure out how to get the marketing over coffee to get six mid-six figure sponsorships. The people involved, obviously, you uh as the podcast owner, obviously corporate sponsors and booking agents and stuff like that, and the audience too would be part of that, and of course, the different agents.

You would have a sales and revenue agent, you would have a project management agent, you would have uh a marketing agent, and then your process would be look at the site, see what's there, go and do some research to figure out what podcasts that do earn, you know, mid-six figures per sponsorship earn, you know, how do they do those things? How do they get those things? Um, build that research, and then build me a strategy, a tactical plan, and then an execution work plan to get to that point. The platform obviously is is going to be like things like the the website is WordPress, the the hosting system is Libson. Um we don't currently integrate with any ad technologies, but like we're saying earlier, go do some research.

What ad technologies should I have in place so that when that that well-heeled sponsor comes in, you can say I got a system ready to ship for you, and the performance is initially give me a plan that will work. And then ultimately the performance is wow, I'm surrounded in Benjamins. How'd that happen? Um Chip offers the opinion, uh, while agentic teams have big benefits, they also introduce greater risk of complicating projects without creating commensurate additional value. I hear you, but I also say this is why we're really hammering in that we want people to constrain their projects to the 5P framework.

If you don't have this plan, don't start building things. So don't start building agents until you know what people you actually need. And with that, you know, obviously generative AI is gonna be like, I'm as helpful as possible. So you need, you know, this person and this person and this person and this person. So it's gonna give you a lot of information.

You can push back and say, what is the bare minimum of people that I need in order to fulfill this? Or you know, get the three scenarios. What is the bare minimum? What is the mid-level? What is sort of like the gold deluxe addition?

And that's one of the ways that you can get ahead of creating additional complexity because you know, and I think this was maybe on the podcast that we talked about, Chris. Like, yes, it's gonna suggest everything, it's gonna say you could do all the things. You, the human with the human intervention, need to be the one saying, yeah, no, I don't actually need a level one and a two and a three uh information architect. I just need one who's sort of mid-level pretty good because I'm not building anything crazy. Like you, the human still have to intervene with these plans that it's uh building.

And I think that's one of the things that people tend to lose sight about when they're hearing all the exciting things that we can do with generative AI and agentic AI is like, guess what? You still got to oversee it. You still have to be the person who at the end of the day says, no, don't do that. Yeah, Katie, when you and I talked about this on the podcast, it was one of those if you were hiring this as humans, would you hire a 100 person team to revamp the marketing over a coffee website? Absolutely not.

Right. Would you hire a small boutique agency with five or six people? Yeah, that's about seems like the right size. Maybe a a a large larger agency, maybe 10, a 10 person agency, because you might want to have some UI UX folks. You might want to have an audio engineer listen to the podcast to see if it's any good.

You might want to have uh a WordPress expert to you know in to see if you've you've got the right tools and stuff. And also if you are good at the orchestration part, which is what we saw in the platform and the five Ps, you don't have to have every agent on deck all the time. In the same way that when you're building a house, not everybody is there the entire HVAC guys are not there while the scaffolding's going up, right? There's no reason for the HVAC guys to be there when they're hammering the frame together. Um so you have to think that through.

And again, if you build, you know, if you use the five Ps and just drop it into a chat, um, any of the tools will help you figure that out as part of the plan. And if you don't know, and if it's suggesting a bunch of synthetic people with roles that you've never heard of, push back and say, what is this person bring to the table? Do I need this? How is this different from this? Or, you know, can I get away with not using that person?

You know, and I think that that's really, you know, Chris, what you've been trying to hammer hone is get that plan squared away first before you're asking agentic AI to build anything. Maybe in next week's show, we should uh maybe do a part two of this and really deploy an agentic swarm on the marketing over coffee website so that we can we can have it build out a plan for John to become you know filthy rich uh as the host of our king over coffee or using uh you know it may take more than a website revamp, but I definitely open to firing it up and see what it will do. So that's but that's what I mean is why don't we build an agentic swarm that has like a business manager and a sales manager and and a researcher and take the entity as a whole. We'll have to gather all the pieces in advance. Um, but I think there could be a there there, especially if we do some deep research ahead of time to say what are the other pieces of what data do we need to have in advance?

You know, pull in the latest research from our friend Tom Webster. It sounds profitable. They have a lot of great industry research about what's going on with podcasts. Pull some of the data from Edison Research and Pew and all these places and say what you know how is it that some shows do much better than others? What's the difference?

And then what can we learn from that? So we'll maybe we'll do that as part two. I like it. All right, I think that's it. That that is it.

That is that is that is it. So uh we are that is using the five piece. Please try it out. Please use it. Uh, you will save yourself so much headache.

I speak from personal experience. You will save yourself so much headache if you just use the five piece and get good at project management. That's gonna do for this episode. We will talk to you all next time. Thanks for watching today.

Be sure to subscribe.ai slash TI podcast and a weekly email newsletter at TrustInsights.ai slash newsletter. Got questions about what you saw in today's episode? Join our free Analytics for Marketers Slack group at TrustInsights.ai slash analytics for marketers. See you next time.