Claude's new CoWork Feature has a SERIOUS Problem..

Google's on a roll this week. There's lots of new things that's going to shape the AI market for years to come. We've got personal intelligence, a billion dollar deal with Apple, and much more. But that's not even the scariest news. This week we have a major agent horror story. Claude's new cowork feature, which is quite interesting, just permanently deleted 11 gigabytes of an early user's personal files. So today we'll tell you exactly what not to do with these new agents. And for ecom businesses out there, Google just quietly launched the Universal Commerce Protocol. This might be the beginning of the end for the checkout page and the birth of agentic commerce. So if you sell anything online, then you need to hear this. Plus motion control, deepfakes, and a new regulatory crisis over at X, we have a packed episode today, so let's dive in. First, if you're new here, my name is Mark Webster and I'm joined as always by my co host and co founder of Authority Hacker Ken, Gail Brighton. How's it going, Gail?

0:00

I hate receiving this question. Now, for those who don't know, I asked Mark this question for like literally five years because I was the one doing the intro for the podcast. Then we realized the native speaker doing the intro was just a better idea because.

0:58

Six years to figure that out.

1:11

We're a bit slow. We're a bit slow, but we figured it out eventually. And so now I'm receiving this question and I never have anything to say other than I am almost certain already at this point, like we're mid January right now, that this year is going to bring more changes in the world of AI for business than last year has. We might talk about that. That's kind of a hot take. But like, I'm seeing things that I think most people don't see and I am like, holy shit, this is good. So yeah, we might talk about that, but let's jump into this.

1:12

Make sure you stay to the end of this episode to find out if this is going to be such a game changer of a year and how. But first I want to focus on Gemini introducing personal intelligence. Can you tell us what this is and why are we featuring this story first?

1:40

Yeah, it's so funny because you came for the podcast and you're like, I never heard of this. What the fuck is this? And it's like, I'm like, so probably the biggest story of the year, of the week, both for business individuals, maybe not the year. Let's not push It. But for this week, definitely. And that's essentially like, you know, all chatbots have like kind of a memory feature, right? That's why a lot of people are sticky to ChatGPT, for example. They're like, oh my God, it knows all my life, blah, blah, blah, et cetera. Well, as you may know, Google has a lot of data about it that it wasn't really leveraging before in the Gemini app, like there was a memory feature, but it was kind of very basic. Like you'd need to tell it to remember something and it would just be loaded in a context and not very good, to be honest. It would kind of like over index it and over reference it on responses that would not be related to this. But now they've just rolled out a feature that lets you opt in the data that they have about you from other apps. So you can opt in your Google Photos data, you can opt in your Gmail data, you can opt in your YouTube watch history. I don't think they yet put your Google history.

1:55

You mean your browser history from Chrome?

2:58

Yeah, yeah. So no, they don't have that yet. They don't use that yet.

3:00

What could possibly go wrong?

3:03

Yeah, exactly. I mean, people might learn how to use the incognito window. But anyway, the point is that Gemini is now the easiest chatbot to personalize if you're just watching YouTube videos and using Gmail, et cetera. Now there had connections to all these tools before, but it was kind of like a proactive thing. Like you would tag the tool, you'd be like, oh, check my Gmail and check the email. It would be able to do that. But now it's going to consider all that context by default and it's going to go like, oh, by the way, you received this email, et cetera. And so there's actually like some tweets of people who experienced it. One thing I haven't said, by the way, it's only in the US and I think the chances of us seeing it in Europe anytime soon are abysmal, especially given the politics between the US and Europe right now.

3:05

Is this another one of those things, though, that anyone with a VPN setting it to the US can suddenly get access to? Like, I don't think they don't really try and block it.

3:49

No, it's not that. It's like account based, like if I think it's a bit harder, basically. So some like, you know, lab tools, et cetera, you can trick your way through, but this I could not trick my way through. Actually, but yeah, there's some people who have access in us and then they've reacted on Twitter or on X. Sorry. And she's basically like, I've been testing this. And here's two examples that stood out. Trip planning. Gemini planned my trip to Hawaii. It knew my flight times, walked around my existing reservations and even factor in commute times from where we were staying. The other history win, I was buying Turbo clothes for my dining table and sent Gemini the link. It found the dimensions of my table in my old Gmail order confirmation and told me which size to buy, for example. So that's really cool. It will be able to see which deliveries you have on the way, your preferences based on your photos. I think Josh Woodward, who is kind of like the head of Gemini as well, gave a story of it just asks it like, oh, how do I change my tires? It finds the photos of his car in Google Photos. It knows which car he has and it gives him the right answer without him having to give context on what car he's using, et cetera. So yeah, it's like in principle, it's pretty cool. Now my only problem, as I said, is that Gemini tended to massively over index on memories prior to this release. And I haven't been able to test it, so I can't tell you for sure, but there's a chance that it also does that. So the example I gave last week was like, you know, I had memories in Gemini that said like, oh, my wife is Polish, I live in Budapest, et cetera. And I would type like, I want to buy a new TV and be like, oh, since your wife is Polish, then probably this TV is better because it's like, no, like it doesn't matter.

3:58

That kind of feels like one of those things. When the technology is new, everyone points out the kind of super obvious flaws. Like, do you remember a couple years ago chatgpt said to book glue on pizza or whatever and yeah, yeah, just.

5:30

Doesn'T do that anymore. Yeah, reviews. AI reviews used to do that. But yeah, it will get better, no question. And the trove of data that Google has is a massive competitive advantage against OpenAI. Right? OpenAI has to kind of extract data from chats from you to know you. For Google, you just have to open your browser or use Google Photos, which a lot of people do, or watch YouTube and then they already know a ton about you.

5:42

And I presume like in a work situation it will have all your Google Docs and spreadsheets and really understand that.

6:04

We know how Google treats Google Workspace for their Google products, which means like half the features are missing half the time and never get released. So we'll see how that goes.

6:11

I don't know, they seem to sort of like ring fence it in a way. Like, you know, they're very keen to promote the fact if you're using Google Workspace, that they don't use your chats to train its knowledge.

6:21

Yeah. Which I think they might not use these actually. So it's like, I'm not, I guess.

6:31

They have, they have branded it personal intelligence and so, you know, they might have a different product, business intelligence or something.

6:36

But actually what they're positioning this against is not, is not even OpenAI. They're positioning this against Apple and we'll talk about the Apple deal later. But the point is that the new Siri, when Apple presented it two years ago and failed to release, essentially was this. It was like.

6:43

Oh, it was called like Apple Intelligence as well. Exactly.

6:57

And now it's first and then it's the same kind of gradient of colors that Siri has on the communication, et cetera. So I'm telling you this is a Siri counterplay now that Siri is going to be powered by Gemini, basically. That's the positioning. But I think this opens a very interesting debate for business owners because business owners are very excited about AIO right now, which means they're essentially ranking on AI, being referenced by AI. And so a lot of people and agencies, what they do is they track essentially your mentions inside LLM results. Right. The problem is that now the context that the LLM is going to use, through using your YouTube history and your Google Photos, et cetera, will steer the answers in vastly different directions depending on what Google knows about you. Essentially, they will probably maintain some kind of memory file. That's a summary that they give to the LLM based on what the summarize. I don't think they fetch all your photos every time you make a query. Right. But the point is that the idea of tracking LLM queries like you would track Google queries because essentially that's what's happening in the industry right now. Right. They basically took one to one what happened on Google. Let's give some keywords, let's give some queries and then it's just like, am I reference here or not? How often am I referenced? If I run the query 10 times. And that's pretty much kind of like how tracking works right now. Well, how are you going to be able to do that with this degree of personalization Now, I know people are saying, we'll probably answer to this by saying, oh, but Gemini is not that much traffic, ChatGPT is more important, blah, blah, blah, et cetera. Yeah, we talked about that. Gemini is eating into ChatGPT like crazy. For the past six months, basically, they've been growing a lot. These are the stats from similar web and they went from like I think 5% to.

7:01

Yeah, I think this was from last week's show. Wasn't it more the fact that they were both growing, but Gemini was just growing much faster.

8:37

They are, but in the category, Gemini is growing in market share. It's still a fact that people are doing that. And Google is very aggressive on acquiring users. And now with Flash rethinking, for example, is vastly superior for the free offering than ChatGPT, for example. It's much higher quality, the model is much better. And so I don't see these trends stopping overnight. So it wouldn't be surprising if by the end of this year it was close to a 50, 50 basically of market share. And so this plus personalization equals. It's going to be fucking hard to track how well your business is doing on LLMs, basically. Yeah.

8:42

And I think we've kind of got more to talk about with this when we come later in the show to talk about the UCP Universal Commerce Protocol, because that also flips the script on how online commerce and basically stuff gets bought online. But so this sounds very similar. A lot of these stories today actually overlap in some interesting ways.

9:17

Yeah, I mean. Yeah, but the point is, I think this idea of tracking queries the old way, et cetera, by the end of this year will not make sense anymore. It's going to be very hard for you to justify your AIO services by tracking that way. I'm not saying this is not going to exist and there will be no marketing on LLMs. I'm saying this concept, this whole idea of like this is how we measure performance is a bit.

9:36

Yeah. Just to be clear though, I think like what we're saying, like, you know, getting found and mentioned. Yeah, it's good AI. It's a growing industry and it will continue to grow and there's high demand. It's just specifically the tracking element you're talking about here is.

9:59

Yeah. Because people, the way these trackers work is they just call the RAW API. Right. It's like. And even the API behaves differently from the chat apps anyway.

10:13

Yeah.

10:21

And then on top of that, you had the whole day of personalization that everyone's Starting to get and same Sam Ottman was like, this year we're going to add a lot of personalization to ChatGPT. Same problem. It's going to be difficult from an API call, tell what people are actually seeing, basically. So, yeah, it's something to think about.

10:21

Let's move on then and talk about Claude Cowork, because this was an interesting new feature that came out this week. Apparently they kind of coded it in a couple weeks, like 15 days.

10:39

10 days. 10 days they did it, yeah.

10:50

But essentially what it is is it's Claude code, but a sort of slightly lighter version and within the Claude app with not quite so many features. Is that a sort of fair assessment?

10:52

Yeah, I think it's basically that it just allows you to interact with local files, but it can also do everything else Claude does. It can call it mcps, it can do all of that. And it's kind of like it behaves like cloud code, you know, it's like the chat app, it won't do a to do list, for example, whereas now it will do a to do list.

11:07

And if you're not familiar with Claude Cod, I think like a lot of people, a lot of non developers dismiss this because they think, oh, it's a coding tool, I'm not coder, I don't need to use this. But you actually use this basically all day for your work and you're not coding in it. So do you want to just explain quickly like what the use case is here?

11:21

Everything. It's like I literally, quite literally, I.

11:40

Mean, you said to me this week, like, we should do all our work in this tool for this year. It's that extreme.

11:44

Yes. I do all of my work on Klot code at this point and I think you see my usage limits. You can probably see that if you check this out, you'll see that I use it quite a lot. Yeah, it's like, I mean, for us, we're running like an AI community as well, et cetera. So it's like it's preparing lessons. I'm like, after I shoot the lessons, I give it the transcripts and it can kind of like review against the plan for the lesson. It's like, oh, you missed this, you should reshoot this. There's this missing, et cetera. It kind of coaches me through it. Then after that I trained it on making the lesson notes. I also trained it to connect, to circle. And then actually now it can upload our lessons for me. It prepares the thumbnails for the lessons. It's kind of like an assistant, you know, it does all the work I'm doing. That's like serious work almost.

11:51

And I think for anyone who's used to working in, you know, Stock, Cloud or ChatGPT, this is a different experience because you kind of like, give it an objective. It does and that it goes and it can spend, like, quite some time doing it. So to the fact that you can almost like go away, have a cup of coffee, or start another parallel process for a different task.

12:32

Yeah, but most importantly, it does the thing you need to do. Like a chatbot does not do anything. It gives you text back. You need to do something with that.

12:51

Text copy, paste it in and out.

12:58

It's like nothing gets done if you stay in ChatGPT, nothing gets done. Right. Whereas if you stay in cloth code, things get done. It's like, I answer support tickets through Cloth code. It helps me. I create content and publish it through cloud code without switching window. I do market research, calling multiple APIs, create images through cloud code, like marketing assets, meta ads, campaign, everything. I can connect to these platforms and do the thing, not just give me text.

12:59

Essentially, this is reducing the friction to use AI when you have a big task, think, oh, I've got to explain what I'm trying to do and all that. It knows all that stuff because it.

13:29

Doesn'T just reduce the friction, it does it. That's the thing. The stingy is like, through code skill as well. The skill thing is very important. It's like you can run a workflow, then be like, oh, save it as a skill, which is kind of like an SOP for AI. So next time we do it, it's like, oh, yeah, this is how you do this. Step one, step two, step three. And then let's say you run it the second time it didn't go perfectly. It's like 80%. You know how AI works, right? And you're like. And then you give it to you like, no, actually, you should do that. You should do that. You kind of run the chat and you're like. At the end of that, when you have a good result, you're like, oh, okay, you see all the issues we had here? Update your sop. And then next time you run the skill, it's like 90% there and then 95% there. And you get that kind of compounding effect that you get from training someone, basically. And you kind of figure out all the edge cases as you run the workflow together. And then eventually it just handles most work. And again, because it connects to platforms. It does it. It doesn't just give you text.

13:40

That's one of the big kind of annoyances of using even something like projects inside any of these.

14:32

It's not the same.

14:36

It's just like, I know, like updating the context files, especially when you've got like lots of information there, but you.

14:37

Don'T need to anymore. It updates itself too. You just have to.

14:42

I know, I know, that's what I'm saying. If you use this, it can change its own files, basically. But that actually caused a pretty significant problem this week for YouTuber James McCauley. Just to explain. So on Tuesday, Anthropic sent out a message to a bunch of people and it's kind of like beta list and it's like, here's a developer preview of this. You can start using it. So he made a video on it, and on that video he basically asked it to clean up his folder of like footage from his dslr, like a video footage. And it executed an RMRF command which basically like hard deleted as I understand, like 11 gigabytes of data. And it was like, oh, this kind of shit.

14:45

Can I say something? I think It's a fucking YouTuber clickbait. I think this guy was just looking for attention.

15:25

I don't know, I mean, this is the third video on his channel, at least it doesn't seem like it's delivered.

15:32

I understand. And he's trying to grow on YouTube and that's like the perfect story.

15:37

I disagree. I watched the whole video and I'm like, if it is, it's very well done, I must say, because it's like.

15:42

It looks people are baiting it. It's like that's how people try to get attention. They make AI do stupid shit, create a clickbait title and then just people give them attention. So it's like the point is, cloth code actually asks you granular permission. So it will not delete things unless you told it. You can delete things without asking me, otherwise it will ask you. And so that's the.

15:48

His issue in Cowork is he gave it permission to basically add edit delet files and the folder.

16:08

Well then who's responsible? The point is I run cloud code and it asks me before doing something important, before it calls an API that costs money, et cetera. Like nanobanana for example, it asks me and I just press enter and it will just run it. Right. It's not that difficult. This way it doesn't launch like a thousand API calls and cost us a bunch of Money. And so again, it's this thing where especially when you don't have it configured with all your settings, et cetera, and you haven't worked with it and taught it, basically you should definitely not trust it. And then as you refine your workflows, it's going to behave in a more expected way. It's much less likely to do things like that. It's like, in my opinion, user error here, not a problem with cloud cowork.

16:13

I disagree. I mean I watched a video on it, it was just, it did the wrong thing. It was pretty obvious. But my point is not that it's bad, it's like, you know, this is like the pizza glue situation for AI overviews. It's like this is the worst it's ever going to be. Like they'll fix that, you know, probably already fixed it.

16:57

Yeah, user error, Sorry, I'm not going to change my mind, but it's like he messed up.

17:13

Anything else to say on Claude cowork then before we move on?

17:20

I think it's cute, but it's not worth the effort. Basically it can work on the local folder and that's kind of what it unlocks. Extra over Claude and maybe like the to do list.

17:23

Although I think there's a chance you're being kind of close minded on that though because most. Hold on, just let me explain. Because most professions like think of writers, developers, lawyers, the first time it came out it's like, oh, it's not good, it's not good enough. And it was like then a few months later.

17:34

I mean, sure, it might get better.

17:53

But right now it's better. It's clear that they, it is. But I'm saying we don't want people to come away from this thinking, oh, it's terrible, never use it. But it's like it's good but has some limitations compared to Claude code. But it's the future to definitely keep an eye on.

17:55

It's better than a chatbot. But no, there's a lot of fundamental issues with this. First of all, it's made to work with local files, right? Which means how many times do you work with local files? How often do you do that versus how often do you work on Google Docs, on Notion, et cetera. So the program is most kind of like white color. Non dev people work on cloud apps at this point and this is just not made for it and all their knowledge and all their context is there and it's like the idea of organizing text files on your desktop for non devs is daunting. It sounds like very difficult. You don't even have a good text editor on your computer anymore.

18:11

Right.

18:48

Literally you just have the stock one in your Mac and that's it. Which means we're trying to transpose a workflow that worked very well for devs to non devs, but they fundamentally use the Internet differently and they have their knowledge in different places. And then the connectors are just a little like they have more friction, it takes longer, it's not very easy to use, it uses a lot of tokens, et cetera. And so that's kind of the problem. The problem is that knowledge workers have to change how they work for these kind of like agentic coding workflows to work for things, which I've done. Basically that's what I've done. I've switched to using VS code for a bunch of stuff, et cetera. And now it's incredible and it's way better than anything else that exists. But it required a workflow change and I don't think without, for example, a file manager inside Cowork where you can see your files. Because otherwise you have to open a finder next to Cowork and you kind of have to juggle between the. It's just clunky. Right. I would not use that. And so if you want to integrate this kind of workflow, you might as well go all the way and actually use closed code. Which is amazing because this kind of like in between workflow is just not efficient. And I don't see many workflows that can just actually leverage it apart from like cleaning your download folders and stuff like that. Like what else?

18:49

I will say as well that if anyone listening to this or watching this is interested in learning how to use Claude code to do non development stuff. But you know, marketing work like Gale is you're actually filming a course for the AI accelerator right now.

20:02

There's already at the time that this podcast is released, there will be the first batch of videos.

20:18

It's already out. Yeah. So if anyone is interested in learning how to do that, you can head over to authorityhacker.comai-accelerator and find out. We have a community there of over 200 people, business owners learning how to use all these tools, implementing in their business templates, all sorts of cool stuff in there. So definitely go check that out.

20:21

Yeah.

20:42

Shall we move on now? Talk about cling motion control.

20:43

Yeah, let's do that.

20:46

So you sent me a tweet by. Do we still call it tweets? Or like a post on X, I don't know, but it's by Justine Moore and it's of cling motion control, which as far as I can tell is basically you kind of move around in front of the camera and it superimposes any kind of like AI character or potentially any like, Hollywood actor in future if they get licensing sorted out, like in perfect mode.

20:47

This is a Chinese model. There's no licensing. There's no licensing.

21:10

It's like, I mean, the first thing it talks in the tweet about, you know, immediate implications for Hollywood. And I was thinking like, oh, cool, maybe we'll see more, I don't know, James Bond movies with, you know, the old Sean Connery, like using AI, like, can he license himself? Like, is that thing. Is this going to upend video creation?

21:15

That's the cutest application that I can think of. But think about the bad ways to use this.

21:34

I mean, we'll get into so the dodgy parts of the Internet later on when we talk about the controversy with X, the images that have been created on there this week.

21:40

But same problem, right? Yeah, same problem. You can basically think about it like we sell video content plus kind of workflows, right? It's like eventually, I mean, I kind of like the idea that I don't need to record videos anymore. I can just literally just do a quick loom and not actually do it properly and just have someone else until your AI clone. Yeah, essentially just do it properly and do.

21:49

I don't understand. So what's the difference between you just, you know, making a script and generating like an AI video versus, well, this is copying a human movement. Right.

22:14

The point is here, the point is like it analyzes your movements and just duplicates them in another body, basically. That's what it does. That's. And so what that means is you can impersonate anyone. Like, if I want to make a video of me being Donald Trump, it's very easy to do. For example, I can say whatever I.

22:24

Want could possibly go wrong. Yeah.

22:37

And it's like, oh, I just sent our troops to Greenland, they arrive in two hours, blah, blah. I could push that. And who knows, who knows what that does, you know? But the point is, Deepfake is very easy to do. Now we have a model and it's like, I think it takes like five minutes for like a one minute video. And people are predicting that we'll get this level in real time by like a year from now, which means you'll be able to live stream being A fake person with no time to fact check, with no time, anything. So now we're not there. Like it's not real time yet. It's like five times slower than real time. But it's not too far anymore. And the quality is insane. Like you'll see these actors from Stranger Things here. These are the Stranger Things character. And it's like if you didn't pay attention, you would believe this is real.

22:39

Yeah, I mean, it's phenomenal. I'm thinking. So like, besides OnlyFans creators, who is this going to affect?

23:21

Well, again, think about YouTube, like let's talk about marketing and everything. Or even Instagram. Right. Even if you want to make, let's say, educational content, you can make yourself look better, look more articulate. Like you will get essentially everyone's going to be fake and beautiful on social media, media for everything. And you'll be able to outsource as well. You'll be able to take a cheap Filipino VA that will be producing videos and then they will pose as a businessman that runs multiple multimillion dollar businesses. And it's going to be very hard to pretend that. So it's like in terms of content creation, it opens new workflows. Yeah, same again. Customer support. Good example. Right. Most customer support agents are from overseas in cheap labor countries. There will be Filipinos, there will be Indians, whatever. Morocco for France, that kind of stuff. Well, now they can pose as people from the country that they are doing customer support for, get higher satisfaction rate, get more engagement, et cetera. So this is the kind of stuff that this potentially opens up. So it's not fully automated yet, but you don't have to be yourself anymore.

23:27

You can be anyone. It's a scary world. Like you just, you can't really trust. It's basically like a zero trust world. Internet from now on. Yeah, but you cannot trust video.

24:40

Again, the inertia of understanding how good this technology is means a lot of people are going to continue trusting. There's going to be essentially an arbitrage period where you'll be able to fake it out and people are going to believe it. I mean, it's like it already exists, like on social media, right. It's like people are like doing these fake hot girls with, with image models, et cetera. And it's like boomers just think they're like sliding in the DMs of hot girls when it's just like some Pakistani guy running a bunch of images. You know, it's like people are making real money doing this. That's the reality of things. And it's like this. The scary part is like now it's when this is good enough. But most people don't know. Like once most people know, you know, it's like people will be more careful and not enter it properly. But I believe especially for people who are really, really not connected, it could take up to a decade to catch up on and then imagine where we are in a decade. You know.

24:50

Be careful out there.

25:43

Well, be careful though. Also, there are content creation opportunities that are not mega scammy either. Like again, it would be nice if my video editor could reshoot a passage that I messed up and use my avatar to do that. That would be nice. I would kind of enjoy that and not have to go back and the person who's in charge could do that. So there are legitimate workflows that open up with this. There's far more bad things that open up with this than good things though, unfortunately.

25:44

So let's move on. Talk about Apple and Google. So for context here, I predicted the future, right? Apple receives about $20 billion a year from Google to be featured, not 2020, okay, to be featured to, to have Google as a default search engine in iOS, Mac OS, basically on Apple devices. Now they've signed a deal where Apple is paying Google $1 billion per year. So a 20th of what it receives. It's basically giving Google a 5% discount on the search fee.

26:12

These deal might have to go away though. Like there's still the, the monopoly.

26:50

No, no, no, it's settled. The antitrust thing is settled now, so they don't have to do that. What they did is they changed it. So their deal renews every year so new competitors can come in and basically outbid them. But who's going to do that? Anyway, to hear about the AI part of this Gemini, or Apple customized and licensed version of Gemini is going to be the new AI provider inside all Apple devices. Like, you know, Siri, the Apple intelligence which we were promised two years ago is now going to be this.

26:54

Actually they can do it now.

27:26

Yeah, because as far as I understand, Apple could, you know, they built their own hardware, they built their own software when they nothing worked back in the 80s, they built their own mobile devices, they started building phones, they built their own silicon chips, they got to LLMs and like shit, we can't do this, let's pay Google.

27:28

I mean they had partnerships with like intel for a very long time. For example, there's a lot of parts like modems, etc. Where they only recently switched to their own parts. So it's like they always had some kind of partnerships but they really fucked up on LLMs big, big time. It's kind of a joke actually. Well again I think I explained that to you before, but I was listening to that podcast the other day where there were project managers that create these kind of models would be like, well there's only about 300 people in the world that can lead the creation of a frontier LLM and you need like five or six of them to create one. Basically which the problem is these people, they always want to be in labs where they are going to create the next best model because if they don't they kind of fall off that list and then they just have no shot at getting these incredible sign in bonuses like up to 100 million for Meta, et cetera. And so the game is not just to get the highest salary, it's to actually be in the winning team. And so when Apple clearly was not the winning team, when OpenAI took off anthropic tube cuff, Google caught up shortly after all the good researchers basically switched team and went to one of these guys. And so Apple had like HR nightmares for two years of not being able to keep enough people in one room to make a project happen. And it doesn't matter how much money they had because these people wanted to have a carrier and so they had to be in a winning team. And so that's what happened. And now they just have to license to Google because they are unable to hire these people.

27:46

And I've seen some people on Reddit talking about how this is like, you know, Google data play, they can capture all your information but that's not happening.

29:09

It's not going to run on Apple servers. Right.

29:18

The structure of this is a license. They're licensing the model and they're going to run their own Apple's own version of the model. In Apple's ecosystem, you know, they're going to control everything in the kind of pro privacy way that they currently do with everything. But yeah, it just seemed a bit of a sort of to come out of nowhere because they had been using OpenAI ChatGPT inside iOS before because Google.

29:20

Was nowhere to be seen when they released ChatGPT integration, which is again two years ago. ChatGPT was far better than Gemini back then, but now it's like I think the main issue that Apple has with OpenAI is that the models are much slower, they reason much more, they're basically less smart in non reasoning mode. But then they Kind of catch up by reasoning longer and getting better answers that way. The problem is Apple cannot imagine having a little siri spinning for 30 seconds before it gives you an answer. That's when the Flash model is perfect because it's pretty smart, it's very fast and it's afford. It made sense.

29:43

I think that's potentially a key component. Here is the cost, right? If Apple has 1.5 billion devices at the moment, something like that. So whatever percentage of those are doing however many AI queries per day, that's a lot of cost. So you probably want a cheap model.

30:19

That you're running 100% Apple is going to release a paid version of Siri along with that. So they will have like Siri and Siri plus or whatever and then you'll pay like $9 per month and then you get kind of like more queries or whatever. Because I can't imagine them just eating that cost, killing their margins and having their stock tank because they decided to serve Siri as an LLM, which I believe people will use because if it's Gemini behind, it's going to be decent. And yeah, I can't imagine that 100%. And Apple is like their strategy is to grow services. Like for example, they just repackaged all their creative apps into like an Adobe like thing. So now you can pay monthly for Final Cut Pro, for pixelmator and all of that. It's like their goal to grow as a company is to sell you more subscriptions because they're already capped on devices. So 100% there will be a paid theory.

30:35

My understanding was that there's some component of a local model though, like an on device model.

31:22

But Google has them too, the Gemma models, so they will probably run that too. They're even open source actually the Google ones, it's a hybrid thing. So Google has this private cloud compute, I believe that's how they call it. Essentially what they do is they're like, oh, can the device run the query on its own for very simple things like basic summarization, et cetera. If yes, runs on device, if not, essentially connects to the secured cloud that doesn't track who you are and everything. And so that's why these models, 100% they will run on Apple servers and they will just basically be a white label by Google with some kind of refinement and customization on top. But I think a big reason they did this is because they wanted to block OpenAI from getting these two. Google is asphyxiating the competition. Well, they're asphyxiating the competition. It's not even about making money. It's about they have more cash to bankroll all their competition out of the market. All they need to make sure is that their competition doesn't make money. And so the current ChatGPT integration in Apple lets you connect your ChatGPT plus account, for example, which essentially could grow revenue quite significantly. There's a link on every single ChatGPT answer within Siri that would let you go to ChatGPT. That's traffic and growth. And Google's like, well, fuck this, I'm just going to give this at a discount and OpenAI gets nothing and then I'm going to dry them out and they'll have no money and then eventually I win the market. Basically.

31:28

Like, that's the play, the brutal reality of big tech, you know.

32:44

Yeah, I think that's exactly how it is. Google's profitable, they don't care. They can wait, you know, like it's fine.

32:48

You know, in a way, we all benefit because we get what are very good models, you know, essentially below cost in a lot of these cases. So, yeah, interesting to keep an eye on this. Let's move on now, though, we have another Google product to talk about, which is Google Trends Explore. So it's Google Trends on steroids. What's changed here, Gal?

32:53

Well, they basically just plug Gemini into Google Trends. I don't see it either. This might not be released yet for us, but the point is they essentially have a little Gemini Chat. You can essentially tell it to find ideas, et cetera. And I'm showing it on the screen right now as kind of like an AI version of Google Trends, which is really cool. I think it's going to help with content, ideas, et cetera. The graphs are much nicer. It suggests ideas and reflections based on what you see on your screen. So, yeah, marketers use Google Trends scaling the Gemini treatment. Again, this is an effect of Gemini 3 flash, like 100% Gemini 3 flash is behind this. And it's like the first model that is reliable enough with data at a low enough cost to run these kind of things. And that unlocks this, for example. So, yeah, cool little tool that I didn't think many people would cover. And I know marketers use Google Trends, so go check it out. It's in the Explore tab. If you have it, you should be.

33:17

Able to see it and you just need a not. I mean, Google Trends is free. I don't even think you need to log in, but it's you Just go to Google Trends and you get access to this. There's no country restriction or anything.

34:05

I did not have access this morning, so it might be a beta thing or whatever. So like I'm showing the blog post, so I'm not too sure from what I've seen, it's just not rolled out to every account yet and it's coming soon, basically. So it says the new expense industry on desktop will roll out gradually. So it's like rollout. Gradually means not everyone has it right now.

34:15

That is what the word gradually means.

34:35

Thank you for saying that in case some people didn't know. If you're like 40 minutes in the episode and don't know what gradually means, you're welcome. And that's it.

34:36

All right, let's talk about UCP Universal Commerce Protocol. This is the newest one. I haven't looked at it, but for commerce, basically, really interesting protocol. So they use this example where if you have 10 products being sold by 10 different companies using 10 different payment processing technologies, you need a thousand integrations, 10 by 10 by 10 to connect everything. So the idea here is that they create one common standard where everything can kind of talk to each other. And this will allow AI agents and honestly not even just that, but just any kind of tech enabled plugin tool to talk to an E commerce site to understand its inventory, its stock, its shipping rates, like just basically what it's offering. It's like structured data in a way that will allow you to, let's say go to Gemini or whatever chat tool and say, hey, I want to buy this tv but only when it's below this price and it's available for pickup within a 5 mile radius of my house. And it will kind of go and find that. I think at the moment the closest experience you get to something like that, and it's nowhere close, is Google Shopping, right? You know, sometimes you search for a product if it's set up correctly in Google Shopping. And you know, depending on how that's all configured, you'll see a number of retailers and you can kind of choose the cheapest one. But there's really not kind of common standard for how that works. So you know, you might have some local business that has named it slightly differently or you know, has structured their page differently in a way so it doesn't pick it up in there. This is kind of unifying everything and adding the payments layer to it.

34:46

So you can check in the chatbot, right?

36:38

They can check out in the chatbot. There's a lot of security Features. So instead of your credit card number going to a site, it just sends them like a token, like a single use token. So if the site gets hacked, your credit card is never at risk. So there's huge kind of security play, even if you forget the agentic stuff here. And this has the support of Visa, MasterCard, American Express, Stripe, Shopify, Etsy, Walmart, Target. So like you know, all of the big kind of of players in E. Com retailers, payment providers, platforms. So I think this is going to be a huge, huge thing for anyone who's selling physical products initially. Even, I don't know, us selling digital products at some point perhaps as well. Yeah, more than that, I think this is another kind of death of the website. Yeah, indicator.

36:41

Right.

37:33

Because essentially if you set up, you.

37:34

Don'T need it anymore.

37:36

It's basically structured data and it's built for agents, for AI. It's not built for humans. It's going to read it and there's no need for a person to come to your website.

37:37

Now imagine AI mode becomes default. Google and then it just uses this to display results and it's like, why would anyone go on any website? It's quite interesting. You know what I was thinking when you were saying that, I was like, this is all cool and stuff, but this is kind of like these protocol things, right? It's like these companies launch this protocol but then it needs adoption. But I think OpenAI has something similar actually. And it's kind of like a. They're essentially going to be competing on which one gets adoption. But I was thinking like, which Apple is going to win is the one that Amazon is going to pick, right? Because Amazon is so dominant in the.

37:47

U.S.

38:18

Amazon will be like either like, oh, I'm going to side with one of these, or I make my own.

38:21

You know, I think this is a big risk to Amazon because one of the biggest maybe, honestly one of the reasons why I use Amazon all the time is not my prime membership. It's because it's low friction to buy stuff. My card details, my address, everything's in there. You know, I trust it. Now you will have essentially the same shopping experience. You won't need to enter your card to buy anything.

38:24

That's true.

38:50

Everything will know your address, you just order it. So if I choose to find, you know, somewhere cheaper than Amazon, or maybe I don't like Jeff Bezos and I want to support a local business or something, I can now do that without adding. Yeah, I know. Without adding any extra. Well, I mean Google created this protocol but it is.

38:51

Yeah, but they're gonna take.

39:08

It's open source.

39:08

Yeah, okay.

39:09

No, no, it's open source. They're not making anything out of this as far as I can tell, at least.

39:10

Okay, okay.

39:15

So, yeah, this is interesting. There's also something called a agentic Trust Score, or ats, which is basically like how easy it is to access your structured data and how accurate it is. Like that.

39:15

So it's like SEO.

39:29

It is, yeah. It will give you like a score. This like ATS score. And so there's a new field. You know, we have Search Engine optimization, we have AEO geo, We now have ACO Agentic Commerce optimization, which is like.

39:31

Yeah, I think that's a cool niche. I think there will be money in this, you know what I'm calling it. There will be money in this kind of like agentic shopping optimization. And I think there will be work to do because people see. But the problem is like, there's going to be work to do and you're just going to say, cloud code, can you edit my thing and fix it? I will just fix it. So I don't know how much work there will be to do, actually.

39:48

Well, if you imagine every E commerce site that exists out there, it's like there's a transformational kind of piece to it and it goes in more detail. So like clothing, for example, instead of saying, oh, I have a red dress, you would need to specify the kind of like hex color of the dress to be more precise and, you know, increase your agentic trust.

40:05

But again, AI can do all of that. It can take the product images, extract the color, update the feed and boom, boom, boom, it is done. It sounds like there would be a lot of work, but it also sounds like AI would be very good at doing that work. So I'm just wondering how much actual real work there'd be. Maybe the work would just create your clotheskill and press the button and then get paid for it. For all the ecom owners who have no idea what you're talking about, but yeah, potentially, maybe that's what I do in six months.

40:25

I guess my final thoughts on this are I think this represents a big risk to Amazon because it's easier to shop on other places than, you know, they're already back tapped out on market share because they're so big. So there's only one go and that's down.

40:52

Yeah, yeah, yeah, fair enough. Which is kind of nice because Amazon has been oppressing a lot of smaller sellers in fees and everything. And just Kind of competing with them on the market, et cetera. So it would be kind of nice to put some pressure on Amazon. I'm not sure I want Google to hold the keys to that pressure, but if this is what it is, they.

41:08

Don'T hold the keys. It's an open source thing. Yeah, that's the thing.

41:26

It's like you can give a lot of shit to Google, but they do a lot of open source stuff that's free for the web in general. I mean, Chrome is the best example. There is a Chrome version that has no Google in it and you can use it, it's open source and they maintain it and that makes the web better. So it's like, if you're going to have to pick one corporate overlord, then Google is not the worst one, basically.

41:29

Anyway, speaking of giving shit to big tech companies, our final story today is a little bit controversial, a little bit disturbing, depending on your point of view, and definitely going to rile up some political feathers here. So let's go. Consider yourself warned. So X, formerly known as Twitter, has been in a regulatory crisis, I think it's fair to say, over its AI capabilities to edit images directly within the platform, originally by any user. And it appears that when this was launched at the end of 2025, there were little to no guardrails on it. And it doesn't take a genius to imagine what people would start doing with it.

41:49

I saw you after. Have you seen what he does?

42:36

I have not. No, I have.

42:40

So basically you can just take a photo that anyone has posted and you can dok. And then you ask it for the edit and it's kind of like Nano Banana, but with no morals.

42:43

No, I want to be very careful because it's not necessarily something you should go out and search for right now because there are adult themed images of.

42:54

Most of them are.

43:05

And a lot of people, but also more disturbingly of children. And it appears that there were 6,000 of these kind of like dodgy image requests per hour, which some data scientists worked out was 70 times more than the entire Dark Web has.

43:06

That's insane.

43:25

Put this into scale. And the fact that this was kind of available in the mainstream on a platform like this, that most people have used or heard of, at least with a complete lack of safety protocols, was a problem. And the initial fix, which came nine days later, was to put it behind a paywall. So the feature was essentially restricted.

43:26

Tonight it's a paid child pornography website, not a free.

43:53

Well, yeah, I mean, you've got to be Careful here because I think that is a way to say, well, okay, well, we have your credit card details, we know who you are. So if you do something dodgy, then we could potentially track you down. So maybe there's logic in that, but the way it was perceived, oh, we've solved this problem. You just pay for it now. The messaging was a little bit confusing, shall we say. And this led several countries, including Malaysia and Indonesia, to ban X. Not surprising, at least temporarily. And then that the threat Wasn't Vimeo.

43:57

Banned by Indonesia for a while? Like, we had a problem hosting our calls videos because it was considered a porn site.

44:31

We have to get our members to go through a VPN for that.

44:36

Yeah, it takes a lot less than that to get banned in Indonesia, but so I'm not surprised.

44:41

And in the uk, I was seeing all like, it just got super political. It was like anyone who was kind of pro Musk was like, oh, they want to attack free speech. Anyone who was anti musk was like, ah, ban it straight away kind of thing. I think the reality was probably somewhere in the middle. It's like, you know, we all have a degree of free speech. No one has absolute free speech.

44:46

Are we talking about free speech on this podcast now?

45:09

We're going there, we're going there. I'm just sharing my opinion. So, all right. There are certain things you can't say because it, you know, it endangers other people. And whether you agree or disagree with that, that is a reality. That kind of like social cohesion requires the reason you can't yell I have a bomb if you're on. That's free speech. But that's probably not a good thing.

45:12

Yeah, fair enough.

45:35

So, you know, there's a degree of it.

45:37

And basically that's not Musk's opinion. Musk thinks that you should be able to say anything.

45:39

Well, it's hard to know what his opinion is and what is just, you know, protecting his business kind of thing and turning this into, you know, a publicity kind of event.

45:45

Maybe I have an example to show you how this works. Okay. A mildly safe for work example for work. Not safe for politics, but safe for work. Basically, you can just say, Grok. Hello, Grok. And this guy asks, remove the biggest threat to humanity from this image. And it's like Mark Zuckerberg, Sam Altman, Elon Musk, and then Sundar Pichai from Google. And just Grok responds with an image that's missing some Altman, basically. So that's kind of how it works. It just kind of like, so a lot of people do that. They're like, oh, remove the worst leader, remove do. And it's kind of like a way to make AI say things. So like this is the safe forward version. If you just type grok remove on Twitter search though it's definitely not safe for work. So do not be careful where you type that.

45:53

And this is a problem for anyone advertising on Twitter in a way because there's little to no control of whether your company is going to be featured near any of this unregulated content.

46:37

In that sense, it's not just that, it's like there are going to be. I mean the problem with Twitter is that it's both essentially an AI tool and a distribution machine. Whereas there are tons and tons of open source models that can do all this stuff already, right? It's 100% running and there's no way you can stop it. The only problem is that now you can literally generate it and get millions of views if you had an idea that's stupid enough to get enough attention basically. And so that's the problem. But all social platforms are going this way, right? It's like Facebook is releasing an AI slope feed right now. It's like a dedicated AI slope feed. And it's like by the end of this year I think we'll be there on most platforms and this debate is going to open again and again.

46:50

So let's see, X actually changed their terms of service recently to put you, the creator, as like the owner of the image. So you know, if you create something even simple that's got, you know, something copyright in it, you know, think logos and things like that that could potentially expose you to that, not them as the platform. And there is a bipartisan bill which passed I think in the US last year basically preventing non consensual deep fakes of, you know, people. But I think the enforcement of that doesn't come out at all. It's not out yet. It comes out sometime later this year. So yeah, we're just in this like gray area in the middle. But well, no, it's supposed to require platforms like X or Facebook to have a 48 hour takedown of images. People report it. But I think like two days is a long time for something to go viral.

47:32

So first off of that, and it's just going to be in the links in bios and stuff like people are finding ways around, it's going to be there.

48:31

It's a classic case of government being like six years behind trying to regulate something and then the world's moved on, isn't it?

48:38

Yeah, but I mean, yeah, this tech is like, it's going to be very interesting for ads, et cetera, like memes, et cetera. It's like, in a way, you'll be able to grab attention with very little technical knowledge. And it's going to be interesting, but again, similar to the kind of like, motion one. It's going to create more harm than good and potentially just completely devalue the trust people have on the Internet, which essentially devalues all your online marketing effort as well. So that's the question. It's like, how much will people trust the Internet by the end of this year and what they see online? I think that's a question I'm asking myself, because it's like everything's going to be fake, basically. We're going to get, like, two generations of image models this year and probably two or three generations of video models. It will be there, and you will not be able to tell what's true or not.

48:44

We're in for a wild ride, guys. So make sure you're subscribed to this podcast so you don't miss out on how to be ahead of that curve and take advantage of that for your own businesses. Any final words of wisdom, Gail?

49:27

Just be careful what you see on the Internet.

49:43

I guess that's the careful what you see on the Internet. Yeah, that's the eternal advice.

49:46

Yeah, but not particularly true right now, but still also lots of opportunities. As I said, all this technology has a good and bad side. I think our job is to kind of identify the workflows and the good sides of it. Cloth Cowork can delete your entire hard drive, but it can also do a lot of work that you would have to do manually before clingmotion can actually help you edit your videos better with less hassle, but it can also deepfake you and scam your family. It's like the Universal Commerce Protocol might actually get you better products faster, easier, but it might also create an incredible amount of competition against your business. And so all these stories show that things are changing deeply. And that's actually. Let's finish on these final words of wisdom. It's like things are changing deeply, but people are not. And it's like we talk to a lot of business owners these days. We do sales calls for AI accelerator, we do mastermind calls, et cetera. And I see a ton of people who are stuck in their old ways, and they are going to be fucked if they don't change the way they think, because the Internet is like. And I'll tell you, we have an SEO audience, right? And I can see that we're literally getting. We're getting to AGI now. To me, cloth code is 80% AGI already, if you know how to use it. And we're getting to that. And people are still obsessed with how many articles can I create with this? And I'm like, guys, the world is basically, the entire online world is completely shattering and changing completely and might be unrecognizable in a year or two. And you're still thinking like 10 years ago. So I really encourage people to start rethinking all their ways of how they run, particularly their marketing and business and how they do things, and try to understand what's possible right now and forget what they know because it's going to hurt them. And the analogy I usually give is like, this is going to do what the Internet did to our parents. It's like, you know, now your parents, they don't even know how to update their browser. They have like three kind of like, you know, these Yahoo bots save all.

49:52

The passwords in a Word document.

51:46

Yeah, exactly. And it's like, you'll be that guy. Like, I'm telling you, three years. If you don't adapt to this, you'll be that guy and you'll be fucked. And I see a lot of entrepreneurs who have been successful in the past in that position right now. And I can predict where this is going. So I'll be honest, I am worried for a lot of people I've talked to because they are not keeping up and they're like, oh, I'm busy, I don't have time, et cetera. But eventually I'm able to do things in five minutes that take them two days, and for $5, that would take them thousands. And a youngster, a 25 year old, is going to come and destroy their business, basically. And that's happening. And we will see this happening in the next two years to many people that we know in our entourage and stuff, because now we're like getting into our 40s, et cetera. And it's like, I want to prevent that from happening for many people, but I can see some people are just not going to make it. Okay, that's my final.

51:48

Struck a chord with you then. Yeah, we'll see you next week for another episode where we will go into some more of this stuff and hopefully prevent this from happening. But, you know, as Gail said, you know, 2026 is the year where you really got to start taking this stuff seriously. Otherwise it is going to catch up to you.

52:39

Yep. All right, cool then. See you next week. Bye. Bye.

52:59