Welcome to the Pulse of AI. Welcome back to the Pulse of AI podcast. I’m your host, Jason Stoughton, here in the Bay Area, and on this podcast, I am joined in conversation with Lutz Finger. Lutz has over 20 years of experience building AI-driven products at companies like Google, LinkedIn, and Snap.
He is currently building a new company around generative AI tools for e-commerce and also teaches a popular eCornell certificate program called Designing and Building AI Solutions. I hope you enjoy the conversation. Let’s get to it. Welcome to the show.
Jason, good seeing you. Nice to have you on. I’m excited about it. You know, I love your podcast.
You know, I love some of the things you write on Forbes, and I think it’s going to be a really interesting conversation for my listeners. So tell my audience a little bit about yourself. Yes. Okay.
Something about myself. It’s a wild time, and I’m trying to make sure people understand the wild time. Those wild times. That’s actually all you need to know.
Background, I’m a quantum physicist. I have built several startups. I work. For Google Health, I was one of the first employees there to build up Google Health.
I helped Snapchat going public or Snap Inc. at that time. I built up a data science team at LinkedIn. I have been doing data and products and AI and products, like, since essentially at times that we didn’t call it AI.
We didn’t wonder about AGI. This was all like easy. Since that time, I’m an old guy and I’m a faculty at Cornell, where I teach. And I live in the Bay Area.
That’s me. That’s fantastic. And I also know you have a startup. We’ll talk about that a little bit later.
We can sort of weave it into our conversation. So here we are, right? Like you said, you help people think through sort of AI, right? And we’re at a very interesting time.
I mean, obviously, AI is in the news. If you sort of jump back, and you and I have been in this for a while, you jump back three years ago, four years ago, right before ChatDBT, and you asked most CEOs, what are you doing in AI? They’d say, well, I’m doing AI. What’s AI?
I don’t think we’re doing anything with it, even though they were across different parts of the organization. ChatDBT comes out. Every CEO says, we need to completely use AI for everything, replace every employee. It’s going to be me in this AI system, and we’re going to take over the world.
None of that’s really true, right? At this point. Right. And so then if you jump back a year ago, a year and a half ago, and you’re in Silicon Valley, which you are and I am, and Silicon Valley is sort of writ large.
Right? And sort of globally in the AI space, it was the year of agentic AI, right? Agents and AI. That’s now gotten out in the wild, if you will.
Right? And so you sort of hit Q4 of 2024, and you listen to most shareholder meetings, even with large tech companies, and it’s all about agentic AI. You know, everybody from Salesforce, Microsoft, et cetera, et cetera. Right?
And so now CEOs are saying, okay, what exactly is this agentic AI or agents in the workplace? And, you know, on one hand, and you and I were just talking about this before the show, you have this huge hype around AI. Right? But on the other hand, when people actually go and think about how they’re going to apply agents in the enterprise, it’s almost like they lose all imagination and they go for not just the lowest hanging fruit, but fruit that’s already on the ground.
Right? And so let’s talk about that. You know, what? Talk about what?
How do you see? How do you see agentic AI in the enterprise? You know, at a meta level? I think one of the key factors to understand is we have there been before.
History is a good guide, like as usual. Right? So as we had the data revolution, people were just like, oh, everything needs to be data. And like, I’m sitting in boards and then suddenly everybody says we need data on our slides.
And then suddenly we need mobile on our slides. I was at LinkedIn at that time. Were the mobile revolution. Kind of took us by storm and we essentially stopped the whole company just to move everything to mobile.
And it’s like very often people are lacking the imagination or they are risk aware because they have a certain business case. They have a certain value creation. And now the question is, how will that change? And yet again, with agentic AI, what is agentic AI?
Maybe let’s start there. What is agentic AI? Agentic AI. So we have general AI.
General AI. And what I mean by that is you go to Netflix and Netflix looks at your viewing behavior and saying based on your viewing behavior and looking at other people groupthink. I think that’s the next best movie for you. That’s general AI.
We have this everywhere. Right? Like if you fly, then like whether predictions are done this way, stock market predictions are done this way. Like whether you get a credit score is done this way.
Whether somebody like in my class. I even score people using this. Right? I mean, this is the general AI.
And what is the new hype? Well, the new hype are large language models. And they are as well predictions, but not the next movie prediction, but the next word prediction or the next prediction in a time series. And that’s actually pretty amazing because the next word, it felt for us.
So, you know, that human language is such an intimate part of ourselves. Suddenly that thing we call AI can talk to us like a human. And everybody is like, wow. Right?
And but that’s an agentic AI. Right? An agentic AI is essentially and I like on the surface, it’s just an interface. So it is the moment of the mouse.
Suddenly you can move a pointer and click. Oh, suddenly you can talk. And suddenly you can use language to describe what you want to have. Yeah.
You and I, we live through a time where analytics, analytics engines were done with SQL. Right? SQL is a language, a computer language. If you know it, you get your job done.
And then there were people saying, oh, I can do this with semantic text mining. And it became so complicated. And all of those startups failed to do this. And they all went back to, oh, maybe it’s still SQL.
Well, now we have a. Better interface. And now we can actually have the analyst agent, the agent who can search for us, the agent who can shop for us, the agent. And they have still the AI tools, but it becomes easier.
Yeah. You know, it’s interesting because, you know, at some level, and you and I have talked about this also, at some level, it’s less about a technology application at this point from a business decision standpoint. Right. You know, the.
C-suite leader in a company, an enterprise. And more about designing, you know, your workforce, designing your customer experience, designing, you know, sort of this, this workflow, if you will, and figuring out exactly where AI can plug into that process. Right. This idea of breaking down a job or a role into very discrete parts.
Right. And saying, OK, who does what? Who does what best? As opposed to, you know, like, you know, you see some CEOs or you talk to them and they say, OK, so this will take over the low level work at my customer service thing.
That’s the wrong way to look at it because it’s going to plug in throughout the entire process from decisions the CEO makes all the way down to the factory floor. Right. And so it’s a way of thinking about it. You know.
Let’s do a fun, fun game here because I actually, I see both sides. Right. I see both sides. I am.
I am a startup. I am a startup person. And because I’m faculty at Cornell, I actually advise CEOs. Right.
So I see both sides. But let’s let’s play a fun game. Let’s let’s test your knowledge here, Jason. Take the S&P 500.
OK. And let’s take some something which is as not even as big as our current large language revolution. Let’s take the mobile revolution. How many companies do you believe have entered?
New into the S&P 500, meaning that they didn’t enter because they move in and out. They entered because they got founded after. And now we need a date. Let’s say after the iPhone launched.
So let’s say the iPhone launch, Steve Jobs going on stage and kind of like showing it’s a phone. It’s a it’s a computer and so on and so forth. At that launch, how many companies made it into the S&P 500 till today? What do you think?
Wow. You are like good. And like audience, this is not this is real life. Jason seven is the correct answer.
But like people overestimate like our mobile revolution changed the way we live and think and do. And everybody is now on TikTok. That has to be huge. Actually, you know what?
Our value creation is still our value creation. And AI is a tool. AI is here to serve us. The mobile was.
The mobile was here to serve us. So if you’re listening to this podcast and you’re a CEO with an existing customer base and an existing business model, you’re actually good. You have customers. You have data.
You have access. Now make their life better using your normal tools you have. And now you have another tool, which is called AI. Don’t freak out.
It’s it’s a tool. It helps you. Yes, it needs, you know, at the time as we get excellent. Suddenly.
People needed to learn Excel. Now, you know, not now you get AI. That’s a reason that I have at E. Cornell.
I launched a certificate so that I can train people relatively simple to get to this to the place that they can actually help innovate with AI, help use AI, help bring AI into company. But if you are having an existing business, you should be glad and you shouldn’t be scared. I agree. I absolutely agree.
You know, so then but you take that. From the other side. Right. Which is, you know, sort of the more traditional story in Silicon Valley.
Right. Because we invest in a lot of companies who are trying to disrupt industries. And we say, hey, you don’t want to be blockbuster. You know, you need to you need to embrace this properly.
Right. And, you know, really sort of. You know, leverage, you know, AI. You know, to its fullest extent.
And so if you probably look across the companies. That didn’t adopt mobile within their existing business model that are no longer on the S&P, it’s probably a fairly large number. Right. And so what you teach and what you help people think through, and I do, too, on the side is how to think about integrating AI into your company.
Like, how how does that work? Because you have to get over the fear of the employee. Right. You know, the employees that are somehow thinking it’s going to replace them, which I don’t believe personally.
And you have to. You know, look at an entirely different process on how to integrate it. Right. So that’s actually an awesome question.
I think there are three parts to it. And because there are three parts to it, you actually will see why I. Despite the fact that I tell you, if you have a big corporations, you’re you’re you probably have a good way to use it. Just don’t be outpaced by your competition.
Why I still invest in startups myself as an angel. But as well, kind of create my own stuff. And you said it is a fear from the people. But there is as well the imagination of the new workflow.
And it’s funny that we talk agentic AI. At the moment, agentic AI is visioned like, OK, we have an AI that is very much like. A human and comes to work and has certain things and gets certain interfaces. Right.
And instead of. Reimagining how the workflow actually would work without the human in the loop. Right. Like, why are LLMs or large language models so amazing?
Because they understand what humans mean because they have the embedding space where they kind of. Really figure out what the meaning of something is. Let’s put it into the embedding space. I don’t want to get too technical.
But that that ability is what we are trying not to bring to the workflow. And one of the examples. I might say for the workflow. And you probably have loads of other ones is as we had the first wired telephone.
It’s like, OK, for what would we use a wired telephone? Well, they used it to transmit symphonies because they thought that is the killer app. You do not need to go into the symphony center and listen to them. Now you can dial in.
As we know, it took quite a while till Shopee like Spotify actually made the news. Right. But that was people thinking the killer app because that’s what their existing workflow is. So here we are.
And you your challenge as a CEO is. To not go through every workflow and say, OK, now I have. I have customer care agents. How can I replace my customer care agent?
Well, that’s an old workflow. You like there was something which made your workflow so that you needed a customer care agent in the first place. Now you replace it with an AI. Maybe build it into the product experience so that you don’t need a customer AI agent.
And that is the change. That’s number one. You probably have other examples on that. No, I think I mean, I actually think that is one of the greatest points, right?
It’s it’s this reimagining even though you have a business, but almost backing up and saying, if I could start from scratch today, what would that look like? Right. How would. If I had a, you know, a grocery store, you know, and you had robots and AI like Amazon did, how how would that look differently?
Right. You know, not just replacing humans, but also integrating humans into helping you think through recipes as you walk through the store or, you know, as opposed to being, you know, just the checkout agents within within the supermarket. And if you did that at every place in your business and reimagined. You know, your purpose and your value to your customer using AI.
And I think that’s kind of where people need to get to. But of course, But of course, potential healthcare issues and then doing preventative care. It worked well, but then the LLM revolution came around and was like, hold on, we should do something. At that time, I was as well sitting at the board of Storytel that is actually the audible for Europe.
So I was sitting at their board and I pitched one idea essentially where I took a book and you stopped reading at a book at a certain point in time and then the AI would summarize in the voice of the person reading the book for you the content up to the point. So to keep you on reading little things you might have missed, like those YouTube things. Yes. Now, so I started building…
There are people like me who fall asleep every night when they read a book. Somehow I’m on my Kindle and I’m 10 pages, it just sort of goes and I’m like, wait. Yes. So I built a platform.
And now I have this tool that I can actually create clones for my own e-commerce, like for my own e-cornell course. So the course where I talk to my students, I explain AI to my students, I explain business cases. I created a clone of myself using the same technology. So you not only get a fake Lutz that looks like Lutz and talks like Lutz, has an accent like Lutz.
No, you as well get the brain from Lutz. So everything I know, it’s like, it’s a rock solution behind it. So everything I know you can now retrieve in the course and you can check with me about business models and questions and frameworks and so on and so forth. So I built this and here comes the problem.
So I was like, oh, this is so cool. Let me go. Let me go to all the YouTubers and offer them this. I mean, I’m an old guy, but I honestly…
Honestly? Like all the YouTubers? All the YouTubers were like, WTF? No, You don’t want to replace.
No, Like, don’t even talk to me about it. It was such a failure of a product. Why? Because this is what we call the uncanny valley.
The YouTuber says, it’s like, like, I love coffee. So take James Hoffman, right? Hey, James Hoffman is awesome in coffee knowledge and he makes an awesome YouTube channel. But if you take now his…
YouTube channel and ask ChatGPT whether they have the same knowledge. It’s obviously that aggregation of knowledge is not the thing. What is the thing is James appearance, James smile, James approach, right? Or like I did a clone from Andrew Huberman, right?
And I mean, there are many like you can ask for the audience like Andrew Huberman got into the news for having had multiple affairs. And you can ask the clone. The clone will actually talk to you about it, that it’s good for your mental stage to have multiple relationships, right? So like all of this becomes an uncanny valley, a PR risk and whatsoever.
It was a product failure. Why? Because the interface is a problem and people are scared about this. In my course, I talk about a very nice study they did with x-ray doctors.
So they would split x-ray doctors into groups like group Jason group lots. So two groups and then they would give a review of an x-ray to them. And one review, it would say it was done by an AI. And the other review, it was done by Dr.
So-and-so. And then they measure AB. And obviously everywhere where it says it was done by AI, the people said, no, I can’t be right. Can’t be right.
No, If they are speciality doctors. If they are general doctors, no clue about x-rays. They’re kind of like, oh, this is awesome. I take this.
This is cool. I love it. Thank you for doing the job. Which tells us low level or low skilled.
In this case, the doctor is skilled, but the x-ray part is the low skilled part. Low skilled tasks can get up leveled quite easy. And we had an amazing. Recently an amazing publication from Eric Bjornjolfsen out there.
And he looked at the numbers and he figured out that, you know, people won’t necessarily get replaced. What will happen is you scale up because you scale up, you are happier. And I published a Forbes article about this. So don’t worry.
You will not get replaced. You will just get more and different things to do. Mm hmm. And we have seen this over and over again.
More technology creates more value that creates more work for us. So if you’re not sleeping well. Then because you have too many emails, right? And you have too many emails because you’re creating value whenever you sit down on the emails.
We are like we have seen that explosion and that’s about to happen again. Will there be changes? Yes. People will need to learn.
That’s the reason why I created the course. People will need to learn how to work with the new AI colleague. Yes. The second biggest problem in this whole structure.
First one is imagine the right workflow. Don’t just replace humans. Second one is, yeah, like humans won’t get replaced. Humans will become the agents and steering.
Yeah, no doubt about it. I mean, you can just, I mean, that’s, that’s sort of been this trend that’s and I, it’s been this trend that’s been happening ever since I started. Yeah. no doubt about it.
I mean, you can just, I mean, that’s, that’s sort of been this trend that’s, and I, it’s been this trend that’s been happening ever since technology came about, right? You know, like my accountant, you know, used to be able to handle pre-tech ex-clients, right? And now that, you know, over time he can add on more and more clients every year because technology helps him do so much, you know, of, you know, the work that he would traditionally have to do himself or hire people to do. And so he’s, he’s able to scale, right?
And scale much better. It does replace people at some level because he needs less employees in his company, right? You know, and I think it’s interesting because- Let me dwell on this because yes, it is true. He needs less employees, but on the same time, he creates more value, which has a ripple effect.
And there are two studies, which are very often shown. Like, look, I don’t want to, I don’t want to diminish the point that AI will create changes to us, like clearly, and people will need to learn. And if you’re a CEO, you need to do change management. And if you are an employee, you will need to learn new skills and tasks.
So there is a, there is a shift and that shift is, might be painful for some, might be accelerating for others. Right? This is a, but there is a shift. However, let’s take the broader picture.
And there are two cases. There is this famous AT&T study. So switchboards, there was a time where we had switchboard workers, like plugging lines, you call, it was normally a white lady. That was the job of white American women.
And then we used a computer to do those switchboard changes, right? And, by the way, my machine makes a slight sound, I will shut it off. That’s fine. Yeah.
So that it doesn’t bother you. So switchboard- You were just able to turn off your robot coffee maker. Yeah, I need to turn off the robots, yes. Switchboards are now replaced by an automated machine.
For the researchers, the good news was it doesn’t happen all at once. We went, stayed, by state by state or city by city by city. So now there is a study looking at the effect. For the people who worked at the switchboard before, many of them left the workplace because they didn’t find a new working environment.
But the next generation, their daughters, they actually went into the industry and found other valuable jobs because now phoning and connectivity became cheaper and that helped our economy. Same is true for OCR, right? OCR replaced the people who read for lawyers a lot of legal documents. I’m an early investor in Flank AI.
It’s a software for legal documents helping lawyers. Now, we first replaced the reading of documents. Now we are replacing the looking up. We are replacing the looking up of legal context and summarizing it out.
And each of those steps is obviously making the lawyer more effective, but it lowers the cost of getting a lawyer. It lowers the cost of getting support. It lowers the cost of creating a product. So we consumers definitely will see the positive impact here.
All that. I absolutely. I completely agree with you. I think there’s, I think we have too much hype.
You know, it’s, it’s doomsday AI porn, right? It’s this idea, you know, that, you know, everything’s going to collapse and AI is going to, you know, you know, take all the jobs. You know, we still have bank tellers, if you will. Right.
And, you know, you, talk about the switchboard, you know, there’s, there’s no doubt about it. Right. Like, you know, traditionally, sadly, work was divided up between men’s work at the time. And women’s work, right.
Women’s work had, it was in the homework. And we automated that in the fifties and sixties to such a point that with your coffee maker right there, with your clothes machine, that women now make up over 50, over 50% of the professionals in, you know, tech and legal profession. And I mean, it’s just, you know, so they were able to make this shift. And, you know, when I was graduating college undergrad in the early nineties, um, you know, if you talk to anybody who’s before Netscape went public, if you talk to anybody about what, what would be the greatest jobs, you know, in the new century, nobody could have envisioned any of the things that we are, they wouldn’t have envisioned what we’re doing right now.
Right. You know, podcasting and, you know, as a YouTube group, they wouldn’t have imagined, you know, all of the amazing things in Silicon Valley, right. There was some semiconductor work. And, and so when people say, well, okay, I is going to take jobs.
Well, what are the jobs? Are going to be created? It’s like, I don’t know, who knows, but there’s going to be a lot of them, you know, it’s going to be fantastic. Now, maybe you should talk a little bit about the hype.
I mean, there is the doomsday scenarios. There is the discussion about AGI coming tomorrow. I mean, like, like a year and a half ago, there was one, one, uh, blogger who kind of counted down till his prediction of AGI would happen. I was sitting there and thinking like, gosh, dude, have you ever read some technical paper?
Well, do you know? I actually know what this is all about. So, um, I think, and I, I recently published an article on Forbes about this. I think there is a very good playbook here for CEOs who think about protecting their business because, you know, when we overcome those two, we overcome the, as we said, there is the issue of imagine the workflow, reimagine the workflow.
And then there’s the issue of train your employees so that they can actually reimagine the workflow. I like be more effective. Once you have that, there’s actually a challenge. And the challenge is how do you protect your turf?
How do you protect your business? This is, this is not something new. This has always been the case in businesses. And if we look at some Altman and OpenAI, they actually flow perfectly through it.
Sam Altman is up against Google and Microsoft. He needs to say, we are the coolest kids in town. The best way to say it is like, we have figured out AGI out. It’s coming tomorrow.
Like subscribe. He. If you really want to get close, right? That is a, is a good story.
Now he as well has obviously a message to their competitors. And we also, that is nothing new in business. Sam Altman goes on stage and says 500 billion debt investment, like hype cycle debt, right? Debt investment, 500 million debt investment.
Dear competitors, watch us. We put so much money in. You don’t want to follow. Right.
And, and I think perplexity pretty clearly read that news. Right. And then, and then he says, open AI, like AGI will be so dangerous that only the very few should actually being allowed to deal with it. AKA me as may I.
Right. So, oh yes. And IP rights. Everybody should look for IP rights.
Except us because we already violated them and we are good. Right. So there is a certain playbook. And let’s take those.
Let’s take those doomsday scenarios out. Understand the playbook because there is nothing better to learn about business by just looking. Yeah. You know, it’s interesting because, you know, you know, you talked earlier, you know, we talked about this hype cycle and, you know, the number of companies that, you know, in the mobile that made it into the S and P and all this kind of stuff.
And, you know, as you say, if you have a business, you have customers, you have data, you’re good. You know, there’s another, you know, it’s an interesting thing because I don’t. I don’t have the data in front of me, but I think in 2024. In America, there were 15,000 AI startups financed across, you know, hundreds of billions.
It’s a crazy number. And the question that I continually come back to, you know. By the way, I just thought I had looked it up earlier. $31 billion in funding.
Went into AI startups in 2024. Right. And across, across thousands of companies. Right.
And so. Yes. We’re outsourcing our future, which is kind of amazing, isn’t it? It’s amazing.
But, you know, and so I keep having this, conversation over and over and over again on, you know, one of the problems that we’re, we have this hype cycle, but then there’s also this sort of level of reality. You know, this, this level setting of reality. Yeah. Where, when open AI or perplexity or anthropic or whoever comes up with a new version, it wipes out 10% of the startups that got financed in AI.
Right. And that’s just, and so it’s thrown this, this confusion into, you know, the funding VC community because it’s like, all right, where is exactly your mode? Right. Eventually.
I get it. GM will continue to exist. And. Eli Lilly in the pharmaceuticals.
We’ll continue to exist. But how many of the startups, you know, in the tech provider, you know, ecosystem will exist. Right. And I can talk about this a little because essentially I decided to leave my Nasdaq company in order to create a startup.
Right. And you can follow two directions here. One direction is you become the platform supplier. You become.
The next open AI. Now I very early on said I would have invested in open AI. So definitely not building my own there. Why?
Because it was clear that you see a radical depreciation of the cost factor, which makes a whole debt investing in Stargate even more critical. Right. So there is a huge depreciation ongoing because we are in a very rapid cycle and it’s software. All the talk about AI.
Being similar to nuclear energy. No, it’s not protectable. It’s not like it’s software. You can copy it and you can learn from each other as we just saw in stuff like deep seek and so on and so forth.
So it’s software. And because of that, we will see a depreciation. There is obviously a platform need and there will be a mode for platforms. Now.
There are good contestants for this space. These are the Googles, the Microsofts. And maybe there is one. There is one really early movement.
Maybe, maybe the judge is still out. It’s open. But that’s one direction. The other direction.
And that’s the direction I went is that I said, OK, now. Let’s look at the workflows. Let’s look at the workflows and reimagine those workflows. And I.
I iterated through like in my course, I actually like from a professional point of view, if you do go to the Cornell course, I go through many different industries. Healthcare, media, finance, legal. I do many of those cases. Now, for me personally, I decided the if we have a new interface.
The area that probably changes first is where interface meets money. And that is an e-commerce. So if you look at Amazon at 1999 and you look at Amazon today. Like, what do you think, Jason?
Different look or same look? Same look, but a lot. But a lot behind the. No, but it’s the same look.
It’s search. It’s like you have filters. But like there are different shopper people. There are people who want inspiration.
There are people who research and they leave an Amazon and go to YouTube and figure out or look at TikTok. And then there are like those high intentional shoppers. So we have different shopper types. But our interface is still the same.
It was like, hold on. It’s kind of not a good idea. I just explained to everybody how generative AI and AI can create together a new suitable interface. So we essentially my company is called R2Decide.
R2 like R square and decide like decisions. So we call R2Decide and we are a toolset for e-commerce shops to actually make the personalized storefront. So how does this work? We are.
We are on Shopify and you can download and test us easy. But instead, like we take over search. So first of all, we are better in search. And like, for example, like we work for a jewelry store, one of our customers.
If you type something like colorful child bracelet, most searches will actually move up because they look for colorful. If it’s not written colorful, but red and green, then they don’t know it’s colorful. So child. If it’s a mother child bracelet, yes.
But otherwise it’s not written. The typical word search and those fails. Now we look at the image. We understand that child or like, for example, we will bring up then in this case a melon because kids like fruits.
Right. So so we understand the image. We understand the context. And we do obviously as well do text search.
And but that’s not the only thing. It’s not only that we have better search. What we then do is. While you are.
Actually searching. We give you small little nuggets. Oh, you’re looking for a child bracelet. Think about the following.
I don’t talk you through. Like if you go to perplexity and it gives you. One thousand words to read about like what to consider for child bracelets. Nobody does this right.
This is the same mistake Google did with your Google home. You kind of saying stop the timer. And Google says consider the timer to be stopped. WTF.
Nobody gives me those seconds of my life back. Right. So you want the AI to serve you. So we build an interface which is exactly this.
Like it’s it’s tiny nudges. We so. But now AI, as you said earlier, it’s not a technology question. It is an interface question.
It’s a question. How do you build it in? How do you make those tiny nudges? Now, once we launched it, it’s actually a little bit more complicated.
Right. we launched it, it’s actually amazing. People love it. We have 90% more engagement.
We have 11% more revenue, like on average from everybody who is doing it because, and the customer doesn’t realize. It’s like a tiny notch saying, oh, if you’re looking for this, consider. And it’s small. They can overlook it and continue with their path.
It’s not uncanny valley. It’s not saying, oh, I’m the AI. I know better than you. No, it’s also not lung.
It’s not the Google. It’s a nudge. And I think those nudges, it’s like if you drive now your car and suddenly the steering wheel starts to shake a little bit because you fell asleep, that’s it. Do it soft.
Do it in the workflow. Imagine how AI can be helpful for you. It’s interesting because there’s a lot of people obviously working on this sort of AI applications in the e-commerce. Right.
And there’s trying to find, for large ones, how to really figure out your sort of high value customers when they hit your website. There’s other people who are trying to figure out how to identify and really build a profile on a customer almost immediately. And it’s a hard thing to do. A lot of people go to e-commerce and they spend half their time and it takes a long time to build up.
So people are trying to solve that. But for what you’re doing, right? Right. May I say something?
Because the technology approach is here wrong. Very often. And again, this is awesome that we have this discussion because it shows how we use AI based on a human approach. It used to be that you have a marketeer and he says, okay, I have three customer types.
I have young women and I have teenagers and I have… The busy midlife guy. And those are my three images now. Dear data scientists, build me a model to fit them in.
And then I play out the best advertisement to them. I’m simplified. It’s obviously a little bit more nuanced, but that’s essentially the approach. We are doing classifiers, traditional AI.
Now, the beauty with large language models, with an embedding model is that I do not need to codify child likeness. I do not need to… The model knows what child’s like. And that’s what we do.
And that’s actually the meaning. If you think about re-imagining the workflow, as we said, this is the biggest challenge, then use the tools as they work. Do you need a break? I see that.
Hold on one second. I like that. Hold on. Sorry about that.
No problem. It’s all that you got nervous. So, totally fine. The UPS person in my door and they’re trying to get me to sign.
I’m like, no, I’m not signing. It’s not happening. Normally, I don’t leave my windows open, but it’s such a beautiful day-to-day right now. And I was…
So, anyway. Yes. So, yes, I agree with you on that. So, let me ask you this.
Right? Why can’t… OpenAI do that? Why?
Like, why? Like, you know, like Anthropic just came out and said, hey, we’re going to create this sort of personal shopper for you, if you will. And it’s going to be your personal, you know, agent that goes around and deals with these APIs. Yes.
So, why can’t they do that? They will try to do it. And I don’t think that Open… So, everybody has now a certain set of cards.
And I don’t think that OpenAI will go there. But, like, Anthropic will try because they are missing out on the big part. Perplexity will try. And so, you have a consumer-facing part.
So, the idea how Perplexity thinks about this is, I will build this personal shopper for you. You, shopper, give me Perplexity, your credit card, and I will guide you through. Now, there is a couple of problems here. Mm-hmm.
ask in if you ask me what espresso machine should i buy and i would figure out what kind of taste profiles you like and how um how much you love to get into this at some point in time i would recommend it to you if you ask oh may i the general public would not use such a nerdy machine so they would say breville is good or go for an espresso it’s easy to clean right and meaning average we get average we like we we misunderstand what those large language models do jason let’s do another game yeah you complete my sentence life is like a box of chocolate why what why the heck like i mean come on said it and everybody knew it before them and once forrest gump said it then it became you know globally known but that means that you are brainwashed by the american media industry go to kenya and ask that sentence they will say oh it’s a like a box full of surprises or a box of happiness or whatever only because we listen to forrest gump you are taking now this average so now here we have a model called perplexity oh may i whatsoever and that model will recommend you average right now it will say okay i learn about you and i will understand what you do and yes technical this is feasible but it’s takes time so after i bought my second or third espresso machine maybe it understands that i’m really into espresso right but it’s it won’t be late though right yes and exactly it’s not the right thing so that’s issue number one models are average so what do we do is we go to the brands we we work with the big brands our like a dream customer from we would be rei or footlocker of those because people come and they come with a certain brand awareness and i built that brand awareness into like like let’s say rei you don’t get any kind of tent you get the tent that is based on the rei knowledge and value okay so that’s number one it don’t do average like build out and number two is let’s say they overcome it and it works for easy You go to Perplexity and say, give me toilet paper. You don’t really care so much. Maybe Perplexity gives you a thousand-word essay about bamboo versus mesh material versus two, three, four layers. And you’re like, I didn’t want to read it, so there’s a UX problem.
But let’s say they overcome this, and now they order the toilet paper. Good for them. That’s fine. And that will exist.
But on the other hand, the brands and the shops, they will be completely replaced by Perplexity. So they will offer you guidance, which becomes more like you go to Macy’s or to Ashley and you get your personal shopper. And that is based, again, on the idea, on the vision, on the style from Macy’s and Ashley’s. And you go there because you want the experience.
And you go there because you want the advice. So there will be both worlds. So yes, OpenAI and Perplexity will try to build those. They will face the average problem.
They will face the usability problem. On the other side, the shops will do the same thing. They will not face the average problem because essentially for our customers, we train for each of our customer and our model. So the model is the customer.
The model is the brand. And they will not face the awkwardness of long discussion because, okay, if you say sleeping tent for Bolivia, kind of like, oh, okay, that’s a tough question because you can be either on the beach. Then it’s the light sleeping tent, like only three temperature, like high temperature grade. Or you go to the high mountains and then you need a really pro sleeping bag.
So I need to ask back. If you do this in chat, GBD, average problem, long list, and now I suddenly need to read again and life becomes complex. Both worlds will exist, Jason, and they will come together. Now, in my vision, we haven’t built this, but in my vision, I do have as well an API to talk to other agents.
So in my shop vision, if not a human comes, but an agentic AI, I’m obsessed. I’m absolutely going to serve them as well. Yeah. I mean, I think that’s the question, right?
It gets back to AI Lutz, right? It’s your own agent, right? And your own agent can provide value to other people. And then in sort of the dream world of AI, it is sort of your avatar out in the world, right?
Yes. And so it learns me. And I say, hey, go book me a flight to Hawaii. I want to leave on Tuesday and come back Wednesday.
Right. And it knows that I don’t like a middle seat. I’m willing to pay up for premium economy, but I don’t like first class too expensive. I like aisles over windows, right?
If I could be in the back of the plane with front of the plane, if I could leave morning is preferable to afternoon. And it kind of goes and just works all that out for me and then comes back and presents that to me and deals with the APIs of these other companies. Right. Yeah.
Sounds like a business. Sounds like business. screen and thinking gosh how stupid is that person right so we are just yeah created an agent that talks like your highly skilled um partner of your investment company or a consulting company that has the knowledge more more likely of a general internet so it sounds good but it doesn’t mean anything so if um if you are a manager today and you’re managing people you know how complicated it is to get even a skilled and intelligent intelligent individual into the workflow um and bring them on board follow the setup follow the structure think like the company does yeah hard right that’s a managerial job well now you have this if you are conduct like if you’re leading your ai team you have the same problem because um like gosh what you just described is a horror scenario if i want to book a flight i i like i’m living in silicon valley i teach in new york right so i do fly quite a lot if i have to book a flight and all those questions middle seat side seat out seat like exit row what is the time should you book now should we remind like which airline i would like leave me alone you know i i do it by myself this is getting too complicated so we don’t it would already know it about you like you wouldn’t deal with that like it would just kind of hold on that is not yet the case that’s not the case we actually need to have a tool that understands the situation and the situation and the situation and the this is an interface question at what point in time do i ask there is a large language models are there isn’t there is another side to larger language models which is like research and how can i use those completion in terms of time series to build better drugs better models and so on like let’s cover this let’s not cover this because we talked mainly about corporate America and the United States and how they can use but if if i have a large language model as corporate America it is not a technology problem it is actually a problem of interface if i have a self-driving car at what point in time does the car tell me i’m not certain anymore please take over if i have chat gpt writing me an email at what point in time does chat gpt say i’m done please read the email because it might completely misrepresent you and sound totally weird and like we humans have agency and one one fact one fun fact i obviously in my course i take um in the e-cornell course designing and building um ai solutions i do talk a lot about bias i talk a lot about risk so for every industry i go in i look at what are the potential potential potential potential um not wanted side effects the unintended consequences the biases the problems and there are technical solutions for it but i do look into it each time and i have done the course a lot with support from my own clone right so i actually talked to the ai lutz to to so that the ai lutz tells me what lutz should have should say in in the course um so but you know what ai lutz as well as chat gpt become pretty lame when it comes down to open ai’s risks or ai risks in general why because we don’t have enough knowledge so now suddenly and you can have two different stories it’s totally a risk it’s doomsday it’s not such a risk it’s like there is and depending on how you tease it and how you prompt it the system will give you an answer we humans are not being replaced we humans have an agency we humans need to steer only because somebody invested in the car does not mean that we humans start stop to exist now we are using this yeah no it’s definitely at this point and and going forward you know for this foreseeable future it’s definitely around you know the amount of agency you’re willing to give up is based on the risk of the decision right on the toilet paper um analogy you used it’s like what do i care two ply one ply you know made here made there whatever on if i’m going to go mountain climbing i’m going to go mountain climbing um scaling you know el capitan and i’m buying something from rei i need to i need to be assured that i have the equipment that works best for me right and so you know i’m willing to give up less agency to make that decision because my life depends on it right you know but you know on sort of a different topic you know you know you talk about um you know the box of chocolates and you know it’s an american sort of thing you know american media how do you see playing out um ai investment you know we talked about you know some ai investments a little bit earlier you know america is just dumping hundreds of millions of dollars into this across thousands of startups and i saw something i think it was in tech crunch today or yesterday that europe for instance and i know you work with uh some of you see firms there and you’re from germany um europe only had eight billion dollars in ai investments in 2024 and then you have asia and china and you know japan and just and they’re more on you know in sort of at the level you know of the u.s how does this play out like what what’s going on with europe and a lot of it has to do with how data is being used and privacy and sort of these rules around that like how does all of how do you see that playing out and and we’re talking about like you’re saying you know sam altman saying this could be a national security risk only a certain number of people should do this and so we were kind of in this discussion about you know how do you see that playing out and i think it’s a really interesting discussion of my gosh deep seek is you know quote unquote chinese right and then you have open ai is the u.s and it’s a winner take all situation and we need to invest in it and everybody’s sort of putting up these barriers and like how do you just see that playing out so yes and i i wrote a lot on this in forbes because i think essentially when i say it’s an interface and when i say current businesses are well set to compete with each other and i think it’s a really interesting discussion and i think it’s a as long as they train their employees to deal with ai whether they do this with e-cornell and go to my course they’re welcome then they can chat with me i have a private chat group then afterwards where everybody goes in either there or any other course but train them because once you are innovating in ai you have a good chance and why because there are two things you need to have you need to have customers if i like i can build they business Facebook clone, a meta clone. Good luck.
Nobody would come. But it wouldn’t take so much technology to build it. So there is access to customer and access to data. If I want to be better, I could try to make predictions like Amazon does, but good luck, I don’t have the data.
So these are the two major assets which companies have. And those are the assets which not only companies, but countries, states have. So it’s a question, how do you protect them? Now then, there is a whole discussion about the actual tools.
And I think it’s very foolish to think you can confine or remove those access to those tools. And there is a discussion on DeepSeek, I wrote about it, that I believe that because the US has pushed China to not have access to H100, they built everything on H800, and now it works and it’s cheaper and it’s faster. Humans are chaotic systems, and we are very innovative. That makes us out, that makes us as a race, it’s a beauty.
Large language models are not, just to be very clear, large language models are just stacked logistic regressions, or linear regression and activation function to be very accurate. If you want to know more, go to my course. But it’s simple. But humans are not that simple.
We are more complex. Now, that’s a reason why countries as an extension of enterprises have the same problem. Access, access to market, access to data. Now, I am, you can’t see this here on the video, I’m 6.4 tall.
I’m a tall guy. I recently did the calculation, and you will love this. Audience, if you can watch the video, then this will be, you will get a goodie. Now, I kind of said, okay, let’s look at the top 10 European corporations that do AI.
These ISAPs and so on. Let’s look at the top 10 US corporations that do AI. Now, let’s say, let’s take all of those enterprise values from the top 10, and let’s say this is 6.4 by height. Like, for the top 10 Europeans, how big, what do you think, Jason?
How big is it? 10 representing what? 10 largest enterprise companies in Europe, focusing on AI. If the 10 enterprise companies in the US are 6.4 tall, how tall will be the European one?
I’m going to say they’re 8. 6.4 is me, and they are 8 feet? They would be, you’re the US, right? I’m the US.
I’m 6.4. So Europe is probably smaller. Well, I was thinking they’re bigger because there’s less companies, and so they take over more space. Now, this is where the visual clue comes in.
That’s this. So, if you don’t see it, I’m holding a 4-inch figure, which I got at Amazon, specified to 4-inch. The US top 10 AI companies are 6-4. European top 10 companies are 4-inch.
So there’s a lot of growth potential. But why is that? I mean, it’s a million things to dig into, but I mean, just really quick, is it a data problem? Is it a regulation problem, I guess?
It’s an innovation problem. It’s a regulation problem. It’s a funding problem, right? I’m, with my startup, I’m out here to seek funding.
If you’re listening and you want to give an amazing round of money to me, like to change e-commerce, do so. I wouldn’t go for Europe. Why? Because America is my market in this case, right?
And therefore, I’m looking at US investors or investors who are focused on the US. Funding problem. It’s a risk. Taking problem.
My friends in Europe look at me and it’s like, what do you mean you quit your president of Nasdaq company job to a startup? Like thinking about changing e-commerce, right? I actually talk to people, it’s like, nah, e-commerce is not going to change. It’s like, yeah, you know, there is a risk thing.
And then there’s obviously a regulation problem. As soon as I say I train on certain behavior, then everybody starts freaking out. No matter whether I actually just look at non-personalized information on it. Yeah.
You know, plus it’s also an employee problem, right? It’s a talent problem, both in terms of access to talent and in terms of how America, we can get rid of our talent quickly, right? We have an amazing team. We are all like, there are a lot of engineers from Cornell that joined this setup because they’re like, we have AI experts and we have engineers and we build really cool solutions.
Trying to put all this off in Europe, it’s just… It’s hard. So before we go, and I know you’ve got to go, what are some last thoughts for business executives as they go down this path? Sort of the high-level takeaways that you want them to have from this conversation and from what you tell them normally.
High-level takeaways are AI is not something special. Do not think that you are replaced. Employees get replaced. The business is replaced.
All of this is a pipe dream. And they create a smoke wall for you to actually see what’s happening. Now go out and tell everybody that those pipe dreams are true because you scare off your competition. That’s exactly what OpenAI is doing, so you can learn from them.
But other than that, reimagine every work. And because you as a CEO, you cannot reimagine every workflow. Use education to train your employees. And how do you use education?
Well, guess what? You use AI. You can use a Lutz clone who talks to you personally about your business models. You can use a co-pilot from eCornell to code without having to code.
My course is technical, very technical, but it’s a no-code course. It’s actually working for everyone. Of course, I believe AI is here to serve us. So don’t be scared.
If you have a business, you’re in an awesome position. Fantastic. Well, thanks for coming on. It was my pleasure.
Jason, see you around. Well, there you have it. I hope you enjoyed the discussion and don’t forget, follow me on Twitter at ThePulseOfAI and connect with me on LinkedIn. Until next time.