This week’s podcast is about the emergence of agentic commerce. This is a big change to traditional omnichannel.
Plus, there is a bit more of a rant about the recent Anthropic and OpenAI ban. My prediction is the future of AI outside of the USA is now:
- Open source, downloadable and non-embargoable models plus
- Low cost, tech agnostic AI architecture
Which probably means China. And probably Nvidia.
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.
The 5 steps to get into agentic commerce are:
- Start building a council of LLMs.
- This means a couple of frontier models and lots of open source (a symphony of models). The net is a composable model with a router than directs all queries.
- The key AI and agents need to be able to code. That is their superpower.
- Get to the tipping point where an end-to-end workflow has agents and AI, with no humans in the loop. Then you can scale up to 100-200 agents.
- This means fixing the quality problems before you scale.
- Get to the tipping point where agents create agents (an agent factory). This gest you to thousands of agents.
- This includes onboarding, training and oversight of agents.
- Centralize enterprise data and context for AI and agents.
- You need updated centralized data that your enterprise AI can access.
- A lot of data is born somewhere else. Salesforce. System of record. ERP systems.
———
Related articles:
- More High-Tech Flex by Huawei R&D (4 of 4) (Tech Strategy)
- 6 Big Events in AI Agentic Ecommerce (Tech Strategy)
From the Concept Library, concepts for this article are:
- GenAI Agentic Ecommerce ABCs
From the Company Library, companies for this article are:
- n/a
——–transcript
Episode 287 – Agentic Ecommerce
Welcome, welcome everybody. My name is Jeff Towson and this is the Tech Strategy Podcast from Techmoat Consulting. And the topic for today, pivot, shift, expand, move from omnichannel to agentic commerce like right now. It’s like my number one priority in life business wise is shifting from omnichannel to agentic within the e-commerce world.
00:33
as soon as humanly possible. It’s just to see change. It’s tremendously important. It’s tremendously powerful. lot of disruption, a lot of risk. Parts of it are a major, major threat to merchants, brands, e-commerce companies. Yeah, this is one of those ones where I would hit the panic button. And I almost never hit the panic button. This would be a panic button for me. So that’s what I’m going to talk about today, sort how I’m thinking about it. There’s a lot of action happening right now, the last month. So I’ll sort of…
01:04
go through some news, stuff like that, but this will be a shorter talk. I’m literally just back from a trip, so I’m restarting here. So that’ll be kind of the approach for today. A little bit more casual. Let’s see, standard stuff, standard disclaimer, nothing in this podcast. My writing website is Investment Advice. The numbers and information from me in AES may be incorrect. 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.
01:32
And with that, let’s get into the topic. Now I’ve been off for a couple weeks. Sorry about that. I’ve been kind of traveling all the time. It’s really pretty fantastic, actually. Great trips. And also, kind of when things change, I kind of need a little time to hide out and do it. I mean, I’ve done a massive amount of reading over the last two weeks because things are definitely changing in a major way. And I needed to sort of deep dive into a lot of it, which if you’re on the plane for
01:57
35 hours to get to Brazil and 35 hours back. That’s a good time to do a lot of reading and podcasts actually. So I’ve been sort of studying like crazy. I feel real good about it. So that’s good. Yeah, I went down to Brazil, which was like kind of a nightmare flight from Asia. Usually it’s not that bad. You fly through Europe and then you sort of cut down from Germany, France. You can go right to Brazil. But because of all the sort of fuel problems and all that, I went through the US, which was just a nightmare.
02:27
mean, it was a long flight to the US. It was a long flight down to Brazil. I got to hang out in Sao Paulo and Rio, which I really like. I was there for uh several days of sort of strategy meetings with the senior management of Mercado Libre, which is public. I’m permitted to talk about that. can’t talk about what we talked about, but uh permission to mention that that’s what the meeting was about. So that was fantastic.
02:52
Super interesting company. I’ve been following for a long long time. So to sit down with the senior manager I’ve met them before on occasion, know sort of one-offs coffee’s come but not like the full strategy process So that was fantastic. I’m not sure how much I can talk about the content I’ll see it maybe at some point there’ll be a there’ll be some public aspect of that but for now there’s not anyways, that was awesome I also met with a company called Natura, which some of you may know, you know, the two big beauty companies of Brazil are uh
03:23
Both DeCario and Natura, which I’m saying wrong in English, of course, but great companies. They’re sort of like beauty, skincare, cosmetics. Like Natura has sort of a rainforest aspect to natural sustainability, all that really cool company. So it’s meeting with them. That was a lot of fun. A lot of interesting stuff happening in Latin America, e-commerce. And anyways, I can’t talk about the subject per se, but you can kind of put it together. Like why would someone who does Asia, China, e-commerce be in Latin America?
03:53
Well, because the Asia, China in particular, e-commerce companies, they are going international and Latin America is like their number two favorite place after Southeast Asia. So you can see TikTok shop is there, Shopee is there, Temu is there, Shein is there. I mean, that’s a really important subject, which is, and this is actually kind of a big deal. Everybody knows that Chinese manufacturers go global. Refrigerators, air conditioners, this has been going on for 20 years.
04:22
And they have tremendous advantages when they do this. Well, that has kind of expanded into the digital world where a lot of these companies that are going international, Southeast Asia, several of them are in the Middle East. Europe is a major target now, not the US for the most part. Well, they have tremendous e-commerce operations back in China that they use, and it gives them real advantages. So you can see these companies like TikTok Shop, Kwai, which is Kuaishou.
04:51
Xi and Ti Mu, Tao Bao is now going international in Southeast Asia, AliExpress is there. They’re going international. JD is now in Europe with JoyBuy. So you see this pattern and they are serious competitors in these regions. It’s kind of a big deal. Anyways, that was kind of the topic overall. A lot of fun. Brazil has been one of my favorite places forever. Like I always take any reason to go to Brazil. And I always have the same experience.
05:19
Like I have a lot of fun in Rio, go out on the beach. And then the security aspects get to me a little bit, some of the, you got to be careful. And sure enough, I had sort of a security issue literally day one in a Starbucks. I had a guy come at me. We ended up having like multiple, three Starbucks employees protecting me. I’m going behind the counter. This guy’s coming at me. I’m just sitting there reading, you know, drinking my coffee. So yeah.
05:47
That’s kind of, I find as I get older, I may be less tolerant of those sorts of aspects of life than when I was younger. anyways, but overall phenomenal trip. Okay, that was just kind of what I’ve been doing. I’m going to catch up on all the articles and podcasts in the next week, but pretty fantastic. All let me get into some content here. Now, a couple of weeks ago, I did kind of a rant, which was just basically talking about, this was June 12th.
06:16
The US government made their announcement where they basically told, you know, anthropic, look, you can’t release Fable 5. It’s too powerful. You know, we can jailbreak, you know, it’s able to hack into things. You got to pull it. But they didn’t say you got to pull it. They got to say, you know, you got to pull it for non-US citizens, which was really kind of weird because one, that there’s a lot of implications to that.
06:45
It teaches basically everyone that you can’t be, especially if you’re not American, you cannot rely on these businesses because they can get pulled from you any day of the week. So it’s a little bit like what happened with China and the entity list back in 2019. When the US applied the entity list to Huawei, it basically cut them off from US tech. It taught every company in China, you can absolutely not depend on this for your business. So therefore you need a separate supply chain for your key technologies.
07:15
Well, this was kind of the entity list moment for every non-American. And even more beyond that, because most companies in Silicon Valley, including like a ton of anthropic employees, well, they’re not Americans either. So technically, they can’t use their own product. Anyways, I kind of thought of it as like the entity list for everybody moment. And the natural result of that was everyone was going to start thinking we need to start doing our own open source downloadable models because we cannot be
07:42
exposed to this sort of haphazard political risk. That was a bit of a rant. It wasn’t terribly well thought through. And I kind of said, the next shoe to drop will probably be, then the US government will ban Chinese open source models. Because if everyone’s going to open source AI, downloadable, well, most of those are Chinese, although Nvidia is becoming kind of a major player now. That was kind of my rant. It was more of a guess.
08:12
Prediction not that well thought through to tell you the truth. Okay, June 25th. The US government does it again open AI releases GBT 5.6 Super powerful frontier model the government basically tells them to stop it and they got to pull it back Kind of the same thing and now everyone’s starting to think okay The first one might have been a fluke but is this just going to be the way it is that if you want frontier level intelligence
08:42
Is this becoming a competitive advantage where certain companies are going to have preferential access to frontier level intelligence and other companies are not? And is it going to be determined that preferential access, which could be a major advantage, is that going to be determined in a rational, predictable way or is just crony capitalism top to bottom? And if you’ve got, know, and it looks like it’s just crony capitalism.
09:11
If you know the right people, you get on the list that’s approved. And they were literally saying to the government, we’re going to have to approve every single company. Well, you know how that game is going to play out. everyone sorts of rants on this, goes on sort of a heightened version of the June 12th situation. Then a couple other things happened, which really the last week, which I think even took this further, which was if you haven’t seen this,
09:38
Listen to the rant by Alex Karp, the CEO of Palantir. He did a rant on CNBC that went everywhere and it was very powerful. He went further. Everything I’ve just sort of said sort of sits in the bucket of political, politically shaped markets that are giving preferential access to technology. The natural response to that is every company has to build alternatives, which probably means open source that you have on your own servers.
10:08
because you can’t risk getting cut off. He took that a lot further. He basically said, look, it’s not even just, I’m paraphrasing, it’s not even just the political uncertainty of this core technology. You are basically becoming dependent on these frontier level technologies, these companies, Anthropic, OpenAI, and they’re going to charge you for tokens until the end of time to run your business. So they’re going to win there.
10:38
It’s not like buying software which you then own and then you maybe pay per seat or you pay per license or per year. No, you’re going to pay per token, per usage forever. In return for that privilege of paying them, you are going to give them access to all of your IP, all of your business know-how, all your work processes. The way he described it is you are basically going to give them all of your alpha. You are going to commodity… Now, what are they going to do with that?
11:08
Well, we already know what they’re going to do with it. They’re going to launch additional services based on your IP, all your business know-how, and they are going to commoditize everything you do and sell it as a service to others.
11:22
That’s like suicidal as a business. And we’ve already seen this play out. You can go and look at laboratories or law firms that are using all these tools, AI for everything. They’re uploading their processes, their contracts. And then these companies turn around and they offer something like OpenAI Legal Services. I don’t think they’ve opened that one. But they’ve opened several like this, where as soon as they get enough of your data and know-how,
11:50
They can turn around and offer it as a service and then everybody can do what you do. So the idea that you’re going to give away, I like the way he described it, the idea that you’re going to give away all of your trade and operational and business know-how, all of your alpha, you’re just going to give it to someone and then they can do whatever they want with it. That’s suicidal and they can sell it. So that really took this argument to the next level, which is like,
12:17
You have got to download open source models and build all these models internally. You know the old bitcoin saying like if it’s not your key it’s not your coin? Like if you don’t own your bitcoin key, if you’ve got it on you know some website hosting it, they’re your wallet. If it’s not your own wallet it’s not your coin. Look if it’s not your model it is not your business. Someone else owns your business at that point if you’re running everything on some model.
12:46
So there’s this idea that everything has to be open sourced. And I think that really the argument has shifted in the last couple of days. Everyone’s kind of saying the same. Now there’s, I’ve been sort of saying for a while, when we look at traditional software and the platforms that you build with traditional software, marketplaces, things like that, the two big superpowers have always been network effects and economies of scale. That’s Facebook, that’s Alibaba, that’s all of them.
13:14
When you move to generative AI into agents, the superpower is economies of scope. Where if I’m offering an LLM to lots of people that’s very good at doing legal expertise and all the lawyers are using it and uploading their contracts to me, I’m getting smarter and smarter. And then I can start to offer, let’s say, accounting as a service. Well, they’re uploading. It’s very easy for me to step from legal services to accounting services.
13:44
to corporate registration services, to some degree to consulting services. But once you get that feedback loop, so I would argue the superpower so far of generative AI is economies of scope with a learning feedback loop that lets you move horizontally across different services and businesses very easily. I think so far that’s the superpower. You could call it a competitive advantage. I’m not sure that it is. And I think that’s what OpenAI
14:13
I think that’s what they’re doing. And I think that’s what Alex Karp was talking about. Anyway, I think you put all that together and these open source models just got like the boost of a lifetime. This is like the mother of all boosts for them. Everybody has kind of accepted, well, most everyone I do has accepted, you cannot build your businesses on these frontier levels, subscription model, pay for token businesses. You can’t do it. You have to build your own model.
14:43
Now, how do you do that? Well, I’m going to talk about that. That’s the point of today. How do you go from omnichannel to an agentic commerce? Really, it’s AI plus agentic, but I say agentic. How do you do that in a smart way? Where, look, we have to use these services, Anthropic and so on. But we want to use them for very select things and definitely not for other things. So I’ll give you sort of my thinking on how I think you should do that. But yeah, I think the argument has sort of been concluded.
15:11
And I think these companies are going to take off like crazy. And when you talk about open source, it’s mostly Chinese companies at this point. Nvidia is emerging as a major player in open source. There’s a couple others. And so then the next shoe, which could drop, which I think will drop, is you’ll start to see the US government try to blacklist these Chinese open source models. And I don’t think that’s going to work outside the US. don’t. I think the whole rest of the world is doing this now. But we’ll see.
15:41
Anyways, so a little bit more of a rant. All right, let me get into the topic here more, but kind of important, I think. Okay, so what is agentic commerce? I’ve written about this a bit. I know it’s like the highest priority at Alibaba. I know a lot of these major e-commerce companies, they’re viewing this as a complete change in how commerce works. And as I’ve kind of said before, like, okay, the world is based on eight billion people. We all do a certain number of payments. We all use our laptops and phones a certain amount.
16:10
We use mobile networks a certain amount. Well, it’s going to go from 8 billion humans to 8 billion humans plus 800 billion agents that never sleep, that work 24-7, that make payments by the millions every second, that do search inquiries by the millions every second. They’re going to do a level of volume that dwarfs anything humans do.
16:35
and they’re all coming online this year, but let’s assume it’s going to be 800 billion. In addition, you could say 20 to 30 billion robots. Now robots are basically physical AI, so embodied intelligence. So that’s kind of how I’m thinking about the future. Like in the future, an operating model for a business, it used to be people plus software plus hardware plus physical assets, buildings, trucks. I think the future operating model is going to be Haro.
17:04
humans, agents, and robots. That’s what you’re going to build a business with. And some businesses, it may be one human and thousands of agents. Other businesses, it may be tens of thousands of robots. Maybe some of them are intelligent. Maybe some of them aren’t. It’s going to be a mixture, depending what you’re thinking about. I like these sort of 24-7 coffee kiosks that they put up with no humans.
17:30
And you just walk up and you get your cup of coffee and it’s got an avatar on the screen and you buy your coffee. And every couple hours someone comes by and fills up all the supplies in the back. Okay. That’s kind of a robotic first, then agent, then human third operating model. So I’m kind of looking into the future for that. Now for e-commerce the difference is these agents
17:59
They’re not just going to be operating aspects. They’re going to be buyers and sellers in their own right. So that’s going to make it very different. They’re going to be kind of quasi-humans. And so almost every week, I find something in the news that I sort of didn’t realize that would fall under the category of agentic commerce. So here’s a couple I noticed in the last couple of weeks when I was reading. Agentic search using search engines.
18:27
absolutely dwarfs human search. They do far more search inquiries than we do, and it’s going up like a rocket ship. And you see these new search engines being built that aren’t built for humans to use at all. know, agents don’t need anything like we would use a search engine. All they really need is sort of uh API and a command line interface. That’s most how they sort of they engage with the world. There’s no UI.
18:57
There’s no user experience. There’s no user interface. It’s usually a CLI or an API. So agentic search, entirely new animal, the volume is crazy. That tees up the idea of AX. Instead of UI and UX, user interface, user experience, you can start to talk about AX, agent experience, how an agent would sort of interface.
19:25
with the service of various types, it’s not going to be like humans at all. So search is going to balloon, agent interfaces are going to be a new thing. When you get into payment, payment becomes a big problem because agentic commerce, you’re talking about millions of transactions, okay, maybe not every second, every minute. Well, none of the payment rails that we use today are built to handle that typically. That’s a volume that’s new.
19:55
According to groups like Coinbase, that’s what they say. So groups like Coinbase, what they talk about is, look, we’re going to have to go to stable coins or something to process this level of volume and to have a record that we can verify. So it might be we need entirely new payment rails, probably based on stable coin, that can handle this sort of massive volume of agent payments. OK, that’s interesting.
20:23
We’re seeing the revival of PCs. If you watch the Nvidia conference not too long ago, you’ll notice that all these companies are starting to talk about laptops and PCs again. Nobody’s talked about laptops for like 10 years. Like, we don’t think about, there’s a new laptop coming out and it has a better hard drive. We don’t really cover this. Well, I don’t. I don’t know anyone who covers these things, right? But now you’re talking about, look, you can’t run agents on a smartphone.
20:50
you’re going to have to run them on a laptop or some more likely a server or a small PC at home. And sure enough, people are using Apple. And now Nvidia has got their DGX Spark, which is basically a desktop workstation that they’re sort of marketing to developers and researchers who want to prototype AI models as opposed to a typical consumer laptop which you might use for productivity or gaming.
21:18
So laptops are kind of getting this interesting revival. I’m probably going to shift over to that soon. I’ve been running things cloud-based. I’m probably going to shift to some sort of home-based laptop. I don’t think I would get the NVIDIA, but probably an Apple or something like that. Yeah, got to say, I as you start building your own models and your own agents and things like that, you got to run them locally. So that’s kind of an interesting thing. And then that brings up the next idea, which is like, how do you run agents as devices?
21:49
Are we all going to be using these smart glasses that connect with the cloud where some of our agents run or they connect through, let’s say, WeChat or WhatsApp and that’s how I command my server at home to do what I want it to do? That’s kind of what people are doing. That’s what Tencent is doing. You can basically use WeChat to connect with Tencent Cloud where you can have your agent, work buddy, code buddy, or you can have it running home on your PC and you can command it that way.
22:17
So this whole idea of agent devices is getting kind of interesting. A couple last ones I heard, scammers are getting pretty good. Like you get these fake emails all the time, hey, you have a problem with your account. And you kind of know the email’s fake. You must log in here, you have a problem with your client, and you kind of know it’s a fake email, it doesn’t look quite right. It is really hard to tell the fake emails now.
22:45
The fake emails by scammers using generative AI are really, really good. And then if you click over, which is foolish, but if you do that, they’re building perfect copies of websites now. You really can’t tell. So the whole scammer fraud stuff is, you know, they’re getting superpowers. And one of the areas you see this happening right now, like full force, is on YouTube, where
23:13
People are creating all sorts of content using AI that is just violating rules and violating any sort of IP rights whatsoever. If you go onto YouTube right now and you type in Bieber and Rosé, you know this black and pink girl Rosé and Justin Bieber, there’s probably 20 to 30 music videos online of Justin Bieber and Rosé doing like songs together.
23:40
None of them are real as far as I can tell, but they look like videos. They even say like Vivo in the corner or in the headline of the video it’ll say official music video. okay, most of the images are sort of taken from other videos and packaged together, but the songs in their voices, very hard to tell they’re not real. I had to watch a couple of them and I’m still not sure which I’m, can, some of them I’m saying, okay, that one’s not real. Some of them I’m not sure.
24:10
The only way I can tell is by looking at the view count. If they have 200,000 views, okay, it’s not a Justin Bieber video. But go and search right now and you’ll see 20, 30, 40 of these music videos that aren’t real. You will see people generating their own version of honestly people like me, where they will create videos, they will just take my content, they’ll run it through a generative AI thing, they’ll create their own version of the whole content.
24:39
swap me out of the video, put someone else in, they may even keep my same background and they will just put it up as a new video and get tons of views. You see that stuff happening all over the place. So, you know, these AI tools, apart from just stealing money, that’s problem, like these phishing emails, but just the content world is just all the rules are gone, all the IP protections are being blown away.
25:05
Now some of it I actually like. Like I really used to like Star Wars before it got destroyed. And they have all these books of Star Wars. I’ve read all of them. There’s like a hundred of them. I’ve probably read all the hundred books like twice, three times over the course of my life. But all the Star Wars movies and TV shows are garbage now so I’ve stopped watching. I don’t even see them anymore. YouTubers are retaking the books and they’re creating generative AI videos from those books. Little storylines. 10 minute, 15, 20 minute videos.
25:34
They’re fantastic. Go look up Star Wars Untold as a YouTube site and they will create videos of like Hayden Christensen as Darth Vader right after the end of this movie. I mean the stories aren’t great but it looks visually fantastic. And there’s sort of this explosion of fan films and creativity that’s almost reviving a dead brand. Now I think the brand is still dead but
26:04
If there’s any life being breathed back into this, like the key will be to tell Disney, just stop making everything because everything you make is terrible and it destroys the brand. Let the fans do it because they actually care. And honestly, the quality is much better. So it’s really interesting to watch all these fan films percolating up. And you can actually get a sense for who people like. Like if you watch the Star Wars fan films.
26:32
They all have Hayden Christensen as Darth Vader, young Darth Vader. And if you watch Superman, they all use Henry Cavill. But none of the fan films use the new guy, David Cornswet, that they did in these new Superman movies. You get a real sense for who fans like because the people they don’t like, nobody uses them in the fan films. It’s a really interesting sort of uh community aspect. Anyways, okay, that’s just sort of news from the last couple, three weeks in this whole agentic world.
27:02
Let me give you the so what, because I’m kind of just rambling today. Let me give this what of what I’m focused on. Like if you’re doing omni-channel, which most merchants and brands are, how do you start moving toward agentic? Okay, here’s what I would do. Okay, for me the biggest thing happening is what I called ABC, right? The agents are coming in, they’re either becoming assistants for the consumer, that’s the A, they’re becoming brokers that sit between.
27:30
buyers and sellers, or they’re becoming concierges, which represent the seller, the merchant. So, you have A agents, B agents, and C agents, assistants, brokers, concierges. That’s reshaping how transactions are done. It’s reshaping how you do discovery. It’s reshaping how you do execution, how you do search, all of that. It’s a huge threat. Within that, your core, your primary objective is the fight for eyeballs.
27:59
Now I would have said before, the fight for attention, the fight for engagement. Well, a lot of attention and lot of engagement is going to be getting the engagement of agents, not humans. We might want the transaction to be done by an agent, but the starting point for all of this is you need human eyeballs on screen. If you don’t have human eyeballs, you cannot tell your brand story.
28:25
You cannot tell them why you’re special. You cannot appeal to their psychology. You cannot appeal to the emotional aspects. You cannot get back and forth data that lets you be smarter about the human at the end of this process. You can’t do personalization. You can’t do cross-selling. You can’t do upselling. You can’t do loyalty program. You can’t do anything if you don’t have human eyeballs. So under an ABC world, the fight for eyeballs is the key.
28:54
Even if you’re doing most of your business transaction wise, selling to agents that represent humans, you can’t do that if you’re not getting direct information from the humans itself. You don’t even know what to sell. You don’t know what marketing works and what marketing doesn’t. don’t know. So you’ve got to have some percentage of your business, whether it’s 10%, 50%, 80%. Hopefully it’s most of it. But even if, even if most of your business is selling to agents, you got to have 10 or 20 % where you’re getting eyeballs.
29:24
Otherwise you’re blind. So the fight for eyeballs is the start. So what do? You got to get used to the fact that for a certain amount of commerce there is going to be a significant change in the user entry point. That you know if we look at any product ketchup, sneakers, fashion, beauty
29:51
I will always ask myself two questions. I kind of do these in my checklist. What does the consumer want or need? And I try and be very crystal clear about the language. They enjoy the shopping experience. They’re looking for discounts. They feel good after they use this skin product. They need the skin product because they have sort chronic acne. I always try to crystallize what does the consumer want or need for B2C. My second question is always,
30:21
map out the consumer journey from beginning to end. And it doesn’t really end, it’s a loop. How did they first hear about you? How did they get interested? Did they watch videos? Did they click in when they went to the first site? Did they enjoy it? Was it all about convenience? Was it about the experience? When did the transaction happen? What about the I literally map out the process in great detail. And I sort of look at every stage and what they’re doing. Okay.
30:50
Now, the first most immediate problem for agentic commerce has been people have immediately started switching from search engines where they click on the search engine, go to your website or your app. Now they’re asking chat GPT. So, you you’ve been intermediated. Suddenly the first beginning point, the user entry point has shifted to a third party and search. You know, if you went to Google search as an intermediary, okay, they would charge you.
31:20
and you’d have to pay them a lot of money to get your clicks over to your website, chat GBT is not even clicking you over. They’re keeping the user relationship and they’re trying to do the whole process themselves, including the execution. So someone says, chat GBT, what sort of skincare lotion should I use for dry skin? Chat GBT answers, it makes three suggestions. The person says, I like number one. Chat GBT says, in the future.
31:48
Okay, would you like to buy this? We will handle it. And they’ll take care of it their agent will go and buy it on the website and give it. And they clearly have this intention. They want to be the everything super app. Okay, that’s a real problem for the user experience, the customer, the process, retention, all of that. So those are your two problems. You got to get the eyeballs and then you got to sort of not be totally intermediated.
32:18
So what would I do? Okay. I like this phrase, the council of LLMs. I think this was from the All In podcast. You got to start building a council of LLMs. Not one LLM, not anthropic, you not, no, you need to build on your own company, on your own servers. You want a combination of models. You want a couple frontier models. So open AI, I don’t…
32:47
Alibaba Quinn something like that and then you want several open source models and you want them to operate basically as a council and the net of that is you’re going to send an inquiry there either from an internal staff part of your workflow or from the customer and The council is going to route that to the right model for the right part of the journey It may well go to the frontier level to begin with
33:16
and the frontier level will serve as the primary routing agent to send aspects of the customer journey, the customer inquiry, the customer transaction, the employee workflow. It will route that to various sub-models, most of which will be on your own servers. So you have sort of a council of LLMs. You can call it a symphony of models. And I think that’s where… oh
33:42
Nvidia is probably going. I think they’re going to try and be a lot of that. I think the problem with Nvidia actually, if Nvidia is let’s call it the primary open source champion out of the West that’s going to sort of fight the Chinese ones, they have a bit of a channel conflict problem. They’re basically competing with their customers. know, OpenAI is buying a ton of their GPUs and they’re also becoming an open source alternative to chat GPT.
34:11
So maybe that’ll be a problem, maybe it won’t. Okay, so that’s kind of number one. Start building your council of LLMs. There needs to be a better term for that. And then within that council, can start to, you know, can use the frontier models, but you’re not moving into a scenario where you’re giving away your alpha from day one. Number two, your core agents all need to know how to code. Agents do certain things, right? They can…
34:39
They can do math badly. They can do words pretty well. They can write, things like that. But the superpower of any model is coding. The difference between a model that has coding ability and a model that doesn’t have coding ability is the one that can code can build what you tell it to build. The other one can only write documents. So you can say, you know, we want to build… uh
35:05
a website to handle customer inquiries for packages that are over 20 days later damaged. If you tell that to Claude Code, it can build the app and it can do all of it. So an LLM that doesn’t have ability to code, it’s like a regular human versus Superman. So Grok XAI, they just bought Cursor a couple of weeks ago and they’re basically referring to it as strategic infrastructure.
35:32
One, they can offer it to their clients, their customers, but also they’re using it internally. So your models have to have the ability to code. One, to create something for your customers or to create anything. I think that’s how coding models, I think about them now. I think they’re strategic infrastructure for any business. Okay, so that’s step number two. Step number three, I call this tipping point number one. You got to get to the point where you have a complete workflow where you’ve built AI and agents across it.
36:03
And you don’t need humans to basically be in the loop, which is actually quite difficult because you can’t scale until you get sort of an end-to-end workflow. It could be something like content creation for a podcast, making JPEGs that go out to sell your, know, to promote your podcast. It could be handling part of merchandising. It could be handling part of customer service. You’ve got to get an end-to-end workflow with humans out of the loop.
36:33
When you do that, that’s when everything starts to sort of accelerate and you go from a handful of agents who are usually working hand-in-hand with humans to something that can scale up independent of humans. So end-to-end workflow, completely agent generative AI. There’ll be some traditional software in there as well. Now, the trick to that is quality.
37:01
know, generative AI doesn’t do the same things the same way every time. So you can’t scale up from uh one agent to 1,000 if you have quality problems within the workflow, because things will fall apart and go badly very quickly. So when you’re sort of doing your first end-to-end workflows, you’ve got to get the quality right. And the way I kind of explain this to people is like, it’s really like hiring interns. You have to treat your agents within the workflow.
37:28
like interns you have just hired, you have to train them, you have to onboard them the same way you would have a human resources office onboard your employees. So literally that’s how you bring agents onboard. You onboard them, you give them basic remedial training. I’m sure people have already built this. There’s got to be an agent version of human resources already created. You bring them on, you train them, you start having them do the things, you check their progress, you do performance reviews.
37:57
You identify the problems, you fix them, and you basically treat them like employees until you get the quality end to end at the level you need it to be. Only then do you scale up. If you scale up before you do that, a couple things will happen. One, you could explode your cost structure, because agents use a lot of tokens. And two, you could have the qualities going to the toilet. Because you can’t have humans watching thousands of agents checking what they do. It has to be running well on its own at that point.
38:26
So that’s kind of tipping point number one. Tipping point number two.
38:32
We’ll call this item number four. At a certain point, you’re going to go from having a couple agents with human handlers and then no human handlers. It’s just agents. And that’ll get you up to 100, 200 agents. That was tipping point one. Tipping point two is where you start to build a factory that can generate agents on its own, where the agents start creating their own agents.
38:57
They go to the human resources, they get trained, they get onboarded, they get performance reviewed, they get checked. Suddenly you’ve got an agent factory. That’s how you go from a couple hundred to agents to 10,000. That’s the next big step up. I don’t really know too many companies that have done that. I’ve heard of a companies that have thousands of agents at this point. I haven’t seen it. But that’s the next bit. So tipping point one, one or two workflows. Get them running.
39:26
Try and get humans out of the loop. Keep them simple. Probably not interacting with the customer too much directly. Maybe internal workflows, so if they screw it up, it doesn’t make your customers mad. Stuff like that. Managing inventory, things like that. Back office is a good place to start. Then the next one. So that’s one, two, three, four to-do lists. Number five, is this the last one? Yeah, this is last one.
39:54
The big bottleneck will be data and context. So you need to start getting your data in context together. And what you don’t want to do is rebuild your whole data architecture. That’s a mistake. Don’t do it. You want to start adapting what you have. Now, most companies, what they have is they have their data warehouse, they have their data lake, and then they got a ton of their data sitting out siloed in various SaaS programs like Salesforce.
40:22
So you’re hiring Salesforce, you’re hiring these SaaS programs, a lot of your data is sitting in their silos. And depending who you’re dealing with, you may or may not be able to port that out into your centralized warehouse that you own. That’s a problem. So you’re going to have to get all of your enterprise data into one data architecture that you own under your own roof, all in one place.
40:52
And then your agents can run on that. Now before you had your big warehouse, your data lake, you’re dealing with SaaS vendors of various types. You’re mostly doing business intelligence stuff. You’re doing analytics. Maybe you’re doing a little machine learning. This is different. The amount of data that agents need, particularly context information, is huge. So you’ve got to start bringing all that in-house. And that creates an interesting problem with these SaaS vendors.
41:21
There’s this argument that agents are going to wipe out SaaS. I don’t really buy that. I don’t think that’s how the world works. I also think SaaS is not terribly expensive. It’s actually a pretty good bargain. There’s no reason not to. You’re not going to save a lot of money by killing off your SaaS spending. The problem SaaS gives you is they want to keep the data. And some SaaS vendors are very good at giving you all your data.
41:47
So you can port it into your centralized repository that your agents run on. And others are kind of at a little bit of a soft war with their clients, where they make it very difficult to get your data. So the problem is not necessarily the cost on the SaaS side. It’s how easily can you get your data. That’s a problem. Systems of record, ERP systems, Salesforce, SAP. I mean, some of these companies have reputations for being easy. Some are difficult.
42:17
But you got to bring it all together. Now what a lot of companies can do, which I think is what they probably are doing, is if you’re using a Salesforce or an SAP or something, I mean, they’re going to give you apps where humans use them to access your information, run your reports. Well, you can have your agents just use the same interface. So instead of a human using the API, it’s an agent using the API. And the vendors probably know that’s happening. But are they going to start creating walled gardens?
42:46
to keep their customers sort of away from that data. That’s a problem. So that’s kind of the five things I’d be thinking on. How do we get to agent commerce? Number one, you got to start building your council of LLMs. And that’s a good way to think about it. Number two, within that, you got to have agents that have coding ability, because that’s going to be the superpower. Number three, you want to get to the point where a handful of workflows end to end
43:14
have removed humans. That’s your first sort of tipping point when the last human removes, because then you can go from five to 100 to 200 agents. Then you got to get to the point where the agents start to onboard, train, and code new agents. You could call that an agent factory. And then the last one, your data architecture is becoming a critical bottleneck for almost everybody. You got to start bringing it all in-house in your central repository. How do you work with that with your vendors?
43:44
I don’t know. We’ll see. I think they’re going to have to bend the knee on this and start giving people their data. I think they can’t do that, but they’ll probably, smaller companies, they’ll probably try not to do that. Larger companies will just demand it. So we’ll see how that sort of internal data architecture plays out. In other those would be my sort of first five steps for going omnichannel to agentic commerce. As for me, it’s going pretty well, as I kind of mentioned. I’m happy to be back in Southeast Asia.
44:14
Probably going to be floating around for a couple weeks. The next thing I think is uh there’s the Shanghai World AI Conference in a couple weeks. Definitely going to that. I was going to sort of do a small tour with that. We kind of decided we didn’t have enough time to put that together, so we’re just going to go and hang out, meet with some companies. If you’re in town, let me know. I’ll be floating around for a couple days at the conference, and then I’ll probably be in Shanghai for a couple days just doing my normal meetings and stuff like that.
44:43
Yeah, we didn’t end up doing a tour for that. was too rushed. But yeah, I’m looking forward to that. I love this stuff. It’s kind of, I mean, AI coming out of China, it’s kind of become the epicenter of a lot of important stuff. A lot of geopolitics, which I don’t really care too much about, but definitely center of the action in terms of capabilities. And I would actually argue that when it’s AI meets e-commerce, I think China is the epicenter. I think this Alibaba in particular.
45:13
is probably the best in the world at this. I’m looking at Amazon. I don’t think they’re quite as good, but we’ll see. I mean, I don’t know what they’re doing behind the scenes, so who knows? Anyways, that’s going to be what’s coming up for me. I did find one fun app the other day, there’s an app called, or not an app, there’s a website called Talkie, and it’s basically a large language model. So like a chat GPT, you can ask questions to.
45:39
But it’s been trained on information up until 1931. Now, I think all Western, like Europe and the US. And it’s fascinating. It’s like talking to someone from 1931. And you can ask it all sorts of questions like, what was the big news in the last 10 years? And it’ll tell you all these interesting facts. But you can also ask questions like, how should I behave in business when my business partners behave? You can ask it like life questions.
46:08
and you’ll get answers like from a different age. To be honest, I agree with the answers more than what I see on like YouTube and, you know, OnlyFans and all this nonsense. I actually identify a lot more with the life and business advice I get from, I think it’s mostly Americans uh in the 1930s. It’s fascinating. So go take a look if you want talky, T-L-A-K.
46:36
Talk IE, so talkie IE. Or just type in LLM1931 and it’ll pop up in a search engine. But I thought that was, it’s really interesting. I’m not sure that’s not a really good strategy. If you want to differentiate in this world, either as a person or a business or as a content creator, maybe just tie yourself to a moment of the past. I’ve actually seen YouTube video, maybe it was TikTok. It’s a,
47:05
A TikTok travel vlogger, right? There’s millions of travel vloggers, but this travel vlogger goes back in time. So the vlog will be, it’s a young woman on the streets of London in 1650, just talking to people on the street. Oh, here’s a chimney sweep. And I think it’s historically accurate. So it’s like travel vlogging into the past. It’s fascinating. So anyways, I don’t know. Take a look at that.
47:33
Now on the flip side, that’s valuable, on the flip side of stupid content, I was watching America’s Sweethearts, which is the Netflix show on the Dallas Cowboys cheerleaders. I think I’m going to probably end up moving part-time back into Texas in a couple years, so I’ve been kind of looking around, seeing what’s there. I’ll probably be 50-50 in a couple years between Texas and Asia. So I’ve been kind of getting into that. Yeah, that’s a really stupid show, but it’s…
48:02
It’s all the women who are trying out to be Dallas Cowboys cheerleaders and all the drama and the arguments and yeah, it’s and then of course there’s women in their little outfits dancing around. So it’s probably doing real well in terms of viewership, but I’m not sure it’s really good for you. It’s probably not like in terms of content, that’s probably in the realm of eating potato chips. There’s two recommendations, America’s Sweethearts and Talkie from 1931.
48:30
Okay, that’s it for me. Sorry I’ve been gone for a couple weeks, but I’m back in content creation mode as of July. Hope everyone’s doing well. Talk to you soon. Bye bye.
———
I am a consultant & keynote speaker on how to increase digital growth and strengthen digital AI moats.
I am the founder of TechMoat Consulting, a consulting firm specialized in increasing digital growth and strengthening digital AI moats. Get in contact here.
I write (a lot) about digital growth and digital AI strategy (3 best selling books, +2.9M followers on LinkedIn). There is a free book and email newsletter below.
My Moats and Marathons book series is a 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.
