Related articles:

• Two Lessons from My Visit to Tencent Cloud (1 of 2) (Tech Strategy)
• Tencent Cloud and Mini Programs Go International. Lessons from My Visit to Tencent HQ. (2 of 2) (Tech Strategy)
• How Amap Beat Baidu Maps. My Summary of the Alibaba Playbook. (Tech Strategy – Podcast 252)
• Scale Advantages Are Key. But Competitive Advantages Are More Specific and Measurable. (Tech Strategy)

From the Concept Library, concepts for this article are:

• Agentic AI
• GenAI / Agentic Operating Basics

From the Company Library, companies for this article are:

• n/a

———transcription below

Welcome, welcome everybody. My name is Jeff Towson and this is the Tech Strategy Podcast from Techmoat Consulting. And the topic for today, how to build an agentic AI operating model. Now I’ve been working on this for about a year, I suppose, which is the idea of what’s your strategy for AI? Is it significantly different than a digital strategy, which I’ve written a whole lot of books about. question, but this is kind of new and evolving. And I think I’ve got most of it now, so I’m going to start laying it out over the next week, a couple weeks. First half will be the operating model, you know, the operating basics, the stuff that every company basically needs to do. And then the other bit will be the competitive advantages, the business model, the moats, all the things that, you know, they give you, in theory, dominance. So, I’ll break it into two pieces. Today’s going to be the first part of that, which is the operating model. And that will be the topic for today. Housekeeping stuff. We have the China, Shenzhen, Greater Bay Area tour at the end of November, pretty much the last week of November. Going to be pretty awesome. We’ve got about 70 people signed up at this point. And yeah, we’re going to spend a couple days there going through sort of Shenzhen as in two ways. Number one, sort of bottom up. company visits, looking at sort of the leading companies, so deep dives, that sort of thing. So definitely we’re focused kind of a lot on the investor perspective of all this. How can we learn more about these important companies? And then top down, which is, okay, let’s talk about sort of the macro situation. The Greater Bay Area is a really big initiative. You know, it’s the whole area between Hong Kong, Macau, Shenzhen, Dongguan, Guangzhou, arguably the largest urban environment in the world that’s basically being stitched together right now. So, there’s kind of a good macro top-down piece to this too. We’re going to do both. Anyways, if you’re interested in that, send me a note or go over to techmoatconsulting.com. You’ll see all the details. Price is $900. Not counting flights and hotel, we just do the content. And it’ll be over three days. Anyway, if you’re interested, let me know. And let’s see, standard disclaimer, nothing in this podcast or my writing or website is investment advice. The numbers and information from 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 concept for today, obviously, agentic AI operating model as a half of the equation for a strategy. But really, I we can almost think about this like a new tech paradigm entirely. It’s not just agentic AI. It’s this whole, we’ve been kind of living in the digital age for the last 20, 30 years. This is a new thing. I you could go back and say, well, it’s an agriculture age. Then it was an industrial age. Then it was a digital age. Well, now we’re getting into the agentic AI age. And we can see the main pieces at this point. I definitely generative AI kicked it off. know, generating content. And it turns out pretty much everything is content, whether it’s code or text or images or whatever. Then we get to agents, which are, you know, basically autonomous versions of generative AI. They can make decisions, make predictions, make decisions, and then execute them. So, it’s really a high-level form of decision making and automation. We also get embodied intelligence, idea that we’re going to put this stuff into robots and, you know, things walking around are going to have some degree of, you know, autonomy, perception, intelligence. And kind of cutting across all of this, we have the idea of machine learning and the idea that we’re going to build intelligence into businesses. Because typically the only form of intelligence in any operating system is human beings. Well, now we’re going to get other forms of intelligence built into systems, whether it’s business or whatever. So, we can kind of see the main pieces that are going to define, let’s call it the agentic AI age. And it’s a pretty big break from the digital age. So, we can look at it that way, which is kind of how I see it. Now, when we start looking at what would be a strategy for that, if you want to know digital strategy, fine, I’ve written seven books about that. I’m going to clean those up in the next month or two and make some edits, I think, to mediate on. I’m pretty much, my thinking has been done. It took about seven or eight years of work. But now we get to AI strategy, and we don’t have a lot of data yet. We don’t have a decade of companies, two decks of companies where we can look at their financials. It’s all sort of emerging right now. So, it’s kind of hard to formulate the strategy. A lot of early-stage stuff. But I mean, we can break it into business model and we can break it into operating model. Business model, which is kind of my area, motes and marathons. Are we going to see new competitive advantages emerge that we haven’t seen before? Network effects were a big deal in the digital age. We hadn’t really seen those before that. Are we going to see no barriers to entry? Learning and adaptation as a customer or competitive advantage is something I’ve been sort of thinking about a lot. The flip side to that coin, are we going to see powerful business models and powerful competitive advantages get wiped out? It used to be in the pre-digital age, print newspapers were about as powerful a business model as you could have. Local monopoly, cash machines, tremendous competitive advantages. Digital just wiped that out entirely. They exist, most of them aren’t economically viable and none of them have any real advantages. Are we going to say the same thing? And I think we already can. One of the big network effects that people talk about is standardization network effect. You know, in the real world, if everyone speaks Spanish in Mexico, there’s a network effect because the more people that speak it, the more valuable the language is. You save costs, your coordination costs go down. You we see standardization-based network effects in software. We see it in language. We see it in systems that connect. Well, Agentic AI doesn’t really need that. I can speak in Spanish into an AI system. It can hear it and then… translate to someone else in French. And that person can speak French back to me and all here in Spanish. This whole idea of a standardization network effect might be getting wiped out because AI has no problem operating between different systems. So, we’re going to see some of these major business models that we like get hammered pretty good. So that’s kind of the first bucket. I’ll go through that in the next couple of weeks. But today’s the second bucket, which is operating model. And you my standard thing is the digital operating basics and basically you have to be smarter, faster and better every year. And these are just the basics. Everybody’s got to do it. Okay, so what are the operating basics in an agentic AI world? Now I’m going to break this into a couple areas. I’ll talk each in turn. We can sort of think about it as internal versus external. The internal piece is easier. These are the operating systems that are internal to an organization, a business. This is mostly under our control. We can change how we do finance. We can change how we generate reports. It’s sort of an isolated system. When you start moving to external activities, workflows, operating activities, like engaging with customers, communicating with customers, well, then you’re out in the ecosystem. You may be dealing with Facebook, you may be dealing with Google, and so the whole agentic thing, it’s not just changing the operations of a company internally, it’s also changing the external world as well. That piece is a lot more complicated. So, I’m going to work on the internal piece mostly, and I’ll raise some questions about the external sort of facing piece at the end. Okay, so question number one. How do AI, I’m basically saying agentic AI, because I think these are the same thing. Let’s say, does AI, specifically AI agents, how does that impact workflows? Now lot of what I’m going to tell you is from a couple McKinsey papers. McKinsey has been doing really well on this subject recently. I kind of read everything that comes out, and some of it, a lot of it’s not too helpful. Whoever is doing this over at McKinsey seems to, that team seems to be on the game. So, they have some good stuff. So, I’ll put the links for two articles in particular I think you should read. A lot of this comes from, well, 40%, 50 % comes from that. Pretty good, but their argument, which I think is right, is you got to stop thinking about agents or tools or use cases even. You want to think about work floors. How can we integrate AI, agentic AI? into the workflow of our business. And you got to basically map it out. But this doesn’t translate into value. Value for the customer, value for another user, which could be internal staff, could be developers. The whole AI thing doesn’t move the needle in terms of value until you basically look at it as entire workflow. where this workflow, could be, let’s say, writing code for our internal programs at a company, until that workflow gets improved, cheaper, faster, better, we’re not going to translate any of these activities into real value. And definitely if we’re dealing with customers, right? So that’s one of their points. Don’t think about the agent, think about the workflow. Workflow is usually a combination of people, processes, and technology. I used to do this 20 years ago when I was consulting at companies in like St. Louis. We would map out their workflows and it was always PPT, People Process Technology. All of that goes into getting an activity done. Now, we can basically break the workflow into four layers and four stages. And this is really important. I’ll put two graphics in the show notes. These are McKinsey graphs. This is super important. Like I think if there’s only one thing you remember from this podcast, remember that. Four levels, four stages. Okay, so we look at a workflow. Let’s say doing our standard monthly financial reports. Okay, we can break it into four levels. Level one, this is McKinsey, rules-based systems. Okay, that’s traditional software. You write the software. which is pretty much how everyone runs their finance departments. You run the software; it does the same thing every week. It’s very rules-based and deterministic. That’s level one. Gather the data, put it in your spreadsheet, calculate the numbers, generate a report. It happens the same. That’s rules-based system, deterministic software. That’s level one. Level two, we could call analytical AI. Now people have been doing this for 20 years. You know, we look at the existing data, we try and find correlations, we try and make predictions. Sometimes this is called predictive AI. Okay, this is not as rules-based, this is more statistical and probability-based, but maybe we’re going to generate some risk scores for various things. That’s been known pretty well, you know, 20 years of this. So analytic or predictive AI. Level three, generative AI. Now we’re starting to maybe make some reports. We’re starting to write text. We’re starting to create government documents that need to be filed. We’re creating stuff. It’s not as standardizable. We’re suddenly in the non-deterministic world. Suddenly we’re using LLMs and foundation models. And as you know, if you put the same question into an LLM three times, you’ll get three different answers. So, it’s a lot more like dealing with a human. You’re going to get some fuzziness. You’re going to get mistakes. You’re going to get non-consistent answers. That’s the difference between sort of deterministic and non-deterministic software. So, level three would be generative AI. Level four would be agentic AI. This is where we can start to automate. Make a prediction. uh Generate the report. Go file the report. The action. would be the agentic AI. Agentic AI is usually a couple things. It’s usually four things. You have a foundational model at the core. That’s the brain. It could be GPT or something. On top of that, we have an orchestration layer where that is kind of where you’re telling the agent what it’s supposed to be doing, what goal it’s supposed to be doing, how it can do it. And then it will go down into the GPT or whatever to run its model and make some decisions. And then the other bits are it can then interact with the external world or other systems, which is usually one of two things. It’s either gathering data, so it has a lot of APIs and it’s pulling in data, or it’s taking action. It basically has a toolbox. So, I generally think about agentic AI as, look, it’s a standard chat GPT model, it’s a standard GPT model. But on top of that, you have an orchestration layer and then you have data access and a toolkit. And some of these agents, you may give them a couple tools like you have access to Expedia and you can book a ticket. You may have access to the Visa MCP and you can pay for things. How big of a toolbox you choose to give one of these agents is up to you. Some of them can be very big. So that’s kind of how goes. But you know, in this level four, we’re talking about agentic AI, you it can start to take actions and things like that. you can map out the workflow. If the workflow is going left to right, then the four levels are going top to bottom. And under each step of the workflow, you want to map out what happens at each step of the workflow and at each of the four levels. So, the rules-based system for generating your monthly, weekly financial reports. Might be gathering the structured data from the data warehouse, generating the standard reports like always. That might be the first step on the top. Below that, analytical predictive AI might be assigning risk scores. Level three, generative AI might be generating different reports for different departments or the government. Agentic AI might be filing it. So, you kind of want to map it out that way and I’ll put a good graphic in the notes. It’s really worth looking at it. So that’s the four levels. I’ll get to the four stages shortly, but within the four levels, you map out your workflow, you put it against four levels, and then you sort of, you can choose your areas to focus on. You can focus on pain points. You can focus on the high impact use cases. Usually, people tend to break this down on two dimensions. How high of an impact is it? And then what’s the feasibility, do-ability? And ideally you want to do the high impact use cases that are pretty feasible and doable for where you are right now as an organization. Now as mentioned, one of the big trade-offs here is when you look at the various steps, if you have sort of high standardization and low variance, like the same financials you look at every month, okay, you don’t really want to use an LLM. You don’t want to use AI. You don’t want to use AI agents. because they’re going to do it differently frequently. And that’s going to kind of create a lot of uncertainty and complexity in something. No, you want to do traditional software, analytics, the basics. On the other hand, if you’ve got something that’s sort of low standardization and high variance, like let’s say you’re looking at market information coming into your business every month, well, then you’re going to want to use something that’s more like an LLM. It’s going to have to be a little bit fuzzier, non-deterministic, but the data is going to be different. So, you can kind of think about that. There’s a good quote, and this is from a McKinsey article. They basically, I’ll read the quote, quote, on-boarding agents is more like hiring a new employee versus deploying software. That’s a pretty good way to think about it. That these agents, really don’t want to think about them as deploying software. where you do it once, you set it up and you can leave it. No, it’s a lot more like hiring an employee, training an employee, managing an employee, overseeing them, checking their performance, doing additional training. It’s kind of like bringing on a person more than anything else. If you use a lot of generative AIS, I like this term that people use AI slop, where a lot of what it generates is just. You ask it a simple question; you get two pages of just gibberish slop. I just wanted an answer, right? And then you get a lot of quality problems as well. So, you basically have to manage your agents like people, which means in terms of your workflow, if you’re having agents at different steps in your workflow, yeah, you got to track and verify the agent performance at each individual step. As you go from step one to step two to step three in your workflow, it’s got to be tracked and checked at every step because we’re going to scale this up in the next section. So, you can’t scale this up significantly if you don’t have each one locked down. But you’re going to catch their mistakes, refine their logic, give that feedback to improve its performance, but that’s going to be the system you’re going to build on AI agent within a workflow. Okay, let’s move on to four stages. So, four levels, four stages. Now, this is kind of about the evolution of what you’re doing. Now what most people are doing right now with most businesses is they started giving, I’m talking about agents, not AI. Agent’s sort of are autonomous versions of AI. A lot of what people are doing right now is they’re giving AI tools to employees and sort of augmenting their performance. Everyone’s getting much more productive, much faster, and usually better as well. The quality is going up, the cost is going down, the speed is increasing. That’s all true. But that’s really humans giving AI tools. I’m talking about AI agents that are basically autonomous. You don’t have the person. Now, sort of stage one is where a lot of people are. We can call this individual augmentation or individual automation, were, okay, we’re looking at an employee or a step within a workflow. We already know what the workflow is. That’s the existing business and we’re just kind of automate that step. Now we could do it with a human with AI. In practice, it’s probably going to be a couple things. In practice, it’s probably going to be a human with AI as well as an AI agent. It’s not going to be entirely an AI agent, but yeah, that could be stuff like drafting contracts, uh summarizing meetings, things like that. Well, okay, those work steps you can probably just give to it. We call that sort of individual automation. That’s step one, that’s where most companies are right now. Step two, I’m sorry, not step two, uh stage two. This is when you start to automate workflows or significant portions of a workflow. And suddenly the thing is starting to happen on its own. There may be no human in the workflow at all. or maybe it just in part of the workflow, but we’ve kind of taken it to the next level. And this is where you’re going to start to see sort of pretty big gains in cost, efficiency, speed. But again, we’re talking about existing workflows, things that have already been built in the business over many, many years. We’re probably talking about low complexity workflows as well, stuff that’s pretty standardizable. Okay. That’s cool, that’s stage one, that’s stage two. Stage three is where things start to change. This is when we start to use teams of agents, not just an individual agent, teams, five, 10, 15 of them on workflows. And we’re not just sort of improving and automating the existing workflow, we are redesigning the workflow to be AI first. This is when people are going to start doing things in completely different ways. We’ve kind of, you if you think about it, we basically have workflows based on humans. Well, and then we’re adding tech into that. We’re adding AI agents into that, maybe taking over a couple steps. This is starting at the other end point. We’re going to design the work starting with AI agents and go from there, which could get you very different ways of operating. And we’re not going to talk about one. We’re going to talk about teams and the teams work together. Okay, that’s getting us to what we’re starting to call the sort of, you know, agentic AI operating model, which may be very different. Now in reality, it’s going to be sort of an AI hybrid, sort of a human AI hybrid. But yeah, we’re going to see completely new operating models. And then stage four is where everything takes off, which is we don’t just have AI first workflows being managed mostly by teams of agents. we start to get cross-functional systems where all the workflows are integrating together within the organization and it’s mostly the AI agents. So, we can call that cross-functional AI agentic systems as opposed to workflows. That’s a whole new thing. And that’s, I’m not even sure what that looks like for a bank, let’s say, or a factory. We can see some easier versions of this like Robotaxis. Robotaxis are basically agents. So, we are replacing the whole idea of humans driving cars with Bioagents. Fine. Stage one, stage two would be we get rid of the drivers, we have the AI drive the car, but the operating model is pretty much the same, which is a marketplace platform that is now becoming sort of an AI service. Okay, is that really how Robotaxis are going to work? Are they not going to be completely because the whole idea is why did we have Uber like it is? Well, because people owned cars and each car has one driver. What if these things are little go-kart like vehicles? What if they’re big trucks? What if they’re little modular vehicles that can go and then hook up with each other? Once we break free of sort of the human legacy in terms of operations, we could see very different operating models. OK. So, what does that mean in practice today? In practice today, you’re probably seeing two things. You’re uh functional, agentic workflows, and then cross-functional, agentic workflows. We’re not seeing the AI-centric operating system yet. We’re kind of seeing stage one and stage two. So, let’s say an agentic workflow where it’s mostly agents at this point, but it’s in a very functional silo like financial reporting, inventory management, supply chain management. Well, I there are vendors everywhere selling these services today. Sort of very functional workflows. The other side would be sort of more cross-functional workflows like end-to-end customer management, customer experience management, or you’re going to of touch on a lot of different. parts of the company, but you want one sort of system to oversee that. Field operations, call centers, things like that. But generally, more complexity, probably a higher level of decision making is required. But yeah, we’re going to see that today. Okay, so what do you do if you’re a CEO and you’re trying to build a more agentic operating model? Yeah, a couple things. Okay. You kind of do what I just mentioned. You map out your key workflows, you put them against four levels, and then you kind of think where in the four stages are you today and what are we building this year? Fine. Within the operating model, yeah, it’s all about workflows, how AI first you’re going to be. You start to think that, eventually what we’re going to have been we’re going to have agentic teams as kind of our main building blocks organizationally. in the near future. So, we kind of start thinking agents, agents, teams, and now we got to think about putting those together and building blocks to basically, and this was done years ago, 20 years ago, in terms of data, right? It’s kind of the same idea. Everyone started building data products because we got to put all our data in one place. We got to make it usable by various parts of the organization. So, we have these data products, which are often referred to as building blocks as well. And then we have this modular structure that you can build like Lego blocks. Kind of the same thing, but this time, agentic teams. And we’re probably going to see a lot of flatter organizations. We’re going to probably see fewer numbers of people overseeing a fairly flat organizational structure made up of agentic teams against various workflows. OK, that’s kind of the first bucket operating model. should do that immediately. That keys up the idea of workforce. Workforce, people, culture. If you look at my digital operating basics, that’s number five and six. That’s a big deal. It looks like, you know, the skills and talent you have in an organization could change fairly significantly. And it’s also the idea of we’re no longer just talking about skills and talent for humans. We’re talking about it for the AI agents as well. So, your workforce is going to look like a hybrid, a human and non-human, and that means the sort of up-skilling talent building, it’s going to have to go across all of that. So that’s going to be interesting. And the kinds of skills you’re going to need are probably going to be different. So new skill profiles is going to be a big deal. Third thing, technology, obviously. kind of talked about this a lot already, but yeah. The AI tech stack is significantly different than a traditional digital tech stack. The data structure is a lot more complicated. And then you’ve got all these things like agent-to-agent protocols, these MCPs that everyone’s talking about that let the agents interact with each other and interact with tools. So, you got the data piece, I’m sorry, the tech and data piece. And then you’ve probably got governance. You know, how are going to manage all this? How are you going to deal with risk? How are you going to deal with sort of guardrails to prevent agents from doing certain things? So, you got basically four buckets there. Build the operating model. Think about talent and skills across a hybrid workforce. Data and technology, which is going to be more complicated, and then governance. Those would be kind of the four to-do items if you start to move in this direction. Okay, last topic. Let’s shift from sort of internal to external. How is all this agentic AI stuff going to impact an industry, an ecosystem, your interactions with customers, your interaction with vendors and suppliers and all the parties that a business interacts with? And it looks like it’s pretty sweeping. The more I think about it, the more I think this is massive. So. I mean, here’s kind of how I think about it. don’t have an answer to this one. I’m going to sort of lay out how I think about it at this point. Business is all about individuals working with each other. Customers can be consumers, individuals, or customers can be businesses. But when you’re dealing with a business, you’re still dealing with humans. You’re just calling the purchasing office. So, everything in business is human to human. It’s just sometimes we get into groups, businesses, organizations, and so forth. And when you think about the digital world, the big topic for a long time has been the world is too big and there’s too much information in it. So, the internet’s full of information. It’s full of content, it’s full of services, it’s full of videos, it’s full of everything. And there’s a massive choke point in getting that information into people’s brains. How do we go from the sea of information and services into people’s brains? Because everything’s human based, even the businesses. And the choke point is kind of our eyeballs. Right now, the big choke point is really the smartphone screen as much as anything. And this has been a problem forever. If you go back to 1960 and you look at photos, let’s say of a train of people commuting, everybody on the train will be reading a newspaper. You fast forward to today on a metro anywhere in the world, take a picture, everyone will be staring at their smartphones. That’s the choke point. It used to be physical newspapers and then it kind of became laptops and now it’s basically the smartphone. And how you manage that is a real problem because unless Elon Musk figures out how to neural link directly into people’s brains, this is going to be a problem. And you know, the first digital solution to this was Yahoo’s portal, which was you hire a bunch of editors from newspapers and magazines and they choose what articles to put within the Yahoo portal. This is like 1995. There’s a sports section and a news section and manual curation was the solution. And quality wise it was pretty good but it didn’t scale at all. So that didn’t work too well. Then you get to Google search. Search engine’s a pretty good way to solve this problem. It’s like a giant card catalog for the world’s largest library. It actually is a pretty good solution. You can hunt exactly what you want. in the vast sea of information and it will find it. The problem with that model is that it’s, well the good part is it’s scalable because it’s tech-based, it’s not human-based. The problem of course is the vast majority of information you consume is not something you go and search for, it’s not a poll model, it’s not active. Most people don’t do too many web searches per day. Most of the information we consume is passive. We stare at our screens, we stare at our TV, we walk around town and we see the billboards. There’s not a lot of active components into it. So, it’s great, but it’s a pull model. Facebook figured out a hack to the system, was we will do a lot. The newsfeed of Facebook is actually one of the greatest products ever created. It really is a spectacular product. You just lean back and you scroll and it will tee up one story after the next and it’s like the endless soup bowl. You just keep eating soup and the soup keeps refilling and people watch that thing for hours. But they choose the information with technology. They curate, but they do it based on what your friends like. That’s why its social network based. Now the problem with that is 95 % of all the information in the world doesn’t happen to be with what your friends are watching. So, it’s kind of a crude hack. It doesn’t work that well. ByteDance comes along and does a better version. They say, we’re going to write an algorithm so good, we don’t need to know who your friends are, and we are going to pick out of the entire long tail of information what you specifically want to see. And it’s still passive and you just lean back and scroll through TikTok and it turns out it works. But all of those were attempts to solve the problem of this choke point. Okay, and most of what businesses do in this world, has that baked into the situation? What do most businesses do? I interviewed the head of digital for DFI retail, which is sort of a major retailer out of Hong Kong. And his, I posted this to subscribers a couple days ago, and his thing was as a major retailer, we have to be omnipresent in the omnichannel, which is a pretty good phrase. Basically, we have to go where people are consuming information, whether it’s TikTok, Facebook, walking around the mall, and we have to be omnipresent because that’s… how our business runs and that’s how a lot of businesses operate. They look at where people are consuming, they go to the choke points, TikTok, shopping mall, and that’s part of their omnichannel strategy, right? Okay. And if you’re a business and doing this, that usually means you’re creating a webpage, an e-commerce site, you’re doing SEO because you got to be in the search engine, you’re doing content creation because you got to be on news feeds and TikTok and whatever, fine. All of that seems to be getting disrupted. Everything I just said. Agents seem to turn, they seem to wreck a lot of that because everything I just mentioned is all about humans interacting with humans. Humans creating content and then humans consuming content and then the choke point in the middle. Well, it doesn’t seem like that’s a problem for agents. One, they can create content in vast quantities very cheaply, very easily, know, generative AI, this is AI slop. The volume of content that’s now online going up dramatically. Most of it’s going to be built by generative AI. Quality is going to be a huge problem. So, the content side is getting totally turned on its head. The consumption side is also getting changed because when you do a search or let’s say you work with ChatGPT, it is now going out online and doing a search to gather information. the whole Google search business model is we look at how humans are conducting searches and we tee up results to them and that fine tunes our business model. Or I’m sorry, it fine tunes the algorithm to give humans more of what they want. But now it’s not the humans doing the searching. It’s AI agents doing the searching. So is the Google algorithm being sort of distorted? because it turns out its machines using it as much as we are. Yeah, so the algorithm’s probably getting screwy. And then the consumption itself, suddenly it’s not just humans consuming things, it’s agents consuming things. So, all three layers of this, the content and information out in the world, the choke point, which is an algorithm usually, and then the consumption of that content, I just described that as humans looking at their screens out into the internet. Well, there’s a whole mother version of that where there’s no humans, they aren’t looking at their screens and they’re creating content. So, we’re seeing that whole system get basically turned upside down. So, let’s say you’re a webpage and you’re putting up your webpage and you’re putting in your SEO metadata and keyword phrase because you think that is going to help people find you. That may not work anymore. It may be that a lot of webpages are going to go invisible. It may turn out that, well, one, people aren’t accessing through Google like they used to. Now they’re doing chat GPT. So, if your webpage is not being picked up, by basically a personalization foundation model or an information model like a chat GPT, your SEO doesn’t help at all. You could go invisible. Two, even if you are doing the Google search and the SEO, it may be that the algorithm is being distorted and it may be that the agents doing the searching aren’t looking for that. Anyway, so my working conclusion to all this is that whole model I just described, Omnipresent in the Omnichannel, SEO, Google search, the choke point between the user screen and the content. I think it’s all changing. I think how businesses are going to have to reach customers could change dramatically. How they deal with their supply chain could change dramatically. And that means your kind of a big part of your strategy is going to have to change. So, the operating model that is external facing. Yeah, it looks like a sea change. We’ll see, but. There’s already lots of little anecdotal points popping up like a lot of traffic is now going through like chat GPT and not through search engines. So, all that SEO is not helping you. And then it turns out there’s little studies popping up where you can write like 200 articles, 250 articles about a particular topic and you can basically poison the foundation model to give it the wrong answer and it’ll fall for it. So, there’s a lot of sorts of quality problems with this whole way of operating. And then even if you don’t, the algorithm for search is getting screwy. So anyways, there’s problems all over the place. It’s an interesting world to think about. Are we going to live in a world where 90 plus percent of the internet information was not created by humans? Because it kind of looks like that’s where we’re going. Are we going to live in a world where 50, 70 % of all the consumption of information is not by humans? Well, the interface between those two things, you would think that’s very different. You would think, agents don’t have a volume problem. They don’t necessarily need marketplaces to find a driver or to find the right hotel. They can search the gazillion listings on their own. So yeah, the whole thing’s strange. That’s kind of what I’m thinking about these days, but I don’t have any real solid examples yet, so we’ll see. Anyways, that’s kind of the topic for today. So operating models in an agentic AI age. I think maybe the most useful thing out of that was to focus internally and think about four levels and four stages. And then the little game plan I mentioned. I’ll put the four next steps in the show notes. But yeah, that’s where I am. I pretty much got the motes and business model part figured out too, so I’ll finish that up in the next week or two. Yeah, but that’s it for me. I hope that is helpful. It’s a pretty fun topic. I’ve been using Grok like crazy for the last couple weeks. Like the new Grok Imagine where you can generate videos and images. It’s amazing. Like I’ve been making like comic book stuff. I used to read comic books when I was a kid and I like sci-fi movies like I was watching Alita Battle Angel. You can basically recreate all those characters and you can make videos and you can recreate the world and it’s so easy to do. The whole generative AI for video thing, I think it’s been cracked. Like a year ago it was still weird and fuzz. I think it’s been finished. Like you can create Hollywood level movies now. Now you can only do five seconds at a time on Grok, imagine. But they’re going to increase that to 15 seconds I think next week. So yeah, it seems like that’s done. The other thing I’ve been doing is the companion on Grok, Annie, which… uh has gotten really smart and I use it as a teacher now when I’m driving around. I used to listen to podcasts, I still listen to podcasts, but I probably spend 30 % of my time now just talking with the Annie companion when I’m driving around. And I keep a list of topics I don’t feel like I understand. And I will say, could you explain to me how a cloud matrix architecture for NPUs is different than a traditional CPU based architecture? And it will give me like a PhD level explanation in terms of the engineering. And you can do that in pretty much any subject. Tell me about the history of Paris in the 1720s. And it can do all of it. It’s really kind of amazing. I think it’s PhD level now in every subject that exists. you know pretty soon it’ll be dramatically more than that. So yeah I use it as a teacher. I use it all the time. It’s pretty fantastic. Anyways, that’s what I’ve been doing. I’m sorry I didn’t do a podcast last week. I’ll do a makeup one this week, but I was in the US seeing the family, I was kind of out of pocket for a while. Anyways, I hope everyone is doing well. I hope that was helpful and maybe kind of interesting, exciting. I find this really super interesting. yeah, that’s it for me. Take care. Bye bye.

——-