"What If You Could Just Talk To Your Website?" · aired May 29 · word-for-word from the recording (light caption cleanup on names only). Use the timestamps to find any moment.
00:00 It's happening. This is Practical AI episode forty-three. This is a special show today. We have a world premiere. We are going to see the lovely Olga talking to an AI and running a website. Yes. What the heck? We are going to have a live premiere today. So you guys are in for a treat. My dream came true. I just talked to my site and it does things for me.
00:31 So I can't wait to show you what that looks like. I'm talking about live, not like some prototype, not some GitHub, whatever. Live website. in the real environment. And we have so many news to talk about, so many trends, so many different patterns. So this is the place to be every week where you can learn the latest and the greatest in AI news and trends. But more importantly, how do you actually use this information to do something with it? Because here's the deal, you guys, we're moving away from who's got the best
01:01 model out there. We're actually moving now to the operational and practical application of it. And whoever is getting this information and using it right now, is winning big time. So congrats on being here and making this investment in yourself. And the worst case scenario, you'll walk away being the most interesting person at dinner somewhere tonight or this weekend, because you'll know a lot after an hour. So let's dive into it. We are actually going to get started with the news, as always. And basically, a couple of weeks ago,
01:34 we talked about the first news of the week is the Glasswing project, the Mythos model that Anthropic never revealed to the public, but they did reveal it to the chosen group. Few? How many companies did they have? Forty companies got invited to Project Glasswing. I predicted that they would open this model to the public by the end of May. This did not happen. It's still under wraps, only used by these companies,
02:06 and they say it is because of cybersecurity. Wow. Do you think, now that you've had some time, do you think they're actually going to open it up? To the real life, to the real people, to the real users. I think what they're really, the real problem that they have is not really about the model. It's about compute. And I don't think they have that shored down. I think that they're currently figuring out compute for their existing models like Opus four point seven and now four point eight that on down the pricing and the
02:37 usage. They're trying to get a handle on all that. I don't think they're going to release mythos until they have a better handle on that aspect of the business. Well, here's the deal. So we talked about how Mythos has been making the news and basically finding the bugs that are like twenty seven year old bugs out there just sitting in the open. But this week we got the report that Mythos apparently found about ten thousand flaws in all kinds of models around all
03:07 the companies that that that they deployed it in. And so I found that many in WordPress last week. So it found more than ten thousand serious software flaws. And so what kind of made me giggle is that's just thirty days. Right. So it flagged ten thousand high or critical flaws across one thousand projects. Cloudflare alone, here's the deal, pointed at their own systems and found two thousand bugs. I don't know if that's a good thing
03:38 to report. I know. Why are they saying this? I don't know. I pointed mythos at PageMotor and found nothing. Independent security firms confirmed over ninety percent were real. So I don't know if it's like a new like flex these days. Look how many bugs my software has. Or maybe it's like a play. I have so many bugs, but look, we're fixed them. So now we're very secure. So we have a mean saying for this. You got to keep that to yourself, cowboy. Yeah, but I mean... You're finding all these bugs. You got to keep that to yourself.
04:09 Well, so I think what's going to happen, we're going to basically think of it that way. Right now, it's only previewed to those fifty. So I guess once they have some time with it and they're going to reveal it to the public, anybody can go out there and find flaws in anywhere. And what was interesting is that how it works, like how do you find bugs in all the secure software? Like humans will have to, I don't know what hackers do, but they have to go and like scan through and find vulnerabilities. And it takes some time.
04:40 And for some of them, it's not worth it. But for AI, it can get through the software code, I guess, a lot faster than the humans would. So that's why they're able to find these bugs so much faster and better. Yeah, there's patterns for vulnerabilities. It can scan huge code bases looking for these patterns. I think it was last week or possibly the week before. One of the things that it did that I actually thought was impressive is it was able to figure out these two-tiered vulnerabilities.
05:10 Like the vulnerability doesn't just exist in the OS, but if you do this one thing to it and get it in a different state... Then there's a vulnerability. That kind of stuff is really hard to spot if you're just a human working through code. But the AI obviously has a capability there that we do not. Well, let's see who's going to be brave besides Cloudflare to come out and say how many bugs they had. But, you know. Try to keep that to yourself. kind of funny anyway uh second news of the day anthropic shipped this new feature
05:40 in cloud code called dynamic workflows and it's switched on with the mode called ultra code you ask for a workflow and clot spins up a fleet of agents up to one thousand that builds and checks each other's work it does a night what used to kind of a dozen a night what used to take um months And it only does sixteen at once, but it can do hundreds across a job. So here's what's interesting about it. It's not just agents working together on one project. So apparently the way it works is like
06:11 a real job or like a real company. Half of the agents are building. The other half is trying to find and poke and critique the work. Yeah, building and checking. So what do you think about that? Kind of what are your thoughts on that and which is something that you think you're going to use? As a theoretical model, it sounds good. It sounds good. In practice right now, we're just creating more slop that's higher checked but faster.
06:41 So that's not so great. What I actually see when I look at this is a pretty big engineering challenge for people who are actually going to be deploying this and trying to build stuff. Your job is to make sure that this thing's building in the right way. To me, this implies imparting skills on it. in the form of markdown documentation and like continuous refinements, like watching this thing like a hawk and running over and over and over this thing. Right. Exactly what I do every day. I'm just having to do it with one
07:12 and not like sixteen sub agents. So to me, I look at that. I think, wow, there's a lot of work that's implied to actually make something that's good. And that's not building on priors. Like whatever the AI already knows may not be the thing we want to carry us into the future. It needs to learn how to build a different way better and all that stuff. But I mean, there's different facets to this. Like AI is good at something called linting, which is checking files for errors. It's good at pattern matching, so we can say, well, we want this particular thing to be done
07:42 this way throughout the system. And then a sub-agent can go out with this in mind, survey the code base, and say, OK, well, we found two violations of this, that kind of thing. So there's aspects of it that are good. But for the whole orchestration to work, a lot of training data specific to what you do is going to need to be added to that. So I'm like, you know, that office that, oh gosh, that show, The Office, I never watched it, but there's a lot of memes about it. So I wonder if there's like an AI office edition coming out, like what agents are actually arguing
08:13 about and what kind of memes can come out of that. The memes are coming. They don't need coffee or sleep or anything like that, but man, they do need a lot of tokens. So Greg Eisenberg, who will come up later in the show, he said he's loving this, but that it is a token hog. Okay. Do you think you're going to try it next week? No. All right. Well, there you go. Okay. Next news of the week is Anthropic raised fifty billion, which we will talk about that later.
08:45 This was very frustrating because everywhere in the news, they're reporting sixty five billion dollars. And I know they're like, we're penny pinching, you know, fifteen billion here, fifteen billion there. But they're reporting sixty five billion raise. And I'm like, but Crunchbase says fifty billion. Crunchbase is like your books. Like every week there's books like, you know, in and out. And Crunchbase says fifty, you know, everything else, including Anthropix, you know, press releases sixty five. So I'm like, are they just trying to like, you know, make it look really impressive and pump up
09:16 the valuation? Because obviously, sixty five sounds better anyway. So the official thing this week is fifty. And then the fifteen other ones are from the previous week's commitment from Amazon and others. And so I thought that was fascinating that, you know, they are they're got fifty bill. I mean, we'll see what's going to happen. But they are valid at like nine hundred and sixty billion dollars at this point to go public. So I saw something interesting. So Anthropic valued it. We'll say it's nine hundred, nine hundred billion dollars. Right.
09:47 Walmart has a similar valuation. That is crazy. Walmart has seven hundred and twenty five billion in annual revenue. So their EBITDA on this sense is barely over one. It's like one point three. I mean, but the revenues are twenty billion. OK. Seven hundred twenty five billion. I just don't know. Do valuations even matter anymore? Somebody who I like on Twitter, Nick Carter, he was saying, he's like, this is just clownish.
10:17 Nothing means anything. Well, so the same week, the Anthropic opened an office in Seoul. Seoul? Is that how you say it? I don't even see the word. South Korean. It's in South Korea. Apparently, South Koreans are adopting. Oh, Seoul. Seoul. Apparently, South Koreans are adopting clod three and a half times the expected rate for their population. Why? Because they don't have any children. No. They have the lowest total fertility in the developed world.
10:49 That's what they're adopting Claude. That's funny. Okay. No, but I'm just curious. Why? Like South Korea? Like what is there? Tech and gaming is like their thing. Well, there you go. And then the same week, apparently the Vatican. So Claude and Vatican is a thing these days. So Pope Leo released forty two thousand word encyclical. Do you know what encyclical means? It's like an encyclical, and it's a big declaration from the Pope. Okay. Well, apparently the Pope has spoken.
11:22 I didn't read about that. And he basically wrote about protecting the human person in the age of AI. And anthropics researcher, Chris Ola, took part in the presentation. So now, anthropic is part of the ethics and morals and all of that in the Vatican. So they're everywhere, you know, South Korea. Oh, yes. When I think about Anthropic or any AI company, I definitely think about morals and leadership in that regard.
11:53 I mean, it looks like they're trying to govern themselves. I mean, it looks good on paper. Okay. One of the things I do like about Anthropic, and you and I talked about this before, is how seemingly open they seem to be out of all the AI companies out there, especially on X. Um, Boris, but he's, you know, like on, on, on, on X he talks like, I feel like anthropics, uh, engineers are more public and more, um, accessible on X sharing their updates and
12:25 news. I don't see anybody on open AI and maybe I'm just in a silo. I don't know. I'm curious if anybody kind of sees something different. The same thing with Gemini. Definitely with Meta. I don't know. What is Meta again? I know. We'll talk about that in a second. Meta rising from the grave this week. I guess the only other person that openly shares what they're working on is Elon. But I'm just talking about how interesting that I understand. I think Anthropic makes it more open to
12:56 see what they're actually working on. Definitely more visible. I feel like there's a communications channel there. where like OpenAI, huh? Only thing I ever see from OpenAI is stuff that Sam Altman has said in interviews. Maybe we need to see what's going on there and who else is there. And Gemini, has anyone from Gemini ever said anything? I don't know. Are they even there? No, no. I guess it's like, why? Are they just trying not to say dumb things? I mean, honestly, I was just thinking through the memory bank here. The only person who works at Google who
13:27 said anything about AI that I've ever seen was a dude on Greg Eisenberg's show demoing like the Stitch... You know, Google Stitch. Well, I like that. I think Anthropic, that's actually one of the endearing things. But somehow I trust them more because I see their people actually talking to us muggles. No question. Yeah. No question. Anyway, moving on. Taiwan is in the news. So NVIDIA announced it's spending a hundred and fifty billion a year in Taiwan. and breaking ground on a new campus amd
13:58 added more than ten billion and almost all of the world's AI chips are made on one island yep okay that blew my mind like what's why this one island why are they doing it all there i did not know this it seems to be very concentrated what if something happens to that island and nothing that island is politically disputed It's what? That island is the subject of constant geopolitical dispute. China wants to claim it. Taiwan claims independence. There's a whole lot of tension there.
14:29 There's a sea between China and Taiwan, and China maintains a naval presence there that's kind of imposing. There's always this lingering threat of like, is China going to take Taiwan officially? Okay. Well, Huang, NVIDIA's CEO, Jensen Huang, he called Taiwan the epicenter of AI revolution. I'm going to have to go and read more about it this weekend because I don't understand why. What's making it so unique that we can't repeat it anywhere else?
15:00 I think they've been making the chips there for a long time. I think they're the number one buyer of silicone. I don't know if silicone is manufactured there. I don't know. But I think they buy it all up basically. I'm going to do some homework because Taiwan is like on the map and AMD is following the lead. Yeah. It goes through Taiwan. Yeah, every autonomous agent in the Thunderdome runs on cutting edge silicon made predominantly on a single island. But why? Very, very creepy or kind of interesting.
15:31 Well, I mean, it must not be that easy to manufacture. And they've had, you know, a forty year run up here. I guess it will be an island in Taiwan, and then it's going to be what Elon is doing, the Giga, oh, gosh, the Terra. In space? No, they're opening the chip maker here in Austin. Oh, yeah. Terabyte? No, God. TeraFab? TeraFab. Anyway, so we're going to have some here, so we'll find out what's going on there. Another piece of news,
16:02 last week we talked about how one of the long-term math problems was solved by open AI i believe last week and this week it's like i swear one thing one company does something the other com the other week the another week the other company has to like up it up somehow so google deep minds AI solved nine unsolved math problems two stuck for fifty six years and verified its own proofs with no human checking OK, why does this matter, Olga? Why are we talking about this in the news?
16:33 Well, because for a long time, AI has been something that's just like, what is AI? It's just the glorified, out of complete stuff. It just, it learns from patterns. It completes the sentence based on the patterns, what's going to happen. So it's really, that's why hallucination rates are so high in the AI, because it's like, it's not, we think it's thinking, but it's really not thinking. Like, it's just kind of like, out of suggest, based on the patterns. Well, these problems in math is actually a really good test of figuring out, like,
17:04 is AI really thinking? Are they solving through the actual problems? And that's why this is interesting, and this is really good to keep track, especially for AGI, because that's the true AGI is when AI can actually come up with solutions on its own and doesn't need to... um kind of out of fill things so is that kind of how i understand this from from non-techie i mean i think that's fair but to me i just hear blah blah blah can it fold laundry yet I know. We're going to check every week. I don't care about some goofball math
17:36 problem that has no relationship to the real world. I have a feeling this might be next. This might be next. Laundry is coming. I'm really hoping that it does. So DeepMind is on the map this week. So that was interesting to kind of see what they're doing. All right. Then... Practical application this week, Zuckerberg, the creepiest of all of our AI partners. Making moves. So far, making moves with this Biohub.
18:07 They released an AI that mapped seven billion proteins and can invent brand new ones to fight disease, so they say. We don't know what... Well, there's never been any mistakes in that realm before. This seems very positive. That's going to happen. So think of it this way. So I didn't know this, but basically just like once AI knows the language, it can write all kinds of things. And once AI knows all the proteins, it can create all kinds of solutions. So this time it learned biology to invent
18:40 proteins that have never been existing. And I'm like, so is that for the drug discovery or is this for something else? Like what can it be for? I don't know too much about all those interactions. I know that like with the COVID vaccine, it was spike proteins. So I think you can influence biology heavily by introducing a protein that is engineered to behave a certain way. I think that's probably the general line of thinking there. Why it made the news, they're saying that they're compressing years of drug discovery into hours,
19:13 and they already designed novel protein binders aimed at cancer and immune diseases. I am hearing this screeching of Eli Lilly and other pharmaceutical companies because I know there's plenty of people who don't want the drugs to be discovered, but I'm curious to see how AI is going to solve this, and more importantly, how it's going to help people. I've said a bunch. I'll say it again. AI for bio stuff, huge. You did. This is your lane for sure.
19:44 Well, humans are going to shift in the next ten years. We are really going to shift our attention towards things that are personal to the body. Because AI is going to be taking all these things that we used to put our focus into kind of to, you know, it took our focus away from the body and onto like our work or whatever. And we end up with all these ailments. We smoke, we get fat, we get sick, whatever, because we're focused on these other things. When those things, we don't have to focus on those, then the focus is going to go back
20:15 on the body. And so I think the interest in all of this is just going to explode. Yeah. Interesting. Well, at the same week, OpenAI shipped ChatGPT add-on in PowerPoint. I was like, wow, a couple months too late. People use PowerPoint still? I don't know. This is dispatches from nineteen ninety seven. What is happening? As we experienced this yesterday at our kids graduation party, I mean, graduation at school,
20:45 like they couldn't play that. Was it a PowerPoint? Yeah, they tried to play a PowerPoint. It wouldn't play. It wouldn't play. And like Chris is like, who is doing PowerPoints anymore? It needs to be HTML presentations. I mean, I've been going to things for twenty years and PowerPoint's been failing. There's always some problem. Like at what point do we just recognize there might be an issue here? I'm glad PowerPoint is getting a glam up. And so it builds the slides natively, which is great. Google wired Gemini straight into Adobe,
21:17 Canva, and CapCut. So the friction is gone. You're not opening a browser to ask an AI. So now it's invisible layer inside the software you already use. So here's the deal. If you are using PowerPoint, Adobe, Canva, and CapCut, check it out with all the new AI that's been wired into it. It may make your life easier, but here's the deal. You have to actually go and play with it. It can have the best AI in the world, but if you're not doing anything... And it's uncomfortable in the beginning. You don't, you don't know what you're doing. You're like, what the heck is this?
21:48 But go try it. So, uh, and kind of, uh, the last piece of news over here before we move on to the, to the deep dive is meta. Speaking of daddy, daddy's a home and. wants his credit card back. Meta started charging for its apps, three ninety nine a month for exclusive features on Instagram and Facebook and two ninety nine for WhatsApp. So they're still free tier. No one freak out. You're still going to have your Facebook profile and all that. But they're just charging for exclusive
22:18 features. OK, there are certain things and you can go look what that means and get some more insight. But to me, what that means is that our A.I. bill is getting higher. I think they're spending like a hundred billion a year or something on A.I., And they need to recoup their investment somehow. And AS is just not making the cut. So that's kind of the news. And the other big piece of news, we're going to talk about it in more depth in a little bit because I thought it deserved its own little time for us
22:49 to drill into it. And more importantly, how can you use this new thing that Meta just did for your advantage? Do you have any takeaways from the news? Anything we missed? No takeaways, but I like to segue into the next segment. Facebook meta actually showing up in a new way for this new world because they have been the silent player. We know they've been doing stuff. We've reported on it for forty three weeks, but really we haven't reported on anything
23:21 real that we felt like people would use. Yeah, it's always been stuff that like guys in labs are kind of messing with. Maybe you hear about this, but like I never see any results. This week, we bumped into a story about a company that's using Manus, which is Facebook's embedded AI. Facebook has an ad platform. Manus is tied into that. So they're using Manus to... optimize their ad delivery, and then analyze their ads, and then also check social responses to
23:52 that to sort of increase customer engagement and kind of get a better handle on the business, especially in the efficacy of their ads. And apparently they've had really, really positive results from that. And this is the first real-world use case I've heard for anything from Facebook where they say, this is making us money. Yeah. So, and I actually, again, you can think about Zuckerberg all you want, but the fascinating thing that caught my attention as I was doing this,
24:23 so Facebook just made their groups, right, the Facebook groups, a thing. It's a new app. And basically, they are betting on having these groups that have been in Facebook for I don't know, probably from early days. But now they're making this, your community, a business asset. And here's why. So we all know Reddit, right? I personally, I've never posted on Reddit.
24:54 I've only seen like jokes from Reddit when people like posting weird things. But apparently Reddit has been used to train AI. So like OpenAI pays seventy million to license the content. you know, for, for their, for the LLM, um, Google, right. Did I say Google this morning? Google is doing the same thing. Yeah. Google, they're paying sixty million to license it. So basically like what I did is a nice little moneymaker for this LLMs. Um, I mean, um, uh, Reddit is a really good place for these
25:25 LLMs to get the data because it has real conversations. Well, I guess Facebook was watching that and like, wait a second, we have our groups that we've had forever that has a lot of conversations going. And when I was starting to kind of poking around, Facebook has like twenty five million groups. that are you know both public and private so facebook decided to make a move and they made the community a business asset and um there are they separated the groups into a separate app that you can download right now on on apple uh i think
25:56 right now forum It's called Forum. You download it on Apple. And you have to log in with your Facebook profile. And it syncs to your profile, but you don't have to check your feed. The thing about Facebook, no offense, I don't like going on Facebook because you get sucked in and you just kind of spend all this time. And I'm like, I don't want to do this. I'm not doing this. But with groups, I do have some really cool groups on Facebook. that my entrepreneurial group that I love kind of checking on that I was actually
26:28 excited. I was this excited that I decided to make a piece about this today. And I'm going to show you what it looks like. And more importantly, how can you use this information early on to go do something with this? So in the AI age, the only thing AI cannot fake is a real human trust. So Facebook now made your community more valuable, but they own the platform. They own the door, right? So when other companies are trying to MCP their way into everything and you have access to anything you want through MCP
26:59 and API, Facebook is the opposite. They are closing it. They're shutting it down. You only can access the community and all the conversations inside the Facebook. on meta through their own apps and their own systems. So basically like strangers are not welcome there. And which is interesting to me because it looks like a Reddit clone, but it really isn't. So first of all, what do you think about this move?
27:31 Well, I'm actually, I was flabbergasted when I saw this. The reason why is because we have talked about how Reddit is In the last few years, Google basically turned its search index over to Reddit. It basically said, okay, we're going to get, Reddit's going to be in the top three for every query under the sun at this point, because this is what, you know, this has actual velocity behind it. People talking about it. There's actual energy going into this. It's not some page that was set up to exploit SEO. They say it's like a more organic thing that springs forth from the internet and
28:02 therefore it's more reliable. That's the justification. Okay. Reddit's full of bias and all this other stuff. There's plenty of other concerns. But the bottom line is Reddit has become the darling of Google search results. That led us into the age of AI, and now AI's got to crawl the internet in the same way. What do you think gets surfaced first? It's Reddit stuff. So we've seen those charts about how, you know, forty-one percent of all the knowledge that's gone into AIs has come from Reddit, and that's far and away the number one source. Right. And Facebook looks at this and they say,
28:34 well, you know what? Groups are basically the same thing as Reddit subreddits. Right. Like little group conversations about things. We have tons of activity. Not only that, Facebook has a unique value proposition that Reddit does not have. Reddit doesn't allow you to businessify things. Your subreddits. They are expressly anti-commerce. You don't go on there and run a business. Facebook groups are business-centric. You know plenty of people who are running
29:05 actual businesses, and Facebook group is the hub of their business. I look at this, and I kind of LOL. I'm like, what a low-tech piece of crap this is, and you're on somebody else's digital fiefdom. You don't own your stuff. But at the same time, I also know people doing over seven figures a year on Facebook groups. So Facebook has a unique asset and they are now rolling this up in a way that makes it sort of like Reddit. They're like turning this into a real thing. I'm paying attention. Right. Well, and so here's – let's see how it
29:36 looks like. So I went in and I downloaded this yesterday. So first of all, there is the Ask button. So this is their AI where you can ask all kinds of questions. Second of all, this is what it looks like in terms of the forum itself. So you can – it's like the feed, right? It's like – so Facebook has the feed, but it's like all the people you knew in high school talking about their kids. Here, there's no – Forum, it's all group feeds. Yes. There's all group feed and you kind of can catch up with all the groups that
30:07 you're a part of. The, the, the interesting thing for me is like, this is all the groups that, um, uh, like that's actually my, my feet over here. But, um, I actually went there yesterday and I deleted a ton of groups that I'm like, Oh, I don't want to hear from them. I don't want to hear from them. I don't want to hear from them. And I kept the groups that actually, and I, and I need more cleanup to do, but I wonder like, all the groups now are going to get all kinds of different like things like, Oh, we haven't had this many members quit at any time. So the major cleanup is happening. But once I cleaned it up, I'm like,
30:38 this is actually something I want to read. And then I just asked the random question. I'm like, is the aura ring worth the price? Just to see like, how is AI is answering that question, right? Because that's what I use. I use this, um, um, ask, you know, beta, this is the AI, um, button right here. And it gave me this answer and it's quoting it and look at the groups that it's quoting from, right? Anonymous participant. This is the group. So you can go and read the comment if it's public. Cause that was my big thing. I was like, are they pulling this from private or
31:10 public? Or like, what's the deal? Like, how are they using this information? I was assured that it's only from public things available, like comments and to my questions, but I don't know about that. So we will see. So Olga, why are we talking about AI and Facebook and why does this matter? Well, guys, so here's the deal. They're making this move. Zuckerberg has been watching kind of been very quiet when it comes to AI, but now he looks like he's using this to actually make money, right? Not just for themselves,
31:41 but also for the people that have their businesses and communities and on Facebook, like what Chris said. So if you have this app on your phone and you can ask any question you want, guess who's going to be cited in the answers? People that are active in the groups, they're going to get this kind of information. You can go and join the group or whatever and follow more. And then you join the group. And then from there, now me as the owner of the group can have this person and market to them if I wanted to. So that's huge.
32:12 The other thing is like promoting any kind of service or event. Like I already created Practical AI group on Facebook to start posting content there. But why? Why do we care about this? Well, I went ahead and I looked Kind of what is the platforms? What do they even look like these days? Like we talk about Facebook, Instagram, LinkedIn X. Those are all the social platforms. But like who is where and how big is the audience there? So if you're watching this and you are thinking about how to market your product, which social media site?
32:44 or, or, or, or place you need to be thinking about, right? Like we're all lies on Facebook because again, like, I don't want to go in and, um, and spend too much time there. I do spend probably too much time on X, but then when you look at the numbers and you're like, interesting. So let's sum this up on Facebook, there's over three billion active users. That's crazy. Their typical user is around forty years old, slightly more man, every income level. This is the broadest and the oldest room
33:14 on the Internet. This is the male versus female kind of. This is the age group and this is the income across all brackets. So if you are selling to anyone like this is like this is a great place to be. You know, the B to C people know this. They have to be on Facebook. Now let's take Instagram. There's two billion active users. They're younger there, late twenties, men and women kind of equally, but apparently in US it tilts more female than male. And so many percent of their members there
33:45 is under thirty five. So if you're selling products on Instagram, this is a good place to kind of, you know, hang out. Now we're going to take LinkedIn. This is your money room. Apparently LinkedIn has one point three billion users. They're registered, but only three hundred and ten million. Look how low engagement of LinkedIn is compared to everything else. I mean, people spend time on Facebook, Instagram, X everywhere except for LinkedIn. But LinkedIn has the highest income of the
34:16 user over there. They're just I guess they haven't figured out how to hook you to come there and be active yet. So people are on LinkedIn. They're in mid-thirties, professional decision makers, and they have the highest income of any platform. So male is fifty seven, female is forty three. And then you look at X, that's four hundred to five hundred and fifty million active users. Late twenties, mostly men. Tech news, early adopters, loud, influential and small. So male sixty two, female is thirty eight.
34:47 Seventy percent are under thirty five. I look at this and I have a lot of thoughts, but I am curious, what are your thoughts when you're looking at this and how now the Facebook's move can benefit those people who are building in AI and market using any of these platforms? Well, what has happened here and what Facebook did this week, what Meta did this week, has defined the playing field, the surface of the new internet.
35:18 I'm not being hyperbolic. This is a total shift, and I want to unpack this. Number one, so who we didn't mention there was Google, but that's because these, like Facebook, Instagram, LinkedIn, X are social media places. They're like places you go to interact. Google, interestingly, probably touches more people than any of these. But notably, Google never established a social hub where activity took place. This is like a huge vacuum in the
35:49 market. And these companies have filled that over the past twenty years. And that's how this is what it looks like now. What's amazing is that when you showed the Ask AI inside Forum, That shows you it's only going to surface information from inside that echo chamber. Yes. It's not reaching out to the broader internet. This shows us the new game. The new game is Silos.
36:20 Ask whatever happens in there. It's going to generate answers and responses. This is a business generation thing. It's a lead generation thing. But if you're not playing in that sandbox, you don't exist. Yes. The same is going to be true for these other companies. This is what we're going to see. This era of broader SEO that touches everybody as an idea is over. It is strictly within each community. The broader SEO play has resulted in a lot of unnatural skews,
36:50 mutations that are not great and some business practices that are cumbersome and expensive, et cetera, et cetera. This new paradigm makes that cheaper and more direct. And it makes it clear, like I am operating here. Here are my people. I am targeting this. I'm not doing this broad targeting in hopes of like putting a bunch of lines in the water and hopefully catching some fish. It's more direct than that. It's like I'm using this type of lure to catch this type of fish in this pond that I know I'm fishing in.
37:20 Yep. So it is a more specified internet. And so this era of that broad open internet is officially over. It is officially over. This move will be so successful in terms of actual targeting, businesses will have to be flocking to it because it's better fishing. And so we're going to see this everywhere. And I'm impressed that Facebook was the first mover kind of was, was leaving them for dead. This is going to be real. This is absolutely going to be real. What is it still water runs deep. So, um, I am actually also my, my,
37:52 my interesting takeaway is that apparently men have more time to be on social media than women. I know I have to be the bad guy here. If we saw this by us only, it would look a lot more female, um, There's certain countries, we could just cut out one or two countries and the male percentage would drop dramatically. It shows the only thing that my AIs, when I was doing this research, that pointed out Instagram was the only one that was female dominant in the US. So I'm just saying.
38:23 So men just have more time. We need our laundry robots. Come on. They need to come ASAP. Maybe we're just lonely IRL and need the online social out there. This is interesting to me. I thought that was a very interesting observation. So, okay. So let's talk about what does this mean for all of you and what can you do with this information and be sort of the first one to do something with this. So the long game in here is...
38:54 So everybody on the web, like I said, they're trying to build MCPs, APIs. If the agents can't find you, you don't exist. That's like the web. This is like the AI paradigm. Like this is what we're moving towards. Well, Meta just closed this fortress down because Meta wants humans to use this, right? Because they're actual humans. They don't want, you can't automate this. There's no APIs. You can't really get in and leave comments and do whatever you need to do. Not with an outside API, wink, wink.
39:27 Well, right, their own AI. I'm talking about the whole world of genetic AI. So it hoards the conversation data for its own AI. And so what it adds up to, they're building the world's largest private human conversation data set. It's a lot bigger than X. And people there talk about all kinds of topics. Because sometimes I feel like X is not very friendly for like normal humans. And it's like, who's like the loudest and the biggest and the greatest? And look at me, look at me. Facebook is a little bit more chill.
39:57 um and there's a lot more like people like normal people i guess over there i don't know and uh meta is gonna have a lot of data what people like real people think about and no one can really copy it because they've got that audience that no one else has There's been a huge trend since twenty fourteen companies that try to gauge real time market sentiment. That is becoming more valuable every day, especially as we move into a bunch of AI opinions. That's not what people, AIs don't spend money. Humans spend money.
40:28 So to suss out real human sentiment is getting more valuable every day. This is a way to do that more reliably. You shut the door. Fascinating development. I think it's something to pay attention to. Real-time sentiment, real human sentiment is a valuable business tool. Yeah, so Meta's forums ask that AI function starts replacing Google for community questions. And if your community isn't there, you're invisible to that layer.
40:59 Communities become the new search. So if you take away nothing from this, and if you have a business or anything, and you've been sleeping on Facebook, now you have this real thing with groups that Right now, it's more quiet. They're quietly launching this. But you can go and start playing with this. But basically, be there first and start structuring your data so AI, Google's, Facebook's AI can start finding your answers. I'm going to start playing with that, with practical AI in our group.
41:30 And I want to start reaching more people, right, on Facebook and kind of what we're doing here. Um, now I can do that because group is a lot more manageable than the whole Facebook app. And so the, also the free internet keeps cracking. Um, like we talked about the premium tier that Facebook and Instagram just made up. So they are mining your fuel and charging you rent and The bad thing about it is that it can still be canceled. You don't really own anything there. But the goal is to try to bring
42:01 as much of that audience to your ecosystem, to your site, something that you own, to your email list, to your whatever. But use that forum for good. We have a real problem. Anyone who's a seasoned operator on the internet understands that getting real customer attention is harder than ever. It's harder than ever. If it's not harder, then it's more expensive. Depends on how you see it, but only real Gs are going to see it as more expensive only. It's harder to get people to even take
42:31 a look at what you got. It's harder to go find your people, to find your buyers in a way where they're actually going to connect. This is clearly a pathway for doing that. But don't you think they're kind of like, I guess they're playing a different game because if open AI, if cloud, if perplexity cannot find the information that it's looking for inside the forum, right, inside the Meta's forum, then all of that geo-optimization inside the Meta is only good for Meta.
43:02 It's not good for anybody else. Yeah, but if Meta's where the people are, what do you think happens? That's true. Very true. Fascinating. Anyway, two different games, you guys, playing two different games. So, GEO is great and all until you're meta that has three billion users and you can do whatever you want. These are all different prongs of a good strategy, digital penetration strategy. You need them all if you're going to be operating on all cylinders. Yeah. But the people one, you can't ignore it. You can optimize for AI all day long and it won't mad up to anything because no human's pulling the trigger to buy stuff.
43:32 Yeah. Well, basically, anybody who's sitting on real communities, probably going to copy this within a year. Reddit, Discord, anyone with a community, Bolton, AI, answer a layer on top. And, you know, maybe that will happen. I don't know who else has this big community. A lot will do it. Maybe LinkedIn would be the next thing. I don't know. But LinkedIn is a completely different ballgame. I'm very curious to see what they're going to do. So anyway, the big takeaway, everyone's teaching AI is on X talking to
44:04 four hundred million young guys. Your customers are the three billion people nobody is really showing up for. So your communities became a business asset. Go ask yourself a question. Which room is your audience actually in? And go engage in that room. And maybe Facebook just gave you an interesting key to reach more people because go-to-market sales and reaching real people is going to be more and more difficult. And Meta just gave us the key to that, to their fortress. Come and play.
44:34 Anyway, any other takeaway on this before we move on? Meta, back from the dead. I know. Who knew? I am, you know, very intrigued by that. It's a savvy move. It's one of the smartest we've seen, I think, in forty three weeks. Yeah, I hope that I hope they'll keep doing that. So next thing I wanted to moving on to the next part of our program um is the demo that we promised you guys this is the the worldwide premiere of um what if you could just talk to
45:06 your website so this was my dream this is what i've been uh Asking my husband for a long time. I was like, okay, PageMotor is amazing. That's awesome that you can do that. I love that my website runs on it so fast. But when can I start talking to it and it will start doing things for me? So this week, I connected Claude to my live business site. Then I built a real page on brand in minutes without opening anything. And here's what happened. So first, I'm going to show you guys the actual...
45:38 video it's just under two minutes and then we are going to talk about what this actually means because this is the future of like you're just talking to your website and it does things for you it So I am witnessing some magic over here.
46:09 I got API working with my revenue hire site that's running on page motor. And I am just talking to it, to my site, that I've been opening anything. And it's created a page with my newsletter. And now I said, hey that's great but there's nothing on the main page that leads to it and it goes well i think here's what we
46:41 need to do we need to create an insights on the main page that will link you to the newsletter i am blown away so let me go show you this page that i already created that can do anything so you guys look at this I just said to Claude, hey, go find my newsletter, first edition that I posted on LinkedIn and create a page on the revenue hire
47:15 page with my newsletter. And it just did it. I didn't have to touch anything. I didn't have to do anything. And look how beautiful it is. The formatting, it matched my homepage. And I did the good job doing that. It has this beautiful call out. It's kind of crazy. So this is a very quick video of what I did for like last two days, but you just kind of see, I'm kind of not believing what's happening
47:47 in my face. And what I wanted to show you guys is the fact that clearly My being blown away with all of this is I'm trying to add. I was going to walk people through what actually happened there. Yeah, yeah, yeah.
48:18 So here's what happened. That's the first screen that I was showing you, right? This is my cloud window. I connected my API from revenue hires that's running on PageMotor to my cloud, to my CLI in cloud. So this is my working thing. So I'm working in this cloud window and I just asked in my terminal, I said, find my newsletter and put it on my site. No code, no page builder. I'm talking to it. I'm talking in the window of Claude, right? You can talk in OpenAI. You can talk in any AI.
48:48 You can bring your own key. It doesn't matter what you're talking about. But the key here is that I'm not in PageMotor's admin. I'm not in RevenueHire's admin. I'm in my Claude talking to Claude about RevenueHire's site. And I said, hey, I want you to add this newsletter and put it on my site. So I just asked, right? Then it goes, okay. It went ahead and built the page. All right. It matched everything from my homepage, my navigation, everything, the theme.
49:21 It put the newsletter in there. I had to do zero design, zero toggling. It even wrote a short summary up top so AI search can quote it. So it's geo-optimized. And I was just like, wait, what? So this is it? It took me like four minutes to do this part of this. And then, like I said in the video, I was like, okay, great. I have this newsletter now on my site, but there's nothing in navigation. So, you know, like before, it was just this thing from about testimonials.
49:51 And Clyde was like, okay. He went and looked at some, you know, at all the brilliant documentation that Chris wrote. And it said, let's add another thing in your navigation page. in the menu that has insights that will lead people to your newsletter and all other newsletters. It gave me suggestions what I need to do, but more importantly, it went and actually created this thing on my main page called Insights. So it edited the page. It edited the navigation menu.
50:22 And I, you see my reaction. That was like the first night I was playing with that. I didn't even know like what to say or how to express myself because it was almost, it was almost like too hard. It's like hard to believe that it's true that this is this easy. So I never had to micromanage anything. And in fact, I just tell her what to do and it goes, reads the documentation, figures out on PageMotor what to do and just does it. So, yeah. Now you can run your website by talking
50:53 to it, and I'm not even a developer. So all of that said, now, Chris, this is your time because you're the builder who built this. What are you going to talk? What can you say about my presentation here? Let's just run through it real quick. It's useful to kind of see what happened here. So here's the page it created first. And like she said, it GEO optimized it. It added some stuff. When it created this page, it added all this formatting. Like her newsletter on LinkedIn does not have any of this native formatting
51:23 features. So the AI looked at this and says, you know, this is probably how it should be formatted to be most visually interesting. So it added this stuff. I've been working with people for twenty years selling software that people use to do stuff like this. People generally are not that good at adding formatting. They don't know what to do. They don't know what's going to look good. They kind of don't do anything. The AI is better at this than the average. Much better at this than the average. It added stuff that was cool. It added a link here at the bottom for SEO stuff and also just for driving traffic around and working with attention
51:53 better. This is a really good page. So, Olga gets the page built. And it's built, like I said, from her email. So she sends out this thing on LinkedIn. That's the content. It would be nice if this was echoed elsewhere, put on her website to increase her footprint, something. But in the real world, prior to this, you'd have to do all these steps. Hold on. I want to point something out. So my site was built in page motor. So my revenue hire site is built there. Now my goal, my next goal is to optimize it so it's citable and searchable by AIs.
52:25 So my prospects can find me because I want to really capitalize on this AI native world. And my site is built on AI native platform. And so I was like, I already have a ton of content. I just now want to put that content into my website in the most geo-optimized way. But I don't want to do a lot of work on that. So this is where this API is coming in so quickly. niche and nice in this because I literally took content that I've created in the last couple months. And now I'm putting this, but guess what?
52:55 The second I put the first page like this newsletter, my second one now is a skill. So it's done it in like even less time. So it's going to be less and less and less time because now it knows exactly what it needs to do. So that's just one little step from this week that I am doing to make sure my site is discoverable by AIs. And I'm literally now talking to Claude on what else I have to do with my site to make it even more discoverable. And I'm just going to do whatever Claude tells me to do because it's just going to do it for me. I will be quiet now.
53:26 Let's keep walking through this. So step one, this page gets created, like I said, from her original asset. She only had to do the LinkedIn newsletter, and then now suddenly the AI has turned it into an asset on the website that has SEO, GEO features, so it's more discoverable. That's a pretty big win. But then she goes and says, well... This isn't linked from anywhere on the website. What are we going to do? The AI properly recognizes this as a new pattern. You're going to be doing a LinkedIn newsletter all the time. I'm going to be creating more of these pages. We are going to need a hub page
53:58 that links to all of these. So it's called this insights. So it created this insights page. Again, GEO optimized, and it automatically adds these links sort of like a blog would work. So in the past, a blog would work automatically like links on your homepage. You publish a new one, and then it bubbles up to the top. We don't need that capability anymore. The AI will bubble this stuff up naturally. So now the AI has created this insights page and is going to keep it fresh
54:29 with links to the latest newsletter pages that it has created. And then it's like, well, we need, so now we have the hub. Now we need this accessible from everywhere. What do we do? We add this insights link to the navigation menu and now the entire puzzle is complete. That is absolutely unique. And she did this all from the AI prompt. She wasn't going into the PageMotor admin. She wasn't making decisions. She didn't have to make the decision about how this stuff should be organized. In fact, when it offered me the ability to make
55:01 insights out of the admin page, panel, I said, well, why don't we make it a newsletter? And it goes, I actually recommend against newsletter because people have biased about newsletters because they think they're going to have to put their email in there. If you call it insights, people most likely read that. And then you can put more than just newsletter. You can put other articles and other research. And then you can still have people kind of get on your email list later, which I'm going to work on that later. But I thought that was very interesting insight that it gave me. So it's like more fresh and new in
55:32 terms of what not to do on your website these days. Yeah, no question. I think this is a pretty compelling result. I mean, obviously, we're just talking about content, not other stuff you might want to do with your website, but... In terms of building out a legit content base that's searchable and discoverable, here's a way to think about it. Obviously, this is a great technique for that. This is unreal. You didn't have to do anything. It did the thinking. It did the connecting. That's beautiful. Yeah, nothing. A lot of businesses can employ this exact
56:03 playbook because you have your top three customer avatars that you want to target. Well, if you don't roll out a strategy like this to target them immediately, you are missing out. Like this is not that difficult to do. You can deploy the AI to do it. You don't have to sit there and do all the research and create all these pages. That's probably the friction that was holding you back in the first place. Now you basically craft a single message to target the avatar and the AI can spin that into all the assets you're going to need to be able to capture attention.
56:34 This was supposed to be for next week that I was going to talk about. But, yeah, this is what I'm going to do next time. It's like I'm going to basically think through all my customer avatars, and I'm going to create a separate page for all of them as a landing page. And I'm not going to do much work because it's going to be done for me, and it's going to be all geo-optimized. And I'm just super pumped because, like, I think the first thing when Chris did this, I was like, wait. Well, it was easy to connect it. That's the first thing. Second of all, I was like, what – I think the mindset was, wait,
57:07 I can tell my site to do anything I want. what do I want? I think that's the biggest thing is you don't know. Imagine you can have your website do anything for you. What are you going to ask it to do if you don't have to deal with formatting, if you don't have to deal with plugins, if you don't have to deal with any technical debt, if you don't have to piece things together and duct tape it together? what, like this, look how beautiful this is and like how easy the process is.
57:38 Like, what will you do with your website when you can do anything and it can create things for you? Right. This is just content and navigation menu and all that stuff. I'm talking about forms. I'm talking about like the behind the gate, um, different reports, maybe for your clients or for other things. Like what, what kind of business platform, what kind of operating system can your website become when you literally can talk to AI and AI creates anything you want on your website that's in your brand in your format and you can accomplish whatever
58:09 goals you you you want to accomplish so that's the new paradigm i think as a business owner is changing how we used to think about the website like it's a pain in the ass um Instead, actually, it's your assistant, it's your agent that does things for you, not sucks your energy, time, money, and whatever else. So that's a big paradigm shift for me, and I like to work in paradise, like my husband says. Yes, indeed. So I want to make another point for
58:41 the skeptics who are watching who know a little too much. They're looking at this thinking, okay, great. What about some hardcore SEO stuff, like schema data, all these overlays, like open graph stuff, the stuff we're going to add to really make a complete package to target, like if we're really professional, we're doing all this stuff. That's not on here. Olga's aware of some of this stuff. She asked the AI, well, what else could we do to further optimize these pages? And of course, it comes back and says, well, you need some more robust schema data. You don't have, for example, an organization schema describing what
59:12 your business is, what customers you want, all this type of thing. Like that would enhance these pages. And she was like, well, can you add that? And it's like, well, it doesn't look like there's any way for me to do that. I could do this and that, this clunky, you know, way of adding this to your pages. Do you want to do that? And I interjected at this point. I was like, no, we don't want to do that. I know that PageMotor is capable of doing these things. And, you know, so I'll set about, you know, solving that problem for her. What is interesting though is the paradigm
59:44 here. With WordPress, you would pay for some fancy SEO plugins such as Rank Math or one of these other ones that enables you to add a lot of this information. It's like a professional uses it. They'll put this on a site and then the professional will go in and make sure all the data gets to where it needs to be. That is a huge burden. Not only do you have to get the plugin, you also have to know how to operate it like a pro. So usually you'll have the plugin or you'll retain an agency and the agency is like, oh, you need to buy this too and then we're going to do it for you. Great. The AI can do all this stuff.
1:00:14 And with PageMotor, the AI can actually create the schema data and stick it into your pages exactly how you want. It doesn't matter about the technical details of it. The AI is actually going to create a plugin for your site and do it. It'll all be bespoke, made just for you and your use cases. But the bottom line is all this stuff, all these layers that have been abstracted out into professional agencies, that have been abstracted out into complex plugins that cost you a lot of money on retainer. All that stuff is disappearing.
1:00:45 There's just no need for it. The AI can simply add it with the same agility and panache that it used to format this content that it added to her site. It knows what to do. It's got pathways to do it. And with PageMotor, the AI has everything it needs to say, okay, I know exactly how to run your site. I know what needs to be added. Oh, now I know how to add it. All these questions are solved. These problems are solved. This is like you will be talking to
1:01:16 your website. You'll be running some agent in the future that's imbued with extra knowledge. It's like your website web dev pro, and it's going to say, oh, you've got these pages, but they need these enhancements. And it's just going to make them. Yeah. So I think the paradigm shift here is we're going to move away from plugins into more of the skills. Because I think once, you know, like my GEO skill, right? Once I figured out what that means for me and my business, I don't have to wait for some developer to develop this or anything like that.
1:01:46 Now it becomes a skill. And I have a feeling that those skills will be kind of unique to me on my website. And then they're just going to keep replicating themselves or using themselves to kind of give me the outcomes that I actually want. So I don't have to connect a bunch of things. And more importantly, I don't have to pay a gazillion dollars for different plugins to do the work that AI can just build for me on my operating platform, such as PageMotor. Because PageMotor was built as an AI native first. So there's none of this AI bolting onto
1:02:19 the very old dinosaur that's duct taped together and can barely hold on to what's going on right now. Plugins will still exist, but it's our focus that will shift. You said it exactly right. Your focus is going to shift to the skills that the AI has to run your website. It'll be creating plugins, but you won't care about that. You're not going out shopping for plugins. The AI is building what you need right there for you when you need it, which is so elite. The best thing is it's a self-perpetuating system. The AI is going to know how to build skills for your website. You will define the skills that you want.
1:02:50 Like, hey, I publish a newsletter every Sunday. I want to make sure this is on the website by Monday. I want to make sure it's SEO optimized and I want to make sure all this stuff is tied together. Well, then the AI is going to say, let's go ahead and build a skill for this on the website so that we know. Then every time it connects, it's going to say, all right, here's what we got to do. We know exactly what to do. You can tweak those processes over time. You can say, hey, I want to make a shift to how we do this. They'll say, okay, I'll go edit the skill. So more to come. I will be sharing my journey with this because I am literally reimagining what my website can do for me,
1:03:21 not what I can do for my website. What can I do for me? And I'm going to reimagine this every week as I'm going to add more and more things because this, what I'm seeing this, the way I'm seeing this right now, this is like a train station with all the kind of agents, like little trains that will come in and out. And then just going to go and take me to all kinds of places I want to go. to all kinds of customers, all kinds of prospects. And I am, again, this is such a different mindset. It's like a hundred and eighty. Like what is happening? A very different, very different than ever.
1:03:53 So we're not talking about a UI here. You're talking about knowing what you want. I want a page for this. saying it, you express it to the universe. I want this page. And it just happens. I shared this story in the email. If you didn't get the email, if you're just watching the show and you didn't get the email, it's worth sharing because this is so funny how this has just totally shifted. I got this customer who is a pain in the ass. She complains. Basically, all her problems are the same. She doesn't want to have to read any documentation. She doesn't want to click through the UIs.
1:04:23 She doesn't want to do squat. She just wants to complain and have some knight in shining armor come and solve her problems and shut her up. That's what she wants. That's totally unrealistic. You got to pay for this kind of service. She doesn't want to pay either. Like I get it. I get that having to confront a new technical challenge is not the nicest thing in the world. It kind of stinks to have to do it. So that's annoying. That can be frustrating. It can be intimidating. She's not going to do it. But she's also not willing to pay for somebody to help her. It's hard to find good help. Maybe they're not going to stick around.
1:04:54 They're not going to be dependable the next time you want them. That stinks too. These are relatable problems. I don't falter for them. But this idea that some knight in shining armor is just going to show up and fix all your problems is ridiculous. Or is it? Not anymore. Or is it? I think the knight in shining armor just wrote in the door because now you can just complain, oh, AI, I need this fixed. And you know what? The AI is going to actually fix that damn problem. You're not going to have to wind to the universe. It's just going to happen for you. I can't believe it. And the thing is,
1:05:26 with PageMotors API and AIs, you're not just talking to hallucinating AI that will make up stuff for you. You're talking to AI that's been coded for the code that's been done for the last eight months that's frameworked with all kinds of documentation that's so extensive of what it can and cannot do that it literally has to operate in this framework. So it's not just going to promise you the world. It's actually going to deliver you the world because that's the big difference
1:05:56 between framework vibe coding and the naked vibe coding is because one gives you the outcome you want. The other one just gives you a lot of headaches of like, okay, I got this pretty shiny picture. What do I do with it now? So anyway, one, one teaser, we've got this nailed down to the point that by next Monday, Two days from now, three days from now, whatever. PageMotor is going to have a set of instructions for the AIs that's super comprehensive, like a menu. It says, if the user wants to do this, here's the stuff you're going to need.
1:06:27 Here's the exact documentation you need to go to. So it uses the least amount of tokens possible. It's saying, if this, then this. A huge list of instructions for how to do everything you can imagine with your website in the most efficient possible way, like... Wow, this is the real deal. This is amazing. This will not only cut down on hallucinations, it will render them literally impossible. So this is exactly the circumstances you need. Here's exactly where you need to go. Don't do too much thinking. Saving tokens. I mean, this is the real deal. I am in love with my page more
1:06:59 site. Anyway, join us on the journey. All right, well, on this note, let's keep moving. We've got one more kind of deep dive to talk to, to talk about, and then we've got our funding. Is this Greg? Yeah, so we've got the... I've got the tweet. You want me to pull it up? Yes, pull up the tweet first. All right. So Greg Eisenberg, who we have talked about, he has a great show. He's kind of like the main show for showing off characters in the AI space. He's got a YouTube show.
1:07:30 He was in San Francisco last week and he did this big post and he says he came back and I feel inspired. He spent five days with Frontier AI model teams, startup founders, and three billionaires. And he gave this list of seven team takeaways and And we thought these were spicy enough or at least relevant enough that we should highlight a few of them on the show today. Yes. So, um, yeah, we're not going to go through all of them, but basically, like I said, in the beginning, we are moving away from,
1:08:00 um, we're moving away from like, who's got the best model and all of that. And we are moving towards the operational, like who is a good operator with AI. So that's kind of where we're moving. And, uh, not to stage you into remove yours. All right. So that's Greg Eisenberg. Follow him. He's awesome. This is his In the Trenches thread. It has over five hundred thousand views. Lastly, Chuck. And we pulled the seven that changed what
1:08:32 you do. So the first point that he made is he said, I had lunch with three billionaires. All of them are buying SaaS companies and rebuilding them agent first. So. This is kind of what Chris is usually talking about. This is the indie version of what he's doing with PageMotor. He didn't bolt AI on it. He rebuilt it agent-native. So the billionaires just have more money to do it with bigger SaaS companies,
1:09:04 and they are rebuilding them agent-first. So what do you think about that? What does that mean for... for normies like us, is that should we go and look for smaller companies that are SaaS companies and try to rebuild them agent first? Well, I love this particular statement. What can I do with this information? Like, what do you think? Yeah, yeah. I love this statement from Greg because it's the most San Francisco coded thing I've ever seen in my life. I had lunch with three billionaires, all of them were buying SaaS companies and rebuilding them agent first. Guys, when you see SaaS companies,
1:09:35 just replace it with plumbing companies, HVAC, electricians, Handyman, whatever, any sort of oil change places, any kind of down-to-earth things that you come into contact with all the time. Replace something high-level like SaaS with something like that. The reality is the same. Every one of these businesses needs a tech makeover, has for years. We've been talking about it, but it's never really been this thing that had this full package that was super efficient. AI is the unlock.
1:10:06 We are now in that time. So rebuilding every company from this perspective, every type of company from this perspective is going to determine the future. Yeah. So these are the trends that I wanted to talk about in this segment. And Cody Sanchez actually talks a lot about that. That's pretty much all she talks about these days is that, you know, in AI era, the plumbing industry, the owners of the plumbing company sleep the best because they know they're always going to have things to do. But the other person who sleeps really
1:10:36 well at night is the plumbing owner of the plumbing company that implemented AI and agents in his plumbing company. So that's the deal of this. If that's what San Francisco billionaires are doing, what can you do today locally with some of the companies and how can you rebuild them agent first? Now, moving on, I heard the phrase agent debt for the first time. This is what Greg says. Hack an agent workflow together fast, prompts conflict,
1:11:06 the memory gets polluted, tools overlap. Six months later, the agent is doing weird things and nobody knows why. Agent debt. So what's your reaction when you see that? I love this. It's so true. Here's why. When you work with AI and really get an embedded workflow, what you are doing as part of that is producing a lot of markdown documents for session logging, for conclusions. We're saying, oh, this is now a working best practice, that kind of thing.
1:11:37 Typically, what we see, especially if you're operating in a few different domains, you get a lot of these files. People are not really super disciplined about this and saying, okay, we have three master files to dictate everything and we watch them like a hawk. It's not what they do. When you have a bunch of stuff looking in a bunch of different places, you've got all this information. There's no one source of truth. Olga knows this phrase well. One source of truth is becoming a very important maxim as you work with AI. With AI agents, for sure, yes. You need a single source of truth. A lot of the people who have done
1:12:08 this work in the past six months, or six months to a year, there has been a real taste for speed. And that speed has come with this call. Ship fast, ship fast, ship fast. We're going to deal with everything later. We're going to deal with everything later. And then later you get agent debt, which is basically a new fancy term for tech technical debt, which is all of these companies that are dealing with right now, trying to bolt AI on it because they have the technical debt. So with agent debt, the solution there is in my opinion,
1:12:39 know what your outcome is. You're building towards and be and simplify. I'm literally this week. my mantra for all my agents I'm talking to is please simplify, simplify, simplify, simplify, because otherwise it's just creating all those things. They'll create anything you want. But then like, once they try to actually connect and do something, it's just like, you've got a pile of thing here. You had a pile of, you know, think here. So like, you got to have the outcome you're working towards and then make sure you kind of stay disciplined on that. I don't care where you're working with AI,
1:13:11 in what way, what your business is. I don't care what it is. Everything about it is exactly the same. You have got to get to the most efficient solution. So like in computer science, we have some things called like DRY, don't repeat yourself. You know, you need to be using objects. Don't like have a bunch of code that essentially does the same thing. Everything's, you know, a single source of truth for everything. If there's a save, it all goes through this save. They all work. Everything has got to work the same way and borrow from the same components. That's like the most efficient way to build a system. That is true for everything now.
1:13:42 Anything AI touches must follow these operating principles or it's going to spiral out of control given enough time. Because you can build all you want, but when you try to operate something and something is just not clicking, you just have to go and try to trail and figure out what the problem is. Cleaning up messes is the most expensive thing to do with AI. We call it code refactoring. I know this from personal token usage. If you have to clean up a big mess, that costs you the most. And now it's not just costing you tokens.
1:14:13 I'm exhausting your tokens. Anyway. All right, moving on. MCP came up in literally every conversation, Greg says. The ones exposing their product as MCP get pulled into deals they never pitched. The rest are becoming invisible to agents. This is the new SEO. If agents can't find you, you don't exist. Vindicated again. Yep. So Chris made this bet about eight months ago, two websites, one for humans, one for agents.
1:14:43 And now it's not a trend anymore. It's becoming the foundation. Right. And what are your thoughts? What's your reaction on that? Unless you're a meta that can block your garden off. What are your thoughts on that? That will still even be important to meta. They'll establish some special connections where you get in. You know, like if your MCP adheres to these rules, we'll let you in here. We'll give you access. It's going to come up, I promise. This is the, it's sort of like having an all access pass.
1:15:14 If you want all access pass to the AI future, you need an MCP. It's as simple as that. Well, I mean, this is businesses that actually, like PageMotor software, it needs an MCP to be legit. Right. It's like a door for AI to come in. You actually build the MCP. What is your experience of very good? So this gets better. So if you use PageMotor, Like Revenue Hire, your site uses PageMotor. Your site now has an MCP. It has both a public-facing MCP that Google's AI bot can come look at and
1:15:45 learn things about your site. And it also has an authenticated MCP where you can be connected as an admin and actually do incredible things with your site. So it has a multi-layer MCP there. So it's not really PageMotor's MCP. It's your MCP. It's revenue hires. So any business using this gets the MCP. I said you need an MCP. You might ask, well, how the hell am I going to do that? You just use PageMotor and you got it. Well, right. But for those other technical companies, like basically,
1:16:16 how hard is it to build MCP for the AI? It is actually quite difficult because it is a fundamental architectural question. An MCP is going to work hand in hand with something called an API. A lot of companies have APIs, but here's how they built them. They already had a business, they already had software within their business, and they layered an API on top, doing the best they can to try and get it to do what they think people want to do. The difference is if software is built AI first, agent first, it's data first, which means it's API,
1:16:46 it's naturally API first. Which means it's like the API essentially should be, in theory, it should be everything the software is capable of doing should be in the API. That's not how ones that get layered on work. You can never capture everything it's supposed to do. So if you had built it that way, though, then laying on the MCP is a very natural next step. But when you've got the API bolted on and then you've got to do the MCP on top of that is a much more complicated setup and it's never going to be as good.
1:17:16 But you still got to do it. Even if that's the card you got, that's what you got to do. But it's never going to be as good. Got it. All right. Next trend from Greg's reporting in the trenches from San Francisco is which model do you use as being replaced by which model for which task? Model loyalty kind of feels dead. What is your reaction? True, like over, like what do you think? I've been saying for weeks, months now that the model really doesn't matter.
1:17:47 If you can do, like if you've got the workflow that you like and are getting the results that you want, you clearly recognize that AI is enhancing this. Like I posted this week, I kind of did a little analysis of my time and input. And I think AI is giving me leverage on my actual work. to about twelve x to twenty x somewhere in there depends on the task twelve x to twenty x that's incredible leverage i mean what i could probably get the same twelve x to twenty x with codex six
1:18:18 seven apples bananas here it doesn't matter it's like whatever you like is what you need i don't need crazy compute i don't need agi i don't need any of that i just need a workflow that's AI enhanced that works for me that's why the model doesn't matter use whatever you want And then for people that are kind of starting with AI and all of that stuff, just kind of start, like, use any model first, and then you'll quickly start figuring out, like, hey, what's better for this? Like, Grok is best for this. OpenAI is best for this. So, like, this is kind of,
1:18:49 but then I feel like it's, to me, it's kind of sort of true because you're still going to have your main model that you use. Like, I'm seeing people, like, you know, they're becoming more codex driven. Like the majority of their work is done with codex and then they use different AIs for different tasks. Like I use mostly Claude, but like, you know, like you can have like the main model that everything else kind of works around, but you know, That's an interesting point that he noticed. Moving on. Next point was frontier model companies can see API calls and token counts.
1:19:22 They can't see the actual workflows. If you're deep in a niche, that's incredibly valuable. Usage intelligence is the new alpha. What do you think? I mean, I think that's kind of a sexy thing to say. I'm not sure how true that is. Like, does it really matter to Anthropic that I have developed my own workflow that is plan, execute, review and revise? Like I have a very clear four-step workflow with AI for everything I do. Is that really valuable to them? Yeah,
1:19:52 but I think what that means to me when I read it is like for my, let's say I have a sales recruiting company. So I have certain workflows that work for me. That's my niche. So now I can use all these models and capitalize on something that's niche, the workflow that I know very well, and I can use agents on. Anthropic, they probably don't know that well what I've learned over the last certain years in my business and my workflows. I think that's what they talk about. They can see what API calls and token counts, but they don't really actually know how are you putting this workhorse together.
1:20:23 And that's why these companies that don't currently have AI, this AI implementer is the one that is going to be getting a lot of cred these days because they can actually figure out how to switch these workflows and make them niche-ified. I think I would add to his statement. I think the agentic workflows are super valuable. I don't know if my terminal workflow is particularly valuable to them. Yeah, I think to me that just meant the business workflow that you're working on, whatever. If it's accounting, if it's recruiting,
1:20:56 if it's plumbing, if it's whatever, it's like you've got your own workflows. But I don't know. Maybe it means something else because I'm not a technical person. Or so I say. Fun things to say in San Francisco to sound important. Yeah. The next trend, the hottest role in San Francisco, the forward-deployed engineer, the person who sits between the agent and the customer, making sure everything actually works. So we covered this in our episode thirty-three, AI transformation hire. So I think it's kind of more,
1:21:26 it's a sexy, of course San Francisco has to name it something with engineer in there. But what do you think about that? Like somebody who understands the AI and the customer and how to make the customer happy with AI. Is that kind of how you understand this? We have a common business associate here who is maniacally focused on exactly this. Like he knows AI is the thing, but he also knows the real world. And he's like, this has got to, he's just solely focused,
1:21:58 all he cares about is how this can be deployed to the real world in a way that is not overtly technical, which is what AI seems like now. I think that's really the question for the future. It has to be brought to the masses in a way that the masses will accept. Because we have this bubble, big bubble on X, where everybody knows what's going on. You go to Facebook, you go anywhere else, Who knows our open clients? Who knows? Who knows? There's the gap between normal people in
1:22:29 the everyday life and like people that are like, you know, AI enthusiasts. It's like becoming so divisively big that we need this kind of forward deployed engineers. I picture this twenty something year old enthusiastic AI salesperson showing up at the fictional plumbing business. Okay. Are you a Claude person? Are you a Codex person? And the guy looks confused. He's like, Buddy, I'm just a person. Exactly. Anyway, and then the last trend that I want
1:23:00 to point out in here, walking around the Mission, the street-level businesses, the taquerias, the barbershops, the laundromats, none of them use any AI at all. I only like to eat tacos that are AI-enhanced. Well, I haven't seen anything yet. So that's what I'm talking about. The gap is the opportunity. It's everywhere right now. So don't, if you're like an ex, don't get all this cloud pilled or AI pilled, like, oh my gosh, everyone is doing this.
1:23:30 No, I mean, the shop on the corner uses none of AI. Like there's so many businesses out there that haven't even like They don't even know how to start. They're overwhelmed because the things are moving so fast they get whiplash. So I thought that was a very cool thing to end this segment with because there's a lot of opportunity right now. You're here learning about AI, how to use AI, how to build with AI, how to sell with AI. Now go out there and do something with it. So who is... The forward deployed engineer.
1:24:01 Who is that person? That person is essentially like a missionary, a religious missionary who goes out and spreads the gospel. Yeah. Instead of spreading the gospel, you're spreading a workflow that will work for normal people. Right. Your job is to figure out what that looks like and how that's patched together. If you can deliver that message, they will convert to your religion and you will make money. Yep. That's the message of the segment. It's a very interesting report.
1:24:31 I love Greg's stuff. He's so down to earth in kind of how he communicates this. And it's all over. You don't have to be in San Francisco. It's everywhere. So go find your niche, something they're passionate about, and go become the messenger. of how do you bring that? But I also think you need to have a certain skill set. Like you can't just be like super, you know, nerdy about it. Like you also need to understand the customer's pain. Right. Because I think this is the big thing with AI. You can be the biggest AI enthusiast and like talk about until the cows come home
1:25:02 how amazing that is, but you're not connected. Like, You know, in sales, if there's no pain, there's no sale, right? And so, like, if you don't know the pain of the customer, like, what are they trying to do with this? You can, like, vomit all your information about AI, but that's not going to pay the bills at the end of the day. So that's kind of the big message of that. Any other takeaways before I move to the funding? No, let's have some funding. All right. So funding, twenty six weeks. Look at this. Anthropix deal made it go way up.
1:25:35 So we're like, I can point out this was Glock. This was OpenAI. Now this is Anthropix. So we're just, you know, up and down, up and down. What was the total share of venture funding this week? It was down. It was the graph on the right, top right. So this one, yeah, that over ninety percent was. Yeah. Oh, my God. Was AI. This is insane. Right. Like, look how how about a share of AI it is. And then let's go and look at our
1:26:09 funding. So, fifty-two point three billion went into AI this week. Fifty billion went into one company. Again, you're going to see some reports about sixty-five billion. It's the total that Anthropic raised for this round to make sure that they're valued at over nine hundred billion. But here's the numbers from Crunchbase. I like everything to be verified and I did a million times. Ninety-three percent of all venture dollars went to AI. Fifty-six companies out of one hundred and sixty-seven got AI funded.
1:26:42 and um two point three billion is the AI funding excluding anthropic look at that though a hundred and eleven companies that got funded received only seven percent of the actual total funding amount that is insane um so AI infrastructure compute fortune companies got the most anthropic is nine nine percent enterprise and development tools uh one billion that's two companies other AI applications Eighteen companies, robotics and physical AI, six companies,
1:27:13 healthcare, biotech, nine companies. And then fintech is five companies. So see, it's interesting to see how it's now kind of being broken down into industry, which I like that, just to kind of see which industries are being funded. Yeah, I love that. And I'm just sickened by fintech being equal with healthcare. come on well you know i can't come on don't shoot the messenger all right and then original breakdown not you know it's it's a big shocker we're talking about AI dollars excluding anthropic so we're
1:27:44 looking at two point three billion dollars united states got seventy three percent europe eleven in china got nine So interesting. So let's see who are the top five that got the money, money, money. Anthropic, we know Anthropic. This is one of the largest and second largest private round in history. Man, even with sixty five billion, if you want to talk about it, there still did not. They did not hit the biggest winner so
1:28:14 far of this, which is OpenAI. One hundred and twenty two billion back in March. The money buys them more computing power and safety research. Okay. The second company is Cognition. Got one billion dollars. It's an AI that writes and fixes software on its own. Okay. The companies pay for code that ships without human typing it. AI software engineering. Sounds safe. I'm going to make a bold prediction.
1:28:46 Tell me more. Cognition got a billion dollars Series D. Yeah. Within five years, it will be clear that PageMotor is actually the winner in the software rights its own. Let's see it. I love it. I love that. PageMotor will beat Cognition even though they got a zillion dollars in funding. Yep. All right. Company number three, Moto. We are writing it down. Don't you worry. Moto Labs got three hundred and fifty five million rents out the heavy computing power AI needs on demand.
1:29:17 So companies don't build their own data centers. AWS for AI workloads. It's picks and shovels, right? That's kind of where the money is going. They're the ones who are printing it. Maffei AI Series C designs specialized chips that make AI run faster and cheaper. We've seen this consistently now for like two months. What? That type of investment. AI chips? Specialized chips or software to run the chips better. Yeah. open router got a hundred and thirteen million a single doorway that lets any app
1:29:47 plug into any AI model claw gpt gemini without locking to one sells the freedom to switch This is interesting about UI, the user experience. This is actually interesting to figure that out because we are now figuring out how important UI is with AI, not like UI of your app. It's like how do you bring all of those things to your AI without worrying about OF and this and MC blah, blah, blah, all these little technical things that
1:30:18 Chris is dealing with right now on bringing this on. What do you think about this? I think sort of the most interesting question to me that's emerged this week is that, okay, a lot of things are available to the AI to do. How does it know what it should do? And then how does it know how to, if it's a complex multi-step process, how does it know exactly what to do to like know that it's doing it correctly? Okay. Like there's an infinity of things it could do. How does it know exactly what it needs to do to solve the question that you
1:30:48 brought to it? That is sort of the, you know, and I'm talking about like in terms of using software to create some outcome. How does AI know what to do? That is sort of the key question for everything now. Yeah. How does it know what to do? And when it figures out, yes, this is what I want to do, what are the steps it follows to do it to make sure it does it right every time? Well, I mean, like what skills you bringing to what, right? Like Claude has different things. GPT has different things. Gemini has different things. Like that's,
1:31:19 I think the cost of switching the model will be interesting because like My files are all stored on my Dropbox, right, in my Dropbox folder that's backed up. And I can bring any CLI or terminal and work on that. But my skills, I don't know how they will translate from Cloud to other things, but that's something to think about. That's a really important question. There's been some traction in this space because right now they're different, slightly different syntaxes. One is not immediately convertible to another.
1:31:50 But there's a catch. MCP we've talked a lot about it quad or anthropic put out the MCP protocol i think the most recent one is is basically a year old now the spec since then other companies that come on board saying yeah okay we think MCP is the deal and we're going to honor the spec as well what that does though skills can be defined through an MCP Oh. And then any model that's plugged in can use them, and they don't really need to speak the
1:32:22 Claude language or the Codex language or whatever. So we're going to see that landscape opening up to be useful to anything, which actually doesn't favor the models. They'd rather create the moat walled garden. Yeah. But, you know, it's like on one hand, it's like a double whammy for Anthropic. Yeah, we came up with the MCP spec. If everyone has to use that, then everyone has to use us. But if it becomes an open standard, well then, yeah, we got credit for launching this thing, but now they don't need us. I mean, you know, like... It's interesting the way open stuff works.
1:32:53 Claude got a bunch of stuff from OpenClaw, so it's kind of a nice little economy of sharing. And so bringing your own keys is becoming a big theme and big topic because of that model not being with one model for everything. So I think that's going to be an interesting trend to watch. So that's it for the funding. Let's wrap this up. What's your big takeaway from today? Well, I think the Facebook forum launch is
1:33:24 easily the biggest takeaway because of the implication. The implication is it's not this huge open internet that's plumbed by Google and then you get answers back. It's now a siloed internet where you're either relevant in a particular space with your information or just not. You know, we've seen search break down over the last ten years especially. Like, it's become more of a, you know, pay to play, less organic. Like, the old ways are breaking down. It doesn't feel as accessible to people.
1:33:55 But this community approach is an antidote to that. It's just that what we give up is this idea that anybody can play and you're just going to win if you're just there and you're playing along. It's like now you have to play in a specific realm and then you're relevant. At least that's sort of clarifying, you know, like there will be pay to play within these new spaces as well. But at least, you know, I have a Facebook group. Okay, I'm in the game and I know that my audience is here. Just even knowing that you have a defined audience that you can speak to is priceless now. And so it's solving a lot of the
1:34:25 problems that have sort of been things that have been starting to break on the internet, in my view. And I think that's something you have to pay attention to. No matter how you feel about it, about Facebook or whatever, or about like Reddit style activities, you got to go where the audience is. Well, and as somebody who has a sales recruiting company that helps companies get salespeople, I am seeing a big demand for go-to-market people who can actually sell all the stuff that companies build with AI. And that is going to be an interesting challenge to solve because like what Chris
1:34:57 said, yeah, you can have the best product, you can have the most amazing product, but you still need to figure out where to sell it and how to sell it. and where the humans actually are that are buying it, especially if it's a bigger deal size, right? But my other takeaway is this. I think that me playing with PageMotor's API on my site this week continues to blow me away with this delight of how my brain has to unthink
1:35:28 something to create something new. So Mark Zuckerberg clearly had to unthink about groups to create some kind of new product. Now, Chris had to unthink a lot of stuff about websites to create the website where things are done for you, not you actually doing anything. And I have to do the same thing. Like, how do I think as much as possible to actually prepare for this new era of creativity? Because here's the deal. Humans are still the leaders of creativity, at least for now.
1:35:59 And that's what makes AI so delightful is kind of creativity. you know, get this idea and go see how it happens. And so that's my challenge for everybody here today. Go and think something, go ask a question. How would it be different? And what would happen if, you know, I talked to my site, what would I want it to be? What would I want it to do? And I'll have fun with that. I think you nailed it. So, so in a world where things are not that developed, being creative and being able to like manufacture new things is the best skill in a world where things are already
1:36:30 established and you're having like an, a change in the era. Things are shifting over to a new paradigm. Unthinking is actually the best skill you could possibly have. Yeah. Be like a kid. I'm feeling so giddy. I'm feeling like a little girl playing again. And it's a lot of fun. So anyway, on this note, we will see you guys next week with episode forty four. Thanks for being here. Thanks for your comments and go play out there and think something. Talk to you. We'll see you next week. Have a great weekend. We'll see you in June. Yeah.
1:37:00 Bye.