This week’s podcast is about what businesses should know about AI Agents.

You can listen to this podcast here, which has the slides and graphics mentioned. Also available at iTunes and Google Podcasts.

Here is the link to the TechMoat Consulting.

Here is the link to our Tech Tours.

Here is a podcast on AI strategy:

• A Framework for GenAI Strategy – Part 1 (Tech Strategy – Podcast 203)

Here is the mentioned Jensen Huang slide.

On the topic of censorship regime and USAID, here is a tweet from today.

 

———

Related articles:

• BYD Is Going for Global EV Leadership (1 of 2) (Tech Strategy – Daily Article)
• A Breakdown of the Verisign Business Model (2 of 2) (Tech Strategy – Daily Article)
• 3 Factors Will Determine the Future of Verisign Inc. (Tech Strategy – Podcast 191)
• A Strategy Breakdown of Arm Holdings (1 of 3) (Tech Strategy – Daily Article)

From the Concept Library, concepts for this article are:

• AI Agents

From the Company Library, companies for this article are:

• n/a

——-transcription below

Episode 238 – AI Agent.1

Jeffrey Towson: [00:00:00] Welcome, welcome, everybody. My name is Jeff Towson, and this is a tech strategy podcast from Techmoat Consulting and the topic for today, AI agents, which is It’s kind of a vague evolving thing sometimes called digital humans, sometimes called AI agentic frameworks. It’s kind of fuzzy, but things are moving really fast.

So, I thought I would do pretty short podcast on, I don’t know where I think it is and how I’m sort of factoring this into business decision making. I think this is going to be the big thing for 2025. I think this is it. I think this is the next, Oh my God, like game changer. Of which we’ve had, you know, a couple in the last couple of years, this is the big wave and we’ll see, but I’m going to sort of give you my thinking on it right now.

And what I’m watching for, let’s see, no housekeeping [00:01:00] for today.   standard disclaimer, nothing in my pocket. One more try. Nothing in this podcast or my writing or website is investment advice. The numbers of information for me and any guests may be incorrect. The views and opinions expressed may no longer be relevant or accurate.

Overall, investing is risky. This is not investment legal or tax advice. Do your own research. And with that, let’s get into the topic. Now the topic, obviously the learning concept for today, AI agents. And this is all fuzzy. Like, you know, I have a pretty I think solid concept to library on the web page. I think these are really economic, you know, business concepts to understand as they relate to digital and tech.

But, you know, oftentimes the concepts, they take a while to sort of flesh out like nobody talked about platform business models 20 years ago. There were all these kinds of other sort of attempts at describing what this thing was the bandwagon effect. People used to talk about the eyeballs. [00:02:00] It wasn’t until I mean, a couple of people really sort of crystallize it.

Look, we’re talking about platform business models. That’s it. And there’s four or five types. And that goes, then we kind of know, okay, AI agents, we don’t really know what we’re talking about yet. So, someone’s going to Crystallize the right way of factoring this into business, but it isn’t there yet. So fuzzy, but important enough that I think it’s worth talking about.

But that’s just kind of where we are now. A couple weeks ago, the NVIDIA CEO, Jensen Huang, a really interesting guy, he gave one of his annual talks and laid out some slides on how this all fits together. And you know, the person designing the NVIDIA AI chips Okay, that’s probably the person you listen to the most.

He’s a very interesting guy, Jensen Huang. He shows up in China fairly regularly. Taiwan, obviously. He kind of dresses funny. He dresses like an 18-year-old. He’s [00:03:00] always got these leather jackets and stuff. It’s kind of a funny guy, but, I mean, obviously wicked smart. Anyways, he put up a slide at his conference and basically argued that there’s a couple waves.

He breaks it into four waves.   I mean, he kind of starts this 2012, 2013, the first wave was sort of perception AI. I’ll put the slide in the show notes, but yeah, just sort of medical imaging, looking at JPEGs and saying, that’s a cat. That’s not a cat. I mean, that was kind of the first wave of AI was just perception.

And then, you know, you start with JPEGs, you start with, you move on from there to more complicated things like video, harder to recognize what’s happening in a video. Okay, fine. Okay. That gets you about 10 years. Generative AI, you know, breaks out in 2022 with widespread [00:04:00] adoption recognition, although it’d been, you know, been going on since 2017, 2018 with the, you know, the transformer model coming out of Google and things like that.

But obviously that’s been the big story. A lot of content creation. Generative AI. Content can be everything from text and poems, to code, to pictures, to video, to music. You know, the whole idea of content creation starts to become kind of fuzzy. Like, it’s like data. What is data? Well, kind of everything.

What is content creation? Kind of everything? Like it’s, it’s a fuzzy term, but anyways. Generative AI, content creation. The next wave he puts out there is agentic AI.   coding assistance would be type, which is, okay, you’re doing content creation, fine. But you’re creating code that can run things and do things, services, apps. [00:05:00]

So, you’re not really talking about content creation. You’re talking about agents, things that can act on their own to some degree. You’re talking about customer service. You have a problem. Where’s your package? Let me get you the answer and fix the problem. Okay. Patient care. Kind of interesting.

Anyways, this next wave is agentic AI. After that comes physical AI. Which is kind of the same thing, but in the physical world. A lot of the agentic AI, it’s stuff that happens in an app. It’s stuff that happens in a coding environment. You know, you can give an agent basically control of your laptop and it can open browsers and make orders and write things and replies to demos.

So, like what we’re talking about, agentic AI is this agent idea, but it’s in very controlled environments that mostly we have created. Like video games, an e commerce site, your laptop, that’s not the same as a [00:06:00] robot or a robo taxi cruising around the streets, just doing stuff. Okay. So robotic AI, Elon Musk calls it real world AI.

That’s kind of the next phase. So agentic AI 2025. Yeah, I think that’s a pretty solid prediction. Now, last week I did a podcast on sort of how China fits into all of this, because, you know, it’s kind of been shocking everybody, you know, deep seek was the one that got people’s attention. Turns out it’s awesome.

I use it every day.   it turns out its super cheap. Now the number they talk about, ooh, five, six million dollars to train and run this. Okay, that’s probably not true. But it is dramatically cheaper than 200 a month at OpenAI. And it’s open source. Which means, look, if you don’t trust their chat interface, oh, the Communist Party is controlling this.

Which is nonsense, but people say this. Fine. Download it to your laptop and run it [00:07:00] yourself as an open source. You can see the way you got it all. Okay. And that’s kind of what we’re seeing from China in the AI space is they’re doing what China does best. They’re taking something and making it dramatically cheaper.

Which is good for everybody. It’s good for people in China. It’s good for people in Brazil. It’s good for Americans. Why is Xi’an so popular? Because it’s really cheap stuff. EVs, cheap. And they’re making it open source as they take it international. So that’s DeepSeek. That was kind of the first one that got people’s attention in the last couple weeks.

Right after that, Alibaba’s Qwen model comes out. Everyone realizes, oh, it’s really cheap too. And it’s really good. They had to look up what this was.   ByteDance joined the, the fray in the last week with OmniHuman. We basically, you can load up a picture of Albert Einstein and his voice, and [00:08:00] it will recreate Albert Einstein giving talks.

And it’s really getting close to looking like a human. Like it’s getting hard to tell it’s not him. Now, if it’s someone new, like they’ll take comedians, I don’t know, Dave Chappelle. They’ll upload his videos. You can put it on. Okay, you can kind of tell it’s not him because you’ve seen him. A lot, probably.

When it’s Albert Einstein or Ben Franklin, you’ve never seen hundreds of hours of videos of this person. So, yeah, OmniHuman got people’s attention in the last week. Turns out it’s pretty great. They haven’t released this wide yet. They’re sort of showing what it can do.   the one that actually is really good from ByteDance is Daobou.

If you’re in China using a phone, you’re using LLMs. I use Daobou most of the time. I mean, it’s all Chinese language, but I actually kind of like it better than, you know, Qwen Model and the others. I think it’s, it’s multimodal, which means you [00:09:00] can, you can do photos and graphs and text and all of it. It works really quite well.

So, ByteDance, which was kind of criticized, rightly so, for missing the whole generative AI thing. Baidu was right on tap. I mean, they were right dead center, front of the pack for China for generative AI. If they were the front of the pack, ByteDance was kind of back of the pack. They kind of missed it.

Well, they’ve been playing aggressive catch up.   Daobou was great. It’s really I use it all the time. So anyways, that’s kind of where the West is and sort of understanding They’re kind of getting a sense of ByteDance this week. The next one will be Baidu probably with Ernie but yeah, they’re all good.

And if you actually just look at the numbers of AI engineers in China They have an incredible bench of talent. I mean it is just hundreds of thousands to millions, I mean the standard number is China graduates about two million engineers every year. [00:10:00] Okay, engineers are smart. You know where they’re all focusing?

1. There is an incredible population of AI engineers in China and their entrepreneurial ecosystem. So, they’re going to be major players like it’s US and China, not neck and neck, but depending what sector you’re talking about. Yeah. So anyways, that’s where we were. That was generative AI. That’s a little bit of a China version of that.

I wrote some sort of high-level theory papers on what your AI strategy about should be eight months ago, and I kind of said it was look, these days it’s not that hard. Step number one, you start putting generative AI into your internal tools. What your staff are using, have them start doing this every single day.

It’s going to make them more productive. It’s going to make them faster. It’s part of productivity, and it’s going to make them better. If you’re using generative AI tools to make movie posters, which [00:11:00] iQiyi is doing, they’re better. And it’s sort of a democratization of capability. Suddenly, you don’t have to rely on your marketing department to create your JPEGs and your movie posters.

Everybody in the company can do it. It’s pretty amazing. So as an internal tool, you start to deploy it, you get smarter. You make things faster, cheaper, and pretty much better. Also, by doing it internally, you don’t risk messing up your customer experience because, you know, if you’re using it for your bookkeeping and your HR and stuff like that, if it doesn’t work out, it’s fine.

Okay, so productivity tools, kind of number one. You can start to then move into the customer facing aspect. Put it into your products. Now, for some companies like Adobe, Okay, it’s transformative. It’s a 10 X improvement in what you can do with the Adobe. This is why TikTok kind of missed the boat because it is clearly an unbelievable [00:12:00] tool for content creators, which is half of what TikTok does.

Tools for content creators, videos for people to watch.   now the risk there, of course, is you can screw up the customer experience, but. Okay, you put it into Starbucks, you’re not, you should put it into your Starbucks, your marketing, things like that, but it isn’t going to 10x the experience of buying a coffee.

But in some products it can. So, number two, you start to put it in your products, you look for a 10x possibility. Number three, you start to build out your intelligence capabilities. This is a long-term build out your models, build out your apps, start to get your data layer in play because it turns out for most businesses, your data layer, your proprietary data is going to be your main advantage.

We’ll talk about that in a minute. And then number four, you start to build a moat. Which I’m still working on that piece. I don’t quite know how to build moats with this yet. It’s kind of my area. I don’t have a [00:13:00] solid answer yet. I’m working on it. The problem is there’s not a lot of data. It’s easy to theorize what might be a competitive advantage, but until we can see it in the numbers, the return on invested capital, the stability of market share, the defensibility against new entrants, until we can see it in numbers over at least one business cycle, you don’t really want to declare it a moat.

You can theorize about it, but I want to see that, I want to see it in proof. Hard to see that now because it’s so new. Anyways, that was kind of the framework I put for generative AI, the standard playbook for most businesses. All right, let’s get into sort of AI agents and where we are, which is, I think it’s going to be another playbook to tell you the truth.

I think it’s a whole another strategy. It’ll be the same tech stack. But I think it’s another strategy. Okay, so we start with use cases. What can we actually see working today as opposed to, Hey, that’s a really cool [00:14:00] potential idea. Hey, we’ve got a pilot, but it doesn’t seem to scale. No, we’re looking for use cases that have gone from pilot to scale deployment.

That’s real. Now, coding assistants. Okay, coding assistant. No doubt. It’s, it’s devastating. I mean, it is just like, you know, Mark Zuckerberg says they’re not hiring any mid-level coders next year. Something like that. I mean, it is just devastating to the number of coders that a company needs. You look at something like Stack Overflow, right?

That’s where if you’re coding and you have an error and you can’t figure it out, you put it into Stack Overflow and people help you figure it out. Okay. I mean, it’s just crashing in terms of usage. Everyone is just going to AI. If you have a piece of code, you know, check it. Where’s my error? It will explain it to you.

I can do this and I, I’m, I’m a minimal coder and I can start to write code just paste it in and it’ll tell me [00:15:00] where the mistake is. It’s kind of unbelievable. But then obviously it’s making massive productivity. You’re in low code, in some cases no code, where you can write apps flat out with it. Okay, so coding assistance 100 percent real and really Because of the nature of what you’re creating, it’s almost basically an agent to begin with.

You’re writing a code that is going to do something in an environment. It’s not like you’re writing an article and you’re just going to put it out. No, it’s going to do something. So, you’re really in the action phase. Personal assistants, getting interesting. We’ve seen this for a year or so. You have an assistant.

Okay, order me a pizza. It goes, it searches the local restaurants, it knows my history, it knows what pizza I like, it places the order, it comes back to me, okay, to spend the money, it fills in the Domino’s webpage, it puts in my credit card, okay, I [00:16:00] probably have to hold it to give it permission, I give it permission, it tracks it, you get the pizza, fine.

Personal assistance in the real world, much less, but personal assistance on like your laptop. You know, answer all my emails for me. Don’t send them out without approval. It reads all the emails, it writes the replies, check them, click, go. Fine. When you move over to the business side, you can see like specialized use cases like streaming assistants.

You know, you can go on to something like Notebook LLM now, which is a Google sort of content site. I can take five articles. Upload them there and instead of saying like, you know, write a summary, I can say create a podcast between two hosts discussing the article and it will create a podcast with two people talking about the article and talking about, oh, you know, digital dude, [00:17:00] Jeff Towson was talking about this last week.

Here’s what he said. Well, what do you mean? He said that, well, you know, we did this. That’s interesting. Do you think that, you know, and it’ll do a 30-to-40-minute podcast. Yeah. On its own with, you know, made up people and at a certain point you can tell it’s not human, but it’s pretty hard. An easier version of that is I can just upload some documents and then have a virtual agent who is a cohost who I will just have a conversation with me.

And you can do it like pre prepared or you can do it streaming. That’s pretty much there now. Simpler stuff. Okay, record a podcast, edit it, post it for content. Pretty much everything you do after you stop recording, you can hand it off and have it done now. It’s not hard. So, type of content, but it’s shifting into the agent role.

you know, sort of like an assistant. You move into sort of areas of [00:18:00] business like marketing, sales, bookkeeping. Yeah, they’re pretty close. I mean, it’s very easy to do sort of a brand agent. Well, not easy, but it’s doable to do a brand agent that I’ll explain more about that. But you can kind of see these agents emerging, at scale working quite well.

Fine, but most everything when we talk about agents, everyone in their mind often will get the idea of like a chat bot. Okay, I’m on the website. I’m on FedEx. I’m on Alibaba. The chatbot pops up. How can I help you today? And you can sort of text back and forth, but okay, so it’s autonomous. Right. When we talk about agents, we’re looking for a couple of things.

We’re looking for independence of action, but we’re also looking for more than just text and chatbots are kind of just texting communication. If a chatbot was a person, it would be a [00:19:00] person stuck in a chair who doesn’t have hands, who can only speak with you. They can’t type on the computer. They can’t go out and do things in the world.

No, they’re, it’s kind of like a human who can only listen and speak. So, fairly limited, but chatbots get you into this sort of idea of a very limited type of agent that can only do sort of call and response, like waiting for you to ask it a question, it can reply. Okay, when you move to AI agents, this is when people start to talk about digital humans, which is, okay, they can plan stuff on their own.

All they have to understand is your intention. They don’t have to wait for you to ask them a question. They can understand your intention. I want to sell more of this product today, this week, this month. That’s the intention. It understands the product. It can start to plan how to [00:20:00] do that on its own without us.

So not a call and response and it can take action. So that’s kind of how I think about AI agents. I think about can you sort of autonomously plan just based on understanding my intention and then can you take action. So, in that sense it starts to look a lot more like an employee. If you have an employee you just want to meet with them once a day and say look we’re trying to get this done and they go figure out how to do it.

And they go and do it and 55 phone calls today.   I sold the item to 5 people. But then I got some feedback on some of the other customers that maybe this wasn’t right and so I came up with an idea And I pitched that to them and then these other people bought, right? Autonomous planning, action, feedback, reaction, really within an environment.

So, responding to interaction planning in an environment. [00:21:00] Now the environment part is kind of the key bit. There’s sort of two things that I think about with regard to this. Okay, so it’s planning and its action. responding, interaction, fine. What is the environment, and what data is required to work? So, most of the AI agents I’ve been keeping an eye on are all sort of in controlled environments that are usually man-made, like a video game.

So NetEase which I talked about before, you know, they’re creating intelligent NPCs. So non player characters you engage with within the video game environment that are becoming smart, autonomous, can sort of act on their own. Fine. Or if you’re in a shooter game, battle bots, battle robots. They’re in the game, but you can have teams with you.

Then you go storm the castle together with your teams. And they’re basically independent, fighters. Okay. So, the environment of gaming makes a lot of sense. The environment of an e commerce site makes a lot of [00:22:00] sense. So you go to JD, you go to Alibaba, they’re deploying sales agents for their merchants that can interact with customers within the confines of the e commerce site.

So, in both cases, environment sort of defines what the agent can do. The one that everyone sort of paid attention to, well, most people paid attention to last week, was OpenAI’s Operator, where you’re basically deploying an agent on your laptop. Which is a controlled environment, and it can open browsers, it can log into your email, it can hunt for information, it can place orders for pizzas, it can download new apps, but you’re giving it a controlled environment and sort of letting it run.

So, I haven’t played with that one. I don’t have a, whatever that pro account is for   open air that you can use that. But, anyways, that’s kind of how I think about it, um. [00:23:00] And when I was listening to some NVIDIA people talk about this, they kept mentioning the same thing. They said, look, AI agents, it’s about three engines that you have to build.

You need a perception engine, you need a cognition engine, and you need an action engine within an environment or ecosystem. But you need all three.   the perception engine would be, you know, if it’s online, it can read text in various web pages, that’s fine. If you’re giving it cameras, it can see other things.

But it needs to sort of gather data. and understand it. And the question then is, well, how developed are each of these engines? Where’s the bottleneck? Cognition engine is reasoning, understanding. That’s all these reasoning models like DeepSeek that can plan in multiple steps, can get feedback, cannot think like humans, but can act with reason and logic in a way that like, hey, create me a JPEG.

There’s no [00:24:00] reason or logic engine in that. And then the action engine would be, okay, how many. How many methods of action does it have? Can it just order stuff on a website? Can it create, can it roam the streets on its own? Physical AI. So, they kept talking about the perception, cognition, and action engines and where they are today.

and the phrase I heard used was the AI agents today that seem to be working are passive intent recognition. So, they’re still mostly passive. They’re watching. But they are having some degree of recognition of your intent. They can’t really understand everything you’re doing. They can kind of understand what we’re trying to do.

And they can more and more sort of try and understand your intent without waiting for you to say something. So that’s kind of where passive intent, [00:25:00] and then you basically pair recognition, I’m sorry, recommendation engine with a limited cognition engine. So, the cognition engine is limited, recommendation engines, we know those well, that’s search engines and whatever.

So, it’s kind of, that’s where we are. They’ve got just enough understanding of your intent and the context to start to make smart decisions on its own. So, what does that mean in practice? That might mean, okay, you’ve got an AI assistant that’s helping you design a building. And it can, it basically has this okay understanding of your intent.

And when you start to change, let’s say the top floor, it will understand how that impacts the rest of the building as regards to your intent. And it can start to say, if you’re going to change the top floor, we’re going to have to take, you know, change the sewage pipes coming out the back. Because it’s going to change the amount of sewage or water coming into that top floor.

And I recommend you do it this way, and it’ll sort of redesign it. [00:26:00] Right? So, you can kind of see that like, okay, it understands what we’re doing, it’s got a reasonable understanding of our intent, and it’s got enough reasoning and cognition ability to take the next steps back from that and say, okay, based on what you’ve said, we’ve redesigned the back sewage system for the building.

Here it is. Proactive. Alright, so based on that, let’s actually talk about a business case that, you know, is something you should probably be thinking about. So, there’s this idea of brand and marketing agents. That, you know, we can deploy these agents, they can go out, they understand our intent, they have some cognitive ability, and they can go out and sort of act proactively and autonomously.

Now, why would that matter, let’s say, if we’re doing branding and marketing? Typically, if you’re doing a branding, let’s do a marketing campaign. We need to do a marketing campaign. We’re new to this market. We want to convince people who we are or maybe to start trying our product. Fine. What do you do? You look at the market. [00:27:00]

Let’s say Dubai and we’re going to do some beauty tech product. I don’t know facial masks with LED lights on it that help you with Okay. We look at the market. We start to segment the customers. Usually, we’re looking at behavior. We’re looking at what they care about. Demographics are fine, but really we’re looking about let’s break customers into buckets of what they care most about.

and their behavior. When do they shop? How do they shop? What do they care most about? Based on that, let’s say we get 6, 8, 9 segments within our segmentation. For each of those, we start to customize the message we want to deliver. We’re not going to if we’re talking with 45-year-old women who are worried about aging of skin on their face Different message than let’s say, you know, 21-year-old who wants to be on top of the new trends and whatever, right?

So, we write a [00:28:00] message for each group then we got to find out where they are Because they’re not on our website. We don’t they’re not on our property We got to go out in the world find where they are and push the message to each group where they are and then we sit back at our website or app and we wait for them to show up.

When they show up, we start to engage them. We have to learn who they are. Hey, how did you hear about us? But really, we don’t know that much about them. This is the first they’re learning about our company. And then we repeat and repeat and repeat. Standard marketing. Well, simplistic marketing. All right.

What if we have brand agents, marketing agents? Okay, the first step now is we tell them what our intention is. Is our intention just to make sales? No, not really. It could be education. It could be increasing awareness. It could be driving them to a source [00:29:00] of content that we like to, you know, put out there that people, you know, to get our message out, to watch some of our videos.

It could be that sales is due, come to our website, fine. But our intention can be more complicated. Okay, we’re not going to necessarily give them our customer segmentation. We might a little bit. We might give them a basic customer segmentation. We might suggest messaging. But to some degree, we want it to figure these things out to some degree on its own.

Because we’re making a lot of assumptions of who our customers are. And we’re making a lot of assumptions about what the right message is. So, we really want them to sort of go out, we want to deploy AI agents that go out. And hunt and that gather data and that look around and look at traffic and we want them to then sort of identify customers that would achieve what we want, and we want them to surface or [00:30:00] push a message or a piece of content or an action at the right moment in the exact right way to the right person.

Right? That’s what AI is good at. It doesn’t need to; it can deal with millions of individual interactions on its own. Volume and complexity is not a problem for it. It is for us; we have to simplify into segments. But, and we really don’t just want to push one message. We want to surface the right message.

Maybe the right piece of content. Maybe the right comment. Or engagement or post, maybe the right promotion, you know, we want to surface the right type of action to the right customer at the right moment when they’re ready to hear it in the right way. So, we were really not talking about messaging. We really start, we’re talking about experience.

We want to engage and create an experience for someone that achieves our intent, [00:31:00] right? Now this is. Now, obviously for those of you who’ve been listening, this is kind of my big thing is I’ve been focusing on what I’ve been calling extreme personalization and customer improvements. All this tech stuff you can get lost in.

You got to move to the moment that matters most, which is taking text, content, experiences, services, deals, promotions, all of it. And targeting it to the customer at the right moment in time in the right way to improve the experience for them. If we can do that, we will get the things we want, which might be data, it might be more engagement.

It might be more interactions. It might be more sales, but we don’t push for sales. We push for high value, high impact experiences at the right moment, at the right time. That’s kind of what this whole extreme personalization thing is. Well, you can see that AI agents are kind of built to do that. [00:32:00] And if you do that, they will show up at your website.

They will learn about who you are. You will raise their awareness. But it’s all at the end of the day business. You’re in the business of. Creating value for your customers, which can be a product, a service, an experience, an interaction, a piece of content. It can be all of that. And you want to really play across the board.

Well, AI agents, compare what I just said for AI agents, as opposed to the typical marketing campaign, right? You can the AI agents, it’s a big part of my playbook to tell you the truth, AI agents. Generative AI is a big deal. Standard AI is a big deal. But intelligence and AI agents are going to be huge. In terms of doing this.

Anyways, now what would you need to do this with a brand agent or a marketing agent? This is where it gets interesting. The first thing you need is you need feedback loops. An AI agent is, most AI really, [00:33:00] is no good without feedback loops. You’ve got to try lots of things. See if it works. The agent gets smarter, smarter overall, smarter with specific types of customers, smarter with individual customers.

You need as many tight feedback loops as you can possibly get so that the agent gets smarter and smarter. That’s number one. The other thing you need is you need proprietary data. This is kind of the funny thing about AI. It’s so data dependent that It doesn’t have any ability to remember things like AI right now, 2025.

It looks like from what everyone says, people I respect it’s about at the level of a PhD. Graduate student now, a year ago, it wasn’t, you know, two years ago as a high school student, now it’s PhD 130 IQ and going up and soon it’s going to be superhuman. Okay? So, it’s an incredibly smart person [00:34:00] who can’t remember anything past 20 minutes.

It forgets everything after 20 minutes. Why? Because it requires so much data to be flooding into the inference that you can’t possibly record all of that. Like everyone talks about NVIDIA’s, you know, GPUs and all of this. You can’t build enough memory chips to remember any of this. It’s, it’s, it’s physically impossible.

The data requirements are so huge that it’s this really smart person that forgets everything in 20 minutes. Which is a weird way to work with someone. If you have an employee, that’s great, but they don’t forget everything you told them an hour ago. AI kind of does. So, who is going to have the historical data to provide context?

Well, businesses. Yeah. Merchants and brands that have recorded their past interactions and their past customer history, they’re going to have the database that’s going to give this thing [00:35:00] more memory. It’s going to be the big weakness of something like Chad GPD. It can’t remember what you asked it yesterday. Not really.

I mean, maybe it has a tiny bit, but it’s physically impossible given its scale to have any sort of real memory. But a business with a good customer database can pair that with the PhD student. So proprietary data, which is going to be a merchant and brand thing, that’s pretty cool. Now, the other side of this equation is merchants and brands should be building this thing.

Humans, customers, consumers, us, we’re going to have our own agents as well. So, we’re going to, it’s going to be like two armies facing each other. Right. All the businesses will have armies that they deploy and we all have our own sort of like teams. Maybe not an army, but we’ll each have like 10 to 15 assistants that deal with this army.

So, we don’t have to. Now that’s really interesting because I’ve [00:36:00] described like how do you get information in this world? Up until recently it was Google search and I’ve sort of made the analogy like Google is the world’s largest library. And it’s so big, there’s no way to find the book you want. And Google controls the card catalog.

So, to find whatever book out of, you know, no infinite books. Now you’re incredibly dependent on the card catalog and Google controls the card catalog and everyone plays this SEO game, which is basically paying Google to move your card. Further up in the card catalog, right? It was a tremendous position of control and power Because it’s necessary.

There is an Unbelievable bottleneck between all the information of the internet and our brains and that’s you know That’s Google and it’s the screen of your smartphone. Honestly, okay This is kind of different then, you know, if we’re talking about armies of agents [00:37:00] You know, roaming the streets as robo taxis or constantly coming at us because every brand has these armies of brand agents and they’re all coming at us all the time.

That’s not really the card catalog. That’s something new. I don’t really have a great analogy for this. The one that I always think about is, is that show The Walking Dead. When they all live in the prison and all those armies just circle the prison and they’re like always pushing the fence trying to get in.

That’s kind of how I view like these armies of AI agents that are just out prolling looking for customers all the time. And how do you get to people? I’m not sure who’s going to control that, but it, it looks like a different model, than we’ve seen before. And generally speaking, it’s going to be better for merchants and brands.

We’re not going to be so dependent. on one company and their endless, you know, auctions to get keywords and all of that. So that’s kind of how I see it for certain brands, is you want, [00:38:00] you know, this army of people out there. I need a more positive analogy than zombies. It’s got to be something better than that.

Maybe traveling salesmen, like people who used to knock on your door to sell encyclopedias. For our side of the equation, consumers, customers, I used to talk about autopilot versus copilot. The analogy I like now is the Jaeger. If you ever watch these movies, Pacific Rim, where these giant monsters come out of the ocean, the kaiju.

And so, then humans get in these huge, massive machines of robots, like, you know, 50 stories tall. And you sit in the head and you run around like the Jaeger. Like, I want my own Jaeger. I think that would be awesome. If I could do stuff like that. Anyways, that’s kind of where I am with, with my thinking. My recommendation for most businesses at this point is like, look, you got to start deploying AI.

If the message from last year was you got to start. AI. The message for this year is you’ve got to start deploying an AI agent, maybe not a sale one. That’s a little bit more [00:39:00] risky. Maybe do it internally. But this is going to be the next year. And you can kind of see in terms of all of this digital strategy, you know, the sharp end of the spear is can you improve the user experience?

That’s when all this matters. AI agents are going to be a huge part of that. It’s clear that’s true. So, yeah, you got to get to where the action’s happening and that means AI agents, so start rolling them out. And I’ll talk about this more, but this is kind of where my thinking is right now. I’m learning myself, honestly.

I’m not even close to being there, but I think I’m getting there a little bit. That’s sort of the goal for today. As for me, just doing well, working like crazy, having a good time. It’s been, it’s been a very good week.   I’m always kind of happy when I’m super productive and I haven’t been on the road this week.

I’ll be back on the road shortly. So, yeah, that’s not bad. I guess the big thing for me this week was the, the U. S. You know, I’ve been talking on this podcast for several years now about information [00:40:00] control. And how, you know, control of information is really control of how people think. And, you know, my, my sort of hot button issue has always, has been for a long time, this sort of censorship regime that was built between about 2015 and 2020 to control information flows.

And it was some unholy alliance between a couple Silicon Valley firms, governments, definitely the US, but some other governments like. Brazil, the E. U. For those in the E. U. You know, the E. U. Digital Services Act is like horrifying what it is. I mean, this is just outright information control. But there was this hunt.

Holy Alliance that was really destructive when it was horrible in its own right, but destructive in terms of censoring people, silencing people, pushing false narratives, which we saw during Cove in and some other things. I know Elon must kind of took a sledgehammer to [00:41:00] that. With the Twitter files and exposed it, but what Doge, which is again Elon Musk, has been doing at USAID and some of these government agencies within the U.

1. is really exposing the machinery, the censorship regime. It’s, it’s unbelievable. Everyone thinks USAID is like some sort of aid and organization. It’s not. It’s the other half of the CIA. I mean, it’s it is everything that the State Department Wants to do that’s probably dirty and they can’t do it themselves They put it there and so they you know, they do all these things like they’re funding Media companies and insurgency campaigns all over the world and you know, some of it’s good.

Of course It’s not nefarious by its design, but a lot of the censorship regime Was coming from there, and we’re just seeing it all get exposed this week, like, all these grants to media companies and censorship [00:42:00] organizations in Brazil, Serbia, Asia, Mexico, you know, all of this. It’s, it’s like the, the machinery’s being exposed, and it’s pretty shocking.

A lot of really well-known Western publications, we’re taking money from them. And look, if you’re a paying journalist, and you’re paying reporters, I’m sorry, media organizations. To some degree, you own them. Maybe not, it’s not a hundred percent, like, but it will shade what is covered. Absolutely. You give money to politicians, you give money to scientists, you give money to reporters and media artists.

It’s, there’s a reason they’re doing this. You know, big pharma companies spend a tremendous amount of money, giving free things to doctors. And it absolutely impacts how they prescribe. Totally works. You know, there’s hundreds of thousands of pharmaceutical reps in the United States, more than doctors.

Like, for every doctor in the [00:43:00] U. S., there’s more than one pharmaceutical rep lobbying them and giving them free stuff, right? Now, you can’t give them free money, but you give them, you know, I used to be in medicine. The pharma reps would come by every couple day and have free lunch, right? Ooh, Thai food today.

And pharma company is, is having what food today? And some companies, some reps, they would bring better food. Ooh, sushi day. And, you know, we’d all go down and get free sushi. So, this whole money for influencing is pervasive and it totally works. I don’t, I never belied this like, oh, it doesn’t work. It totally works.

So, when you see huge amounts of money from the U. S. government going through the National Endowment for Democracy, USAID, and funneling it into media organizations, activist organizations all over the world. It’s an, it’s a, it’s a mechanism of statecraft. Some of which can be white hat, some of which can be black hat.

Anyways, listen to like Mike Benz, go look at what Doja’s uncovered this week. [00:44:00] It’s the joke that I’ve heard a bunch of times today. It’s like we are all, you ever see that, that movie, the Truman show where Jim Carrey, you know, discovers he’s in a TV show and he wakes up like the joke is, have we all been living in a Truman show?

And we’re just discovering like that. It’s not real. And we’ve just. you know, ran our boat into the side of the wall, and we realize it’s like a good portion of reality has been falsely constructed. Is that true? I don’t know, but it’s an interesting analogy. Anyways, Doge, USA, National Endowment for New Arts, take a look at it.

It’s, it’s completely blowing my mind when you actually see the money spent line item by line item by line item. Anyways, that’s what I’ve been thinking about, just because it’s kind of in my in my hot button issue of information control and you know, freedom of speech. That’s kind of my political rant is always in that area.

Anyways [00:45:00] interesting. That’s it for me today I know today was a bit sort of more general and hand waving, but yeah AI agents I mean, I can’t recommend strongly enough get on this topic. It’s going to be huge. Anyways, that’s it for me. Take care I’ll talk to you next week. Bye. Bye

———

I write, speak and consult about how to win (and not lose) in digital strategy and transformation.

I am the founder of TechMoat Consulting, a boutique consulting firm that helps retailers, brands, and technology companies exploit digital change to grow faster, innovate better and build digital moats. Get in touch here.

My book series Moats and Marathons is one-of-a-kind framework for building and measuring competitive advantages in digital businesses.

Note: This content (articles, podcasts, website info) is not investment advice. The information and opinions from me and any guests may be incorrect. The numbers and information may be wrong. The views expressed may no longer be relevant or accurate. Investing is risky. Do your own research.