Publishers Block Just 22% of AI Bots. Here’s What They’re Missing

0:01In terms of AI bot traffic, have you met someone more knowledgeable about it than you? I'm sure they're out there, but I've got to say I'm probably in the top five. I stare at this every day. My background is in network performance at big tech companies. I spend more than a normal workday every day on it. So this is kind of what I live and breathe right now.

0:39Welcome to Media and the Machine. I'm Rob Kelly. This show is about AI and how to profit from it. I believe this is the biggest technology shift of our lifetime. I'm a three-time founder and CEO of content-driven businesses, and each week I talk with the people on the inside of AI and content, The ones building this new world and learning what works firsthand. I'm an optimist about AI, but I'm also a realist. So every show ends with me asking our guests what AI means for our jobs, our families, and our future. My guest is Gavin King, founder and CEO of Known Agents. He was a software engineer at Meta for seven years before that.

1:17And as you just heard, he lives and breathes AI bot traffic. Gavin and I dig into a remarkable shift happening on the web. One third of website traffic is now coming from bots. We break down what those bots are actually doing, and bots often get a bad rap. But according to Gavin's data, 96 % of the bots actually follow the rules. Meanwhile, websites block only 22 % of bots. We go over live data about what percentage of bots are blocked by such companies as Gannett, CNN, and Adweek. We also get into the bad news. Just 0.1 % of human traffic is currently coming from AI chat referrals. and what publishers can do about this, from paywalls and subscriptions to MCPs and building businesses that AI agents can transact with directly.

2:00After all, as Gavin says, an AI agent's money is just as green as a human's, right? And at the end, Gavin shares his advice for young people entering a job market being reshaped by AI and whether his own future kids will even go to college. A special thanks to Tom Januszewski for connecting Gavin and me. Now, please enjoy my conversation with Gavin King.

2:47Every AI Google would make, in effect, a photocopy of the web. The bots would go out, grab the content that it wanted to help serve to people doing Google searches. What's changed now with AI? Yeah, so instead of just kind of one category of bots, these search engine crawlers, now there's dozens, probably like about 15 to 20, at least the way that we categorize, different categories of bots. Okay, so we're looking at your knownagents.com forward slash insights page. One headline here is 33 % of traffic now to websites is coming from bots. These are good bots, average, medium bots, bad bots, all the bots, about 33%.

3:30And I double-checked because I have three sites up now that Node Agents is running on, and I range from 32 % to 38%. So I'm sort of in the normal kind of range there. But there's kind of a surprising stat here on the robots.txt effectiveness. Can you first off just define what that means and what's kind of the surprising part of the data here? Yeah, totally. So what a robots.txt file is, is it's a file that you put on your website. Before a bot starts looking at any of your pages, it should read this file, see if it's allowed to visit particular pages. And if not, then it should back off. It is sort of an honor system.

4:11And yeah, there's sort of a current narrative right now that many bots, if not all bots, just ignore this completely. And that's just not true, right? Like this data set, this is thousands of websites that we're looking at the traffic of. Many of the big bots, especially of the bigger companies, they actually do follow the rules. So if you put something in for OpenAI or Anthropic, they actually will back off and not visit your pages. So yeah, I think a lot of people, like I said, immediately jump to firewalls or bot blocking enterprise software when they can get 96 % of the problem solved just by having a full coverage robots.txt.

4:46So let's talk about the four types of AI activity. And we're still on this knownagents.com forward slash insights. So you got AI training first. These are the major bots like ClaudeBot and GPTBot and so forth, right? Yeah, Anthropic or OpenAI or Google or Meta. They have their own bots. It's collecting website data. And then that data is basically used to train AI models, whether that's LLMs or something else, like the first-party scrapers. How can you tell? How can you tell that someone's training on someone else's data? Yeah, so a lot of the times, which is good, they'll identify themselves.

5:23So they'll say, I'm GPT bot, for example. Actually, there's literally sometimes a URL in the request itself that says, hey, if you want to learn more, go to openai.com slash whatever bot. And then you can visit that website, and then it will describe what this bot's for and it will say, hey, this is used for training. You can block it by adding it to your robots.txt. Here's maybe some IP addresses that it should be coming from. So in this case, this is saying roughly 5 % of traffic of the... How many sites do you have in the Insights page? Yeah, this page, the Agentec Web Index here, is built by about 5 ,000 websites.

5:57Okay, 5 ,000 websites. Around 5 % of those websites' traffic is coming from what we can just call AI training related bots. Yep. You got Anthropic number one, Meta number two, Amazon number three, OpenAI number four, ByteDance, owner of TikTok, ByteDance is? Correct. Yep. As number five. Can you describe why the numbers are like they are in that order? Meta and Amazon kind of surprising, given that they're not as big market share as Anthropic and OpenAI do? Yeah, but they're trying to be, right? So that could explain some of it. Okay. So they might be playing catch up just to - Exactly. Okay. Right.

6:34That could be indicative of maybe they're more kind of in the training phase of their new models. Like, you know, Meta's coming out with a bunch of new models. Amazon's starting to get in the mix with their own models. So yeah, maybe they're just doing more training these days or they have less already crawled because they're, you know, kind of falling behind a little bit in terms of timeline of developing these big LLMs. Yeah. So if you see OpenAI at number four, it's possible by far there's 7 % versus Anthropic up top of 26. It might just be OpenAI had a bigger head start on training and now they need to train less.

7:05Their bots need to hit less. Yeah, maybe there was a big surge months ago or years ago, and now they're kind of just incrementally crawling web pages, whereas that might not be the case for these other websites who are kind of crawling things for the first time. And what's going on sort of the first unknown name or sort of non-big company is Parallel. What's going on with Parallel at number six? Yeah, yeah. So this has been sort of a recent surge in the past couple of months, there's been kind of like this proliferation of third-party crawlers. But now you have a bunch of SaaS businesses spinning up who are kind of like doing that on their behalf.

7:39So you might get a crawl from Parallel or you.com or like Exa. So they're kind of like these multi-purpose crawlers. And there's this whole other side thing that we can go off on about Cloudflare kind of like getting them to split these up into use case. But this is like another sort of cohort of bots that are visiting websites. These, you know, one size fits all bot. So Cloudflare is asking them to do more than one at a time or saying they'll be... Yeah. So they're basically saying, break up your use cases or, you know, we'll start blocking you. Taking kind of the common denominator of what this bot is used for, they're basically going to assume that it's for the worst unless you start breaking it up into multiple bots.

8:17So they're saying, we'll block you unless you start splitting up your use cases. You've also got the top visited website categories for these AI training bots. And it's surprising. So the top two are law and government and online communities. Can you just give me your take on why those are the top two categories? I think online communities is a good example. Like you want them to talk like a human talks. And the best way to do that is to crawl user-generated content, right? So if you think about, you know - So that's why Reddit is the number one source So training data for AI models, it's because they have this nicely formatted question and answer that real humans, last I checked, are providing there.

9:01And that's very easy and effective and valuable for an LLM to train on. Is it as simple as that? Yeah. Like if you want your LLM to talk like a human, you want to train on humans talking to each other. Like you don't want to, I mean, you can, but you can train on like an article, but that's not going to tell you how humans interact. It will just tell you like, you know, what concepts go together. So yeah, online communities, it's top here for a reason. Wow. So it's not just on Reddit, it's not just the actual content and getting the answers. It's also how humans are speaking to each other when they're getting to that question and answer.

9:32Fascinating. Exactly. Yep. Yep. Yeah. Because it wants to mimic human interactions. And that really is the beauty and the big breakthrough of LLMs, right? Is it's looking at language and training for now. It's the main trainer of AI, right? Of AGI. Getting to AGI. obviously physical world models and stuff cue the dog i love the dog barks keep them coming so then um that's number two online let me just quiet her real quick before we yeah what's her name joe zoe she's like this little yorkie all right zoe the yorkie barking if anyone cares okay so that online communities was number two trained on type of category number one is law government.

10:13Why is that number one? If you think about common use cases of why people use LLMs and especially like why enterprises would pay for LLMs, you know, you want them to be good at things like legal or making decisions. The best way to do that is to actually just read legal documents or read government websites. This has more to do with what people use LLMs for, especially like enterprise customers. And it's authoritative, kind of the primary source in case of government. Okay. So So that's the AI training tab there. Should we click on what next? AI search indexing? Kind of going in order of traffic roughly?

10:49Let's do fetching first. Fetching next. Okay. What are the insights to read into here? This category kind of encompasses two types of bots. By far, the most important one for most websites is AI assistants. So this category is when there's literally like a human in the loop on the other end. They're working with an AI. They're asking questions. They're doing some research. And the AI is going off and pinging websites. based on what the user is asking it. RAG is an acronym that people are probably familiar with, but that's exactly what this is. So yeah, for example, if I'm looking up sports scores or I'm writing a research paper, let's say I'm typing something, I say, what's the current score of the game?

11:27It might go off and ping ESPN and say, hey, what is the actual score right now? I'm going to respond with that to the user. Maybe I'll include a citation as well that the user can explore more and click into the page. But yeah, that's basically what this is, is that kind of human in the loop going off and pinging websites. if you're ChatGPT. And by far, the number one operator is OpenAI with 88.4 % of the fetching activity. And then Anthropic is next at 4.7%. Is that just really market share of people using chatbots? Yeah, so it is mostly just market share. Like, you know, more people use across the world OpenAI products and Anthropic products.

12:05Gotcha. What should we do next? Search indexing? Yeah. So in this case, give an example. but what's an AI search indexing type of tool? This would be the Apple bot perplexities in here too, or? Yeah, so AI search indexing, these are how the chatbots know what's even available, right? Like visiting like a science website, it has to know that that science website exists in the first place. So it might have already kind of like done a rough crawl of that website says, okay, it probably has answers for these things. It's kind of saving what could be a good reference. That's what this bot does. It builds up that index, which, you know, Later, an AI assistant might ping that website, the same website, to get fresh information.

12:43But this is just kind of similar to Googlebot. It's how it knows the lay of the land, what's out there, what's available. All right. So in this case, some of the top agents are Perplexity, and then the search part of Claude, and the search part of ChatGPT. But one surprise here is there's no Google. And you explained it to me in a past conversation. The reason there's no Google is because? I mean, Google, because they already have such a big index. They might be using some crawlers for multiple use cases, which is something that people are pushing back against. So that's why it's not here, is because they already have a big index, which they are allegedly pulling from, instead of crawling this as a special case.

13:22Okay, so next is AI browsing. Yeah, so this is really interesting. These are the actual AI agents. People use that word a lot. They throw that around for bots in general. You, as a human, are still on the other side, but instead of it doing research, and it's kind of like pinging random things, you actually give it a task. So if you say, I want to book this dinner reservation or I want to book this flight or eventually I want to buy this thing, it's the thing that actually uses a browser. It literally has a window of a browser. It has a cursor, it's going off and clicking on things. So it's doing multiple steps on a website with some goal in mind.

13:58So if I went to sleep and said, I want to build a new business, go ahead and build it for me by the time I wake up at 5.30 a.m. all of that activity would be the type of activity you're describing here in AI browsing. Yeah, it could be. So if it's building a website, some of that might be done by those AI coding agents, but some of it might be just testing what it built. So it could be spitting up a browser, it's clicking through the website, making sure it coded the thing correctly. So yeah, that would be an AI agent. Okay, taking over the browsing of pages and the web. Exactly, yep, yep, exactly.

14:33And this is interesting in this case. I mean, I've looked like you had a lot of Google Analytics in my day from different blogs and websites created. So it spends on average one minute, but looking at 13 and a half pages. So that would be, you know, roughly more than four seconds per page. So that would be way more productive than if you or I were going to look at pages and actually use them ourselves as a human, right? Yeah. Yes and no. I actually think generally the models are pretty slow at this still. They literally, most of the time, take a screenshot of the page and then they look at the image of the screenshot and then take an X and Y coordinate from that of where you want to move the cursor.

15:14So it's a very slow process. If you look at the actual session replays of what they're doing, a lot of the time it's just randomly reloading the page or clicking on something twice or going back because it messed something up. So there's a lot of stumbling along the way. But yeah, I expect over time as the models get better especially, and there's actually been a lot of progress on this in the past month or so, they'll get more efficient at completing their tasks faster with maybe fewer page views. But yeah, I think this is very up and coming. Sam Altman tweeted about this the other day, or I guess retweeted something from ChatGPT.

15:43They're doing this more and more at SpaceX AI. They also have GrokBot, which was just released. That heavily relies on browser use. They're partnering with Stripe and Link to do shopping. So there's a lot going on here, and I do expect this is going to probably ramp up pretty quickly over time. And these products are going to be doing more and more interesting things. Okay, so now we're looking at robots.txt and sort of the big headline here is 95.7 % of bots are following the rules, basically. If you put in rules in your robots.txt, they are following them. And what about this list of top blocked agents here?

16:20Can you read through who each one is and what that means? Just maybe the top five. Yeah, so as part of this data set, we look at the top thousand websites, according to SimilarWeb, and we look at their robots of TXDs every day. We take a snapshot of all of them, and then we derive these insights from them. So yeah, the top block bots by far are AI data scrapers. So these are the first party scrapers that are collecting data. CCBot is also sort of a multi-purpose crawler. It is used to train many of these AI models. Does CCBot stand for common crawl? Exactly. Yep. Yeah. So that's the top one. ByteSpider, that one is just kind of particularly aggressive.

16:55It might be hitting many pages in a very short amount of time. So that kind of makes the people running the websites angry. So they might block that one over the other ones. So that one's kind of at the top. That's ByteDance's spider. Yep. Exactly. Yeah. Then you have like GPTBot, CloudBot, Metaspot. So kind of like the big AI companies. What's a worst case scenario you've heard about an AI bot hitting a site? Yeah, I mean, you will 100 % see some of these bots. A lot of times it's like the scraping ones kind of hit your website hundreds of thousands of times, depending on the size of your website, but within a couple of minutes.

17:32That happens a lot. That's actually a pretty common pattern. You'll see nothing for a week and then it will just hit everything at once. A lot of CDNs are pretty good at handling that kind of traffic, but some aren't, one thing. And then two, that's just costing you money, especially these bot visits that aren't giving anything back. You pay per request a lot of the time because these are usage-based billing systems with the CDNs. So yeah, it could just cost you a lot of money and potentially take your server down. Although I will say that's not super common. Yeah, I remember when I first started using your known agent's tool, it was amazing, like 67 milliseconds, 60 milliseconds.

18:07I'm looking right now in real time, 482 milliseconds. Bots just hitting like crazy. Right, yeah, it's like this fire hose of visits. All right, so now we're on AI chat referrals on your insights page. First off, just define this. What does it mean? It looks like it's around 0.1%. So what does that mean? Yeah, so this is the fraction of your human traffic that comes from people clicking in from AI chat responses. Okay, so this is the big bad news, only 0.1%. Obviously tiny compared to the days of Google. That's the bad news. What can publishers do about this? One, I think this is just bad. You can spin it, but I do think this is just bad sometimes, especially if your business relies on people coming to your page specifically and seeing ads.

18:52That's going to be a bad time. Yeah, it's just bad. Right, right. We've got to say that. I see a lot of people converting, for example, more of their content to paid or behind a paywall or they're offering subscriptions. So these are things that an agent can't get past a paywall. So they can say that, hey, if you want to sign up for this thing, you have to click through, you have to subscribe to this publication. and that might be a better model in the AI world if you want that premium curated content. But something needs to change. There's not a real future where we're going to get as much traffic from AI chat referrals as we got from Google.

19:25That's just not going to happen. So something needs to change, and for better or for worse. So there is sort of a silver lining, not enough to capture the whole thing, but I've heard anecdotally that these humans are more valuable. They're more likely to subscribe or to buy something or to sign up or to convert in some way. But even if they were 10 times more valuable, that would still be 1%, right? Which is nowhere near the value you're getting from Google. Okay, so now we're looking at this publisherrobots.txt leaderboard on your site and back to the AI bots are kind of doing their jobs very well.

19:5995.7 % of them are honoring robots.txt rules, but publishers are only blocking 22%. And I just remember that most people are just going to be listening. So I'm just going to read off some of these so people can understand. So the ones with 100 % coverage are Adweek, Arkansas Online, Golf.com, San Diego, UnionTribune.com, Slate, The Star. And then it starts falling off to way less than 100%. And the list goes down to some folks who have tiny coverage. I don't want to pick on anyone here. But if you look at Fox5Atlanta.com, you know, 2 % coverage or fool.com. Is that Motley's Fool? I think. Yeah.

20:42And Country Living, only 2%. A lot of big companies. American Banker, only 2%. 404 Media, they love to report about things like bad bots. They have 2 % coverage. TVGuide.com, only 2 % coverage. Meaning they're only blocking 2 % of the AI bots. So a huge, huge delta here and bad coverage. But as you were saying earlier. So basically across these 1 ,000 top publishers, yeah, only 22 % of the bots are actually blocked. So if you were to click into the robots.txts of some of the websites that you just mentioned, maybe they block two or three of them. But there's over 100 AI bots that we know about that are actively scraping things to train LLMs.

21:26So yeah, the coverage is surprisingly low here for how big of a topic it is. And for, I guess, this new insight of how well bots actually do follow it, there's a huge discrepancy here that I think is pretty surprising to people. So to summarize, bots follow the rules like 96 % of the time. So the bigger issue or another issue is that coverage is just really low. So I'm kind of on this crusade right now to get all the publishers to actually make sure that they have a full coverage robots.txt. because that's the first line of defense and that will solve 97 % of the problem of AI scraping at least.

22:01But say an ad week or golf.com, one of the few in Slate who are at the 100 % blocking, is that a good strategy to block 100 % AI bots? So these AI bots in this list are exclusively the ones that do training without giving anything back. So this isn't including like those AI search crawlers or those AI fetters. Okay. These are only the AI training bots. Exactly. Yep. Yep. And also, these are like running up your traffic bills, right? If like 30 % of your traffic is this type of bot, then your CDN bill is 30 % more expensive. So even if you don't believe that against the data, if you don't believe that many of these bots are following the rules, the ones that do will still save you money on your server bill.

22:43And why do you think people are missing the 100 % coverage? You know, I'm looking here at some of these others. CNN, 39%. It's like, they got a smart team over there. Why are they missing this? Yeah, I think it's just hard. It's a hard problem. There's just so many bots. There's new AI companies popping up all the time. Within the companies, they're popping out new bots all the time. So it's just hard to keep track of. So I'm looking here at all your individual web properties, and I see a bunch of Gannett ones. And I know Gannett came up recently. This is Herald Tribune, IndyStar, Jacksonville.com, a bunch of them.

23:13They're huge. Got a lot of local sites in addition to USA Today. But anyway, in your numbers here, it looks like Gannett properties very consistently block around 65 % of that AI bots. I know I heard recently they said 100%. What's the difference? Yeah, so it could be a few different things. I mean, first of all, 65 % is far above the average, which is that 20%. Yeah, they're way high up. They're way, way above average, probably in the top five publishers or something doing it. That's why there's like a leaderboard. Yeah. But I was just curious, what's missing? Are they missing some bots? Is that simple as that?

23:47It could be a few things. One, it's like, I mean, they're like licensing deals play into this. Like sometimes you do want to allow certain bots. If they do have some deal, like they might not block certain bots to let them in. They could mean that they're blocking them in some other way. So maybe they block them with their firewall, which is like hard blocking instead of kind of like this, like soft blocking, like a RoboSat TXT. Or yeah, they might just not know about all of them. This is our full-time job. We have our own AI that scans the traffic of the thousands of websites that are connected to us.

24:14It finds new bots literally every single day. So we also probably just know about more bots. So there's this new New York law, it's called the Stealth Crawler Prohibition Act. It looks like there's fines of$15 ,000 per day if there are stealth crawlers that fail to comply with disclosure requirements. Is that going to hit up these bad bots out there? Yeah. So I mean, what these laws are is basically just making it so they can't pretend to be human. There's a few versions of this. There's the New York one. I think there's an analog in Europe. Basically, what that will do is make bots have to identify who they are and also, in some cases, comply with robots.txt.

24:53If you already have a full coverage robots.txt setup, it will only become more and more effective as more bots start following the rules. Just for some positive news on AI bots in general, Like I saw one the other day, which was Yelp did a deal with OpenAI, where OpenAI will now with ChatGPT use Yelp's localized info. And of course, you know, they're probably getting paid some amount of money for that licensing deal Yelp is. But also Yelp got, as part of it, they said, and soon there'll be a link when folks are on ChatGPT and looking up Yelp stuff. There'll be links for requesting a quote from a plumber.

25:29You know, as part of Yelp, you know, they provide local contractors and plumbers and other stuff. True commerce coming from it. I was like, what a great use of AI for a content-driven company with Yelp. Can you just give me a few other use cases like that that have got you excited to counterweight some of the bad bot news that comes out constantly? No, I mean, I think that's my point. Like, I think some of these businesses will have to adjust their model for this new reality that's definitely not going away. So yeah, whether that's selling to AI agents rather than selling to humans, making that easy, like working with some of these new paradigms to be able to offer these things through AI platforms.

26:07That's a really good example. Yeah, a lot of these marketplaces or services, you really don't care if a human books your thing or an agent books your thing. So if you're booking a flight, you're making a dinner reservation, you want to make sure that your website is easily accessible and optimal for this new type of user, which is like bots, because their money's just as good. Their money's just as green as a human's, right? They can do the same thing, and you really shouldn't care. So I think it's important that you make sure you're not blocking them or your website's set up correctly. But yeah, these are these new kind of business use cases that especially for when your business isn't page views, it's some kind of conversion, whether it's buying something or booking something, AI agents will actually probably be sort of a tailwind for you.

26:49Great. Zoe. Okay, there's Zoe, good. If you were building a new site today and wanted a lot of AI traffic, what would you do? I mean, luckily you don't have to do much. For example, if you want to attract, this whole GEO thing. 95 % of that is the same as SEO. If you have good SEO, that's 95 % of the problem if you're trying to attract them. For the next cohort of AI agents, the ones who are clicking through your website, that's where some specific tactics come into play. For example, you want to make sure you have a good accessibility tree because that's something that Google has said that their agents value and help them navigate through your website more quickly.

27:30If you can, many people have heard about MCP, but it's basically an alternate way for agents to access your data or access your service. This is something that's basically through an API. It's not through a browser. And it's a lot more token efficient. The agents usually can get something done much faster through this. So if you were to set up an MCP in parallel to building your website, that would a lot of the times be more optimal for AI agents to access the same data. So last time we chatted, you talked about some good things to put behind an MCP. for a data owner or a content owner. You talked about surveys, polls, research.

28:06Yeah, yeah, for sure. So I think there's opportunity for publishers, like content creators, to pay well more stuff on their website. Or another way to do it is if you want this accessible for AI agents specifically, you could put that behind an MCP, charge some kind of access for AI agents to be able to retrieve that data. So yeah, great stuff. A lot of these publications do a ton of good research. They do polling. they have these proprietary data sets that would be valuable for AIs or the businesses that use AIs. That could be another revenue stream. It's like putting that behind an MCP, locking it off, and then paywalling that, and then getting revenue from unique data.

28:43What's something that leading publishers are doing in this space that has your respect? I think a lot of adjustments to the business model to depend less on inbound traffic are always good. You could have your free articles, but you might want to decide to put more behind a paywall. Or maybe before you didn't have a subscription offering, but now you implement a subscription offering. But also there's more creative things, like offering things like games or doing live events, hosting conferences. These things can also generate revenue without being disruptible by AI. Going offline. Man. Yeah, I was talking to Jimmy Hutchison, the CEO of Spin, who's got Spin Magazine, but they also do have events and other things.

29:24And basically, AI is really not hurting his business much at all, because he's got so much offline business and just a strong brand. Totally, yeah. And I think the AI companies see this too. You see these mega salaries for event producers within Dropik and OpenAI and things like this. So I think just live events in general are going to become more and more valuable. What's your take on, this just happened recently, Reddit, they went from receiving 4 % of all citations on AI searches to less than 1%. What's going on there? Any idea? Yeah, that's what it was like black box decisions. I don't know if there's some behind the door deal going on or yeah, I don't know.

30:02I haven't actually kept up with that. It can be a function simply of an open AI or Google just changing their algorithm in terms of what their bots are hitting. So it doesn't surprise you that that could happen to a publisher. You can just deny list someone from being surfaced as an AI assistant, or you can change the algorithm so it subtly edges someone out. There's many ways that you could start blocking someone. You came from Meta. What's your take on their AI strategy overall, just how they're doing AI-wise compared to the open AIs and the anthropics of the world? I mean, obviously I can't disclose anything confidential from when I was there, but But I will say in general, I think that these incumbent tech companies that have kind of a portfolio of other products that they can kind of just plug AI into to make them that much better are in a better position than I think people think.

30:54Kind of the front runners right now are the OpenAIs, the Anthropics. They have the best models right now, but they have to invent ways to use them from scratch. But these are use cases that they're having to develop zero to one, whereas a meta or a Google, it's really easy to plug in AI into Gmail and make it super, super good. Or to use AI to do creation tools on meta or Instagram that make people use the product that much more. So I don't think they fully realize that future state yet, but I do think they're in a much better place to capitalize on AI than people give them credit for. So cool.

31:30Now we're just on to a couple of humanitarian questions and we wrap up. What are you telling the younger kids in your life? I know you don't have kids yet, but just the younger kids around you, what are you telling them about how AI is going to impact their life or prepping them for this new world of AI? I think the biggest change, you know, not super young kids, but one thing I see is, you know, my old university, they are seeing a lot of people like entry-level engineers unable able to get jobs or there being like fewer jobs for computer scientists, we're kind of probably going to get into a trap where, you know, we're not up leveling these people who are just kind of coming into the market.

32:07And then therefore, as the market matures, there's not going to be people to like develop into those more senior roles, which is bad because I think you are always going to need a human controlling a lot of these AI, especially with coding and with engineering. You need someone who's orchestrating it all. So, I mean, the best one is just to like leverage AI as much as possible, especially if like you're in that situation to kind of hate on AI and say that it's ruining everything. But at the end of the day, this is the reality and you need to learn how to leverage these tools because the genie's out of the bottle and you need to learn how to use the genie to do what you want.

32:38If you have kids, do you think they'll even go to college? I don't know. They'll probably have brain implants that just puts college into their neurons directly. So who knows? Well, this is Media and the Machine. A few things about you and me. If you want to hear about the next new episode, make sure you hit follow on the show on your podcast app. And if you like something about this episode, the guests, the topic, even one part of the conversation, leave me a rating or review and tell me. That helps me know what you want more of so I can make the show better for you. And if you want to go a little deeper, head to mediaandthemachine.com and subscribe to my newsletter.

33:13You'll get my favorite highlights from podcast episodes as well as essays I write about AI. You can also email me directly from there. Recommend a guest. If you make it on the show, I'll give you a shout out. If this show made you think of someone, please pass it along. That's how people find us. And I'm genuinely grateful. Thanks again. I'll see you next time.