A (3:03)

So I suspect maybe I'm wrong, maybe I'm naive, but that most marketers today or consultants have figured out, yeah, I need to use this AI thing. You use the term AI native. Full Stack consultant. How does that differ from someone who, say, just uses it in their workflows?

B (3:21)

Well, one of the things that has happened in the last few months, the easiest way to put it is the AI has gotten really good. So there was this. We're on a short timeline for the podcast. We don't have time to dive into it deep, but there were studies done through OpenAI, so take that with a grain of salt, but it's called the GDP val. And what they did was they took a bunch of subject matter experts, they got a bucket of tasks across most knowledge work and they said, give us your best blank and there would be a task and so they would do it. And then they would have the AI do the exact same task and they would give it to another subject matter expert and they would say, which one of these is better? And so 2024 Windsor ties by the AI was like at 10%. Now it's a caveat that the AI has all of the context it needs in order to do the job and most people don't get anywhere close to giving the AI all the best. So 10%. Like people would say, yeah, it's not a 10%. Well, it's probably not a 10% because you didn't give it all the context. So anyways, move forward into 2025. Middle of 2025, we're approaching 50% Windsor ties by the AI. So we're getting close. We call that the AI tipping point. As of a couple of months ago, the wins and ties by AI were 70%. And since then, you can feel if you're using AI, this is getting better and better. So what we mean by Full Stack Consultant is if you understand what good looks like across any subject matter expert domain, you can get pretty close to doing an incredible job for your clients in all functional areas of a business. So a marketing agency can do sales enablement, but they can also do some CFO work, they can do some strategy work. And that is what we mean by the full stack Consultant. And the idea is that if you get very good at some new skills which are not obvious to most people, like how to produce a really good robust context library for your client, you and the company can do amazing work. It can be done incredibly quickly. And we're learning more and more every day about what that looks like. So today, this morning, I did about, and this sounds ridiculous when you hear it from the outside, and I understand that it sounds ridiculous, but I did about, in about two hours with 30 minutes of my work and an hour and a half of just waiting around for the AI to finish its work, about 260 hours of design interface, interface work, copywriting and development. I don't do any of those things in the past life. I have no skills in those things. But I know what to ask for, I know what I want and I know what good looks like. And now I can get it. So it's an amazing time that we live in.

A (6:21)

I want to go back to a point you made there because, you know, I have jokingly but seriously said, you know, marketing is everything. And what I really meant by that was in a lot of small businesses I would go into, there were a lot of things I had to fix that weren't under the, you know, the heading of marketing a lot of times, because if marketing was going to work or we were going to grow the business, I had to get involved in this area over here. Sales was a typical one. Customer service is another one that, you know, that you don't always hire a marketing agency for. But I would get into it out of necessity. And I think what you're really pointing to is a great point, this idea of, of if you go into a business, it's not just a matter of offering a suite of like, what do you need? It's more like I can be more effective at doing my job that you hired me for if I can actually easily fix an area over here that I may don't. May not have true expertise in, or I couldn't spend the time because I wasn't being paid, you know, to fix their P and L, you know, for example. So I think that's a great point. Let me back up again because I want, I circled, I wanted the word context that. Let's spend a little time talking about that because I think AI has gotten better, but I also think prompters are getting better and we're realizing, you know, we get better output with context. How do you give a how in your view? This is probably a really long answer. How in Your view. Do you give AI the proper context?

B (7:43)

So when most people talk about contexts and there's a term called context engineering, they're mostly talking about it in the terms of like, a single task that's going on. What we mean by context engineering is how does the AI know everything about your business so that whenever you pull up a task to do that, you actually, the AI can. It's really hard to explain without getting into the weeds, but here's my best shot. So imagine that you have like the world's best employee on every single task that could be done in your business, but they have amnesia. Every single time you give them a new task, they know nothing. So you have to train them. And that sounds like a really painful thing to have to do. The. But if you build the context library and only has to be done once to start, you can train that. You can give that AI like 10 years of training in about 10 seconds. So it forgets all the time, but it learns, like years in seconds. So all you gotta do, like, this is how I boot up my instance of how running AI in my world, as the CEO of our company, I onboard the AI. I say, go look at the. I literally create an onboarding file. I say, go look at the onboarding file and get yourself onboarded. And 10 seconds later, it knows exactly, like from a meta perspective, how we're going to do our work today. Then I'll say, go look at that folder and let's do this task, like, redesign our interface for this page in our system. It knows exactly how I like to work. It knows exactly how I want design options. And you, as a marketer, you'll appreciate this. So you can go in and you're doing, let's say you're doing ad variations. You go and ask for. You don't even have to be that specific. If it knows everything about your business, you just say, give me five ad variations on this one topic, this one offer we're making sends it back. You look at it, you're like, I like that one the most. And I've had the AI give me the rationale for like, this is, scored it and ranked it. Then I could give me five more variations on that one and then five more on that one. And one of the things that's not obvious to people is that the cost of variations is almost zero. So you can ask for an infinite number of variations.

A (10:08)

I do that with subject lines for emails. I mean, same thing. It's like, I kind of like the idea in this one. Iterate on that 10 more times. Yeah, yeah.

B (10:15)

But you can do it for really big things too. So it's not just like a single subject. You can do it for an entire interface, like an entire set of code. So, like, because it works so fast and the cost of its work is so low, it transforms the way you approach the work. And so customizing campaigns down to the individual level, not a problem anymore. Like, I can find your LinkedIn profile, I can scrape it and I can send you like an entire landing page that's speaking directly to you. And it cost me a penny to do. And so there are things that we can do that we're quite literally impossible before that now makes sense.

A (11:02)

So this might be a good time to talk platforms and technologies a little bit. Are you agnostic or have you really dove, you know, gone all in? And maybe it's a, maybe it's so complex that you can't really say it in one sentence, but is like, are you a Gemini person? Are you a ChatGPT person? Or have you really building your own stuff?

B (11:22)

We're definitely not building our own stuff. We have a very particular point of view, which is we're serving companies and companies will eventually choose their LLM of choice and that's what they're going to do all of their work on. So we are, we're not hitching our wagon to any1llm. We, we also have the point of view that for the most part, most AI wrappers go away. So an organization is going to build their own software. That's our long term bet. And so I, we're just using whatever one is most productive. I personally have the max subscription on Claude, OpenAI and Gemini. I'm mostly using Claude right now because Claude Cowork and Claude code just came out. And so if you listen to this, like a month later, maybe I'm on something else by then, but Claude Cowork has been, that tool alone has transformed our operations in two weeks. Like we literally operate day to day differently now because of that tool. So. But whatever one's working the best when the next time we talk is the one we'll be on then.

A (12:29)

Yeah, you know, I contended for a long time that just what you said, it's, it's going to be plumbing. It's not going to be, oh, I use this tool or that tool. It's going to be, no, this already works with what I use. And I really feel like, doesn't that give Microsoft and Google because of their installed user base? I mean, you know, I fire up Gmail and all of a sudden it's like, oh, there's a new tool, you know, I can opt into, you know, I mean, doesn't that give them an advantage? And also I. The other thing, the first version of AI tools is everybody kind of used them by themselves. Well, now all of a sudden we got collaboration built in, which I think was a big missing part. And so it's like working the way people work already.

B (13:07)

Yeah, I think like all other things being equal, Microsoft and Google have the biggest moat around it. However, for the longest time, ChatGPT was the best tool and now Claude is by far the best tool. And so I don't. I would think, I would have thought that if it was true that Microsoft and Google would be like, for sure would win. It would have happened right now because they had. Not only they have the user base, they have all the documents, and that's what the AI needs for context. But as it turns out, the AI does not read and work with the file formats that we all produced over the last 20 years, which is PowerPoints and Word documents and all those kinds of things. So there's going to be a shift around that as well, which I think will loosen the moat that they have because we're not going to be stuck on PowerPoint anymore. Like, I, the. In the last couple of weeks I've been on this kick of and I think this is just true. I will never use Keynote or PowerPoint ever again. And I'm not using like another AI tool. I just build HTML documents and it does exactly what I want. And I. That's my presentation style. We do SOPs in our business. Everybody in our company builds HTML SOPS because you can just speak into a computer. HTML file is open everywhere. And it's also a good language for the LLMs to understand because it's way easier to read than a PowerPoint. There's other. If you pay attention to how software engineers are using AI, you'll have a glimpse of the future. They're mostly using file formats that they're comfortable with and that work well for development like markdown files and things like that. So that's what you'll see them suggest. Like you have to use markdown files for these things. And what our point of view around this, and I think this will just prove to be true, is that we need to be building like artifacts or deliverables, whatever you want to call it for humans and AI, not just humans only, is like PowerPoint, like LLMs. Hate PowerPoint.

A (15:18)

I hate word document can't read them

B (15:19)

at all, there's too much code around them. Right. So then you have the like markdown files and things that most humans look at it like a picture is worth a thousand words. It's a real thing. Like we need to see visuals and workflows and all those things. And so HTML happens to do both of those incredibly well. So whether or not we're right on that, I don't know. But for now it's like transforming the way that we do work because we can now build things that both our humans understand, the LLMs understand. And also there's this magical thing that happens when a non technical person speaks a website into existence and presses a button and it's live. So it's also a really good AI adoption tool because it's cool, right? Like it's cool to like you build something beautiful. If so as a marketer, if you have a design system that most companies would never spend 30, 40, $50,000 on, you can speak one of those into existence. Like you can do it right now. Then everything that gets designed looks great in your company. And now everybody's sharing beautifully designed SOP documents. Like that's a weird thing to think about, but it's like I like doing it, it looks nice, it explains what I, explains my thoughts and my process. And so yeah, I think this year is going to be transformative in how we all do work and I don't think it's going to look the same by the end of the year.

A (16:47)

So, and I know you have an answer for this, but this, I'm guessing listeners are out there going, okay, if I can just speak this stuff into existence, am I going to just start creating stuff without any kind of guardrails and without any human intervention and you know, where's the, you know, the pushback you're getting from people that hate AI? You know, so imagine the people that love AI but don't want to be embarrassed.

B (17:13)

What do you mean by guardrails specifically?

A (17:15)

Oh, just meaning like if I can design all these things, who's going to actually go and make sure that they're being done right, that they look good, that they say what they're supposed to say. Because you know, some of, particularly some of the image know generators today, I mean there's, it's like appalling. Some of the things that show up in, in some of those, the way

B (17:35)

we look at, the way we look at how work is done is by deliverables. So it's, you can look at it as a process, you can look at it as tasks, you can look at as deliverables. But if you look at it as a deliverable, because that's when the thing ends. So, and that's when the human has to look at it. So when the deliverable is done, okay, first of all, you should have the AI do a QC process on itself. You can do that. And it actually does a really good job of QCing its own. So that's the thing that most people don't understand. But then once it comes off the, the press, whatever, you know, whatever metaphor you want to use, like a human looks at it and says, are we sending this out? And if you treat it, and this is a language that most, like most knowledge workers don't like, it's a factory now. And so you don't QC every energy drink can that comes off the line, but you look at some of them, right, and you know that if this one is, this one's off. Well, we gotta look at the ones that just went out the door because maybe they have a defect as well. It becomes more of a factory mindset knowing that if you have a good manufacturing process and that this like, again, like, marketing agencies will hate this. Like, but that is even the word

A (18:44)

factory they're gonna cringe at.

B (18:46)

But like, if you do not turn your marketing agency into a factory in 2026, you will be out of business. Like, let's go have a. We'll do it next year and we'll see whether or not that's true. Like, you'll have to, you'll have to learn good workflows, you need to learn good work instructions, you need to learn good QC process. And so, and once you do, you can start mass producing things that are top notch and knowledge work will be turned into a factory. And then what gets layered on top of that is a new skill set which is not 100% clear what that is yet. But we will invent new things to do that will just add value on top of that.

A (19:32)

Fascinating, Steve, appreciate you stopping by the Duct Tape marketing podcast. Where do you want to invite people to find out more about? I think it's simple consultants, AI and obviously about your book.

B (19:44)

Yeah, well, if you're up for it, I would love to give everybody in your audience free access to our black belt training and we'll create a page specifically for your network. It'll be ROI association AI slash ducktape.

A (20:03)

Awesome. Well, as I said, we'll put that in the show notes as well. So Steve, again appreciate you stopping by. This is awesome. And hopefully we'll run into you maybe one of these days out there on the road.

B (20:13)

Absolutely. Thanks, John. It.