Social Media Platforms Are Coming For Your AI Content

Everyone's trying to automate social media with AI. Just plug in a prompt, hit post and watch the followers roll in. Sounds great, except it doesn't work, at least not in the way you think it will. Today we're breaking down where I actually helps with social media, where it's a total trap and why X just launched Operation Kill the Bots to crack down exactly on the kind of content most people are creating. Plus we cracked profitable meta ad creatives with AI and we'll look at some huge Claude updates, including remote control enterprise plugins for cowork and a new model. I'm joined as always by my co host and co founder of authority hacker Gael Breton. How's it going, Gail?

0:00

Busy. It's quite hard to keep up these days with the AI industry. It's kind of interesting though, because it's changing. Before like last year you'd get like 100 product companies releasing each kind of one thing. And now we see the big AI labs releasing 100 features basically. And I think we've hit that point where they're vive coding most of what they release. And so it's going faster and faster. And as you'll see with Entropic, there's almost one new thing a day with Entropic right now, if not multiple. And it's accelerated a lot. And it's kind of scary because it tells me that it's going to get faster and faster as the models get even better. We are entering the area of consolidation.

0:42

I think that is a good opportunity to remind people if they want to hang on and stay up to date with this stuff, to subscribe to this podcast, either on YouTube or your favorite podcast platform, to hopefully stay up to date with some of the latest stuff here. But today I want to talk about automating social media with AI because it's probably the number one thing that people ask us or people imagine they will do when they learn these new crazy AI skills, or they set up Claude code for the first time, or any kind of automation. Like how do I automate my social media? I'll just do that and get millions of followers and I'll be billionaire next Tuesday.

1:22

That's it, right? That's the best.

1:57

It doesn't quite work out like that. Why not?

1:59

I mean, first of all, traffic doesn't equal money. I think people's poor business plans answer is always like, I get more traffic and that will solve my business. When most of the time you need a better offer, you need a better product, you need better everything. We've had a lot of traffic. We had hundreds of thousands of leads per month. We had all of that, et cetera. And we had time. And quite often actually there was no correlations between the traffic we would get and how much money we'd make. We'd often make more money at times where we get less traffic and vice versa, make less money. Because it's not just a matter of getting eyeballs, it's a matter of converting these eyeballs into customers and making them swipe their credit card. And often what gets the most eyeballs is what is the lowest. Like the conversion rate is often the lowest and you're better off niching down and so on. And so that's first. That's kind of like a fantasy of all the business owners.

2:02

It's like I want the case in point that if you wanted millions of followers, just do prank videos on YouTube but none of those people are going to buy your shit from your.

2:49

You post about that monkey in Japan, right? This punch monkey. I don't know if you've seen that on social media. You want views? Go and repost these videos. You'll get millions of views really easily. How much money are you going to make? No money. You're going to make nothing but.

2:57

Okay, but to push back on that, is it as easy as using AI to recreate monkey videos or whatever to get views? Because as I see it, it's not even that easy to get views using AI for social media.

3:12

I think it's pretty easy if you just pick the right topics. I think it's not like baby podcast video was one as well. Etc. But the point is most people, they're trying to make, they're not trying to get views. Like views don't pay your bills, et cetera. And so it's either you're an account that tries to monetize with the ad system of these platforms, but it's very, very difficult. You need tens of millions of views per month to consider making a small living from that. So the best way is just to run a business and get the right eyeballs and get people to convert. And that's the problem. It's like if you outsource the whole job to AI, if you ask it to both come up with ideas and create the idea and refine it, it just smells like an AI answer. Even if the models are writing better now, they don't have the. They tend to not be up to date. They still rely very much on their training data. So most of social media likes the fresh stuff like the stuff that just came out. And those models, even when they use web search are not very good at surfacing what makes the most sense and what is the most current right now. And it's like asking them to write a joke. They tend to be kind of terrible jokes. Same thing. Social media is rewarding punchy, high impact copy and without a lot of messaging and refining, it doesn't do that much of a good job. Now I use AI to help me with social media, but I use it in a way that it works like an assistant. So I have this close code skill where I do social media manager, I think social manager and then I might either give it context, like this podcast for example. Quite often I give the transcript of the podcast and it takes my words and it knows how to repackage them. Like it has a library of hooks, for example, to stop people, it knows the length, et cetera. And even then, and I'll still give feedback as it builds it. It's kind of like a collaborative process. It's much faster. Right. I have it connected to my type folder as well. So I can just send the post directly to the scheduler, I can preview them.

3:28

I mean I think there's a few elements to that if we break it down. There's this like sort of support layer in social media where you're talking about kind of organizing some thoughts, drafting, maybe scheduling, posting, repurposing that, you know, thumbnails. We've been doing a lot of recently as well that kind of stuff. It's helpful for. But you're saying we're quite far away from fully automating, hey, just do social media and then suddenly your likes follows and sales go up.

5:16

Basically if you have nothing to say and if you have nothing to add to a discussion, you're extremely unlikely to be successful by letting AI running your social media. And that's what I think now just

5:44

to home in on that. So if you do have something to say, yeah, you know, you have experience or whatever in your industry and you, you know there's something valuable there, but you just suck at thinking of what would make a good post. Remembering it can help all your good ideas or explaining it or you know, just the organization of it, is it going to be good? Or you know, is there something more like on the authenticity level that it kind of erases during that process?

5:54

I think the challenge is that when you tell AI you want to make a social media post, it by default tries to teach something valuable to the audience and information is irrelevant at this Point, information is everywhere, right? You can just type to a chatbot and you'll get a better answer than your shitty social media posts unless you have exclusive data or something. And so one thing that I've found that works really well to run social media with AI is actually to tell you we're not teaching. Nobody cares for information. Focus on emotional response. Get people to react to this and talk to their personal situation and how it makes you feel to talk about that thing. And then that got me a lot stronger post. Another thing that has worked quite well

6:22

as well, hold on, Is that not just basically rage baiting or flaming people to get a response, though.

7:00

Not always. It's like, for example, if I talk, let's say, about a new AI feature, let's say we're going to talk about this remote thing. It's like I can talk about like, oh, I can connect to my local remotely. But what does that mean for me, for example? What does that mean? That means that, look, I could say that often my best ideas come when I walk my dog, for example. And then it's going to essentially talk about that and how it unlocks this for me and unlocks this for founders. You're not like, right now, the way for most people to work is they either have a basic chatbot on their phone or they go on cloud code and they have a better setup, but it's kind of very much like a desk thing and that has unlocked that. And so the emotional response is like, oh, I can finally take my best ideas that I always forget when I get home and then just capture them in my workspace on the go. And that's like liberating.

7:06

That's not radiating, but that's sharing a use case of how.

7:55

Yeah, but that's how it makes me feel like. It's like, oh, now you can like the way did the announcement post work? The angle? No, no, no. The angle was developers can finally touch grass. Again. That was the angle. That was the emotional thing. It's like the feeling of touching grass. And that's what got people to react on social media, not the feature. And that's the idea. It's like if you tell the model to think about what this means emotionally for people, it will craft better models.

8:01

I understand, but there's. You're interweaving two points there, right? Because if I'm reading your social media and you're like, oh, I can use the remote thing while I'm walking the dog in the morning.

8:27

Yeah.

8:37

You know, and that makes me feel Great or happy or whatever. It's like, I don't really give a shit how it makes you feel. I'm like, my takeaway from that, my practical takeaway is, oh, there's a new feature I maybe hadn't heard about. And here's a use case for how I can implement that. When I walk my dog, that's fine.

8:38

But the way it's going to work is the people who are in the same pain as you are going to react to that post because they react, they like it, they feel more strongly about it. This is going to enhance its distribution and enrich more people. Whereas if you focus on like, here's a feature, like, people will not like because they don't feel something. They're not going to click the like button. They're not going to do the engagement actions as much.

8:56

I understand, but I'm just saying that the difference is from here's a thing you can do to here's what I did with this.

9:18

Sure, but what matters is engagement. If you want social media to work, you need engagement. You don't get engagement by making a message that talks to everyone. You get engagement by writing a message that talks to a small group of people that feel very strongly about what you wrote. These people trigger high engagement and that triggers distribution of that post because the network is. And so it's kind of counterintuitive. You write for a smaller group of people and that's kind of like what the model doesn't get. And so the way I've got my best posts is by niching down, talking to the emotion of that small group of people, having them react and engage more strong comment like, oh my God, I had this problem, I hated this, et cetera. Oh, finally, etcetera, People would not do that on just a feature. But it's like, it works better with storytelling. That's kind of the point of content, right? Content is now storytelling wrapped in value, wrapped in storytelling rather.

9:27

I feel as well that when a lot of people are kind of getting into social media, automating or using AI in social media, they kind of look at the gary Vees or Mr. Beast or these really the people at the very top of an industry and they're like, how do I get there? When actually saying like, well, you know, everyone talks about niching down, like, why don't I just try and become, you know, like king of this, like, small little castle over here rather than the whole industry? And that's a much more realistic way of approaching it.

10:16

I mean, that's Kind of what I do on Twitter. So, like, my Twitter is quite active. Like, for example, what I do is like, I post videos of like our workflows when we release them and I show the output, I show the like. And again, you can see the hook is very much on the emotional. I spent 3k pounds, I got 19k back. I didn't even review a single ad before launching. A little bit exaggerated. We did review them a little bit, but playing the platform here, that's the point. And the point is like, this is here to trigger emotions. And then we go, and then it's like, how does it feel? Like, old way you hire an agency takes weeks, new way you do that and after that you get into the value, how it works, what ad formats we got, et cetera. But I needed to earn your attention through emotions first. And I did that. And so the point is, people read this sometimes they don't read the rest of the post and then they watch the video. But I need to earn six minutes of their time, which is not easy. And that's the point. And you can see this post has 1k, like 122,000 views, two 6k saves and 200 comments or something.

10:48

And you used AI to write that?

11:42

Completely. But the video is me, obviously. The video is like I recorded a loom. But the point is that the copy here is meant to trigger an emotional response so I can buy six minutes of your time. And so emotion above the fold, details for people who want the details after that and storytelling and then the actual how and the technical breakdown comes later in the video, in the link, in whatever you share. Basically, because people are saturated with teaching. They don't care for teaching. I mean, this one was like money, so it's easy, but I have some that are not money. I think I posted the one on topical map, right? It's like I built an SEO consultant inside cloud code. So this is not money. Just because this is a harder one. It didn't do as well. As you can see. It only got 34,000 views and almost 500 likes. Still decent. But the point is, again, it kind of went the same way of all the way, new way. And then it's like, I built an SEO consultant Inside cloud code is interesting because it appeals to the fact that AI has reached human level. And it's like, oh, are we there yet? And it's the curiosity, the emotional reaction here is like, is this really as good as a consultant? Is this guy full of shit? Is this another X shit post? Or is this real. And so that's the point. And then the way I ground this is like, I actually use data from the tools you respect, like data for SEO and ahrefs. And then the first frame of the video is here to show the output. It shows the interactive topical map. And the idea is that again, I'm here to. Every first touch point is here to buy more of your time. And then as I retain you, that is metrics that creates distribution for social media, for example. And so again, this copy was written by AI. It wasn't me who wrote this.

11:45

Yeah, but that's my point. You can use AI to do this. It's not a barrier. It's just the way people initially try and use AI to do social media is usually wrong.

13:26

Yeah. And you need to create stories and you need to create things that stop the scroll. So you need to find. If I'm like, here's a keyword research, nobody cares, nobody's going to stop. You have to kind of storytell around it. And the storytelling is like, this is now SEO consultant level. And then the emotional reaction is like, is this game full of shit or is this real? And it works very well. Again, talking about emotional reaction, when you create content where there's two sides to it, where people are for or against it, because what that does is it gets people to comment and fight in the comments and so on. And like, ah, this is nowhere near as good as what I would do. Or like, oh, my God, this is amazing. No, you're like, obviously you guys are terrible at SEO if you think this is good, etc. The point is, like, that's what social media is today and that triggers reactions and that gets people to react, basically.

13:36

And so you've had some experiences, maybe not quite so recently, but in the past few months where you were like, I'm going to stop posting on X because, like, the AI content I'm doing there is, like, not good.

14:24

Yeah.

14:38

Like, what was causing that at the time?

14:38

It was not enough of me. Like, not enough. Like, the point is, these videos here, they like, they are real.

14:41

Were you like, tell us what you were doing to create the content there? Was it like just fully automated or were you not?

14:47

Yeah. I would take this podcast, for example, and then I would just. So, like, I would post. I would not be critical enough of the output of AI and post too many posts that would underperform. And then obviously the average reach that you get and the average performances that you get, they depend on, like, your past performances. So if you start like posting lower quality content that doesn't trigger engagement, then you lose that engagement.

14:52

And is that, do you think that's like a symptom or a cure, so to speak? Is the human in the loop quality layer? Because you could argue that that's relevant whether it's AI generated or human generated or if you have an assistant or something helping you with it. It's still like who's deciding if good shit's going out on my social media or not.

15:14

It's really taste. Right. And it's also obviously as other creators also figuring out this and getting better, the bar rises. Right. And so your fight, you can also

15:35

learn from them because they're essentially sharing what works.

15:46

Yeah. So my skill actually what it does is every week I run it and it runs apify and scrapes all my posts for that week. It has a little note taking system where it just learns from what has worked and what has not worked and it updates its documentation. So even if the platform changes, it kind of adapts as well. Like things that used to walk and don't walk anymore, it learns from that and so on. And so that's kind of the.

15:49

You essentially created a skill in Claude code to scrape your own social media on a weekly basis and create qualitative review or do you look at like data likes and engagement and things?

16:11

Yeah, it looks likes, engagement saves, reposts, etc.

16:23

And can you give us an example of some of the insights that it might give you in a week? Like, you know, this type of post does well or like what is.

16:27

I think it's mostly around what doesn't work. So like for example, like this one was just like the new model and just like an emoji. This is not AI generated. I didn't, I, I didn't need AI to come up with this, but it will see like this is underperforming, for example, so it would be like, oh, there was no value added, wasn't that good, et cetera. Same with this one I tweeted me or Opus 4.6 is getting dumber already. And that was like two days before Sonnet 4.6 was released. So like is this that? And then sometimes it works. So it's like, it's quite tricky. So for example, yesterday I tweeted this like, you know, Claude released these kind of remote control and I said this is an open claw for grownups. And this was shortly snappy. But again because there would be for and against people and that triggers an emotional reaction versus like oh, here's just a new model. Boom. 140 likes, 22 comments, 24k impressions. And so it's going to see that and be like, okay, you can do short and snappy posts, but you need to have some kind of duality in your content so that it triggers engagement basically.

16:33

So it sort of reminds me of the whole social media influencer boxing matches or any kind of rivalry or like dividing people into camps. Like it just gets attention and that's kind of what they live off of. It's. You're kind of doing the same thing here in a way.

17:35

I'm trying to keep it classy. Like I do play a bit on it. So for example, this image here that I put like the people who are the keep 4. 0 people, like they're crazy basically. And so like I kind of like made that image and just added this little kip4.0 here thing and actually didn't get. I mean it didn't get that many four people in the comments. I expected to get hundreds because last time I did it I got hundreds of people telling me that 4.0is basically their husband or something like this.

17:50

They've all blocked you now.

18:15

Yeah, I actually deleted the tweet because literally I started getting harassed from this. Literally. The point is you just do these little winks or whatever to just trigger engagement and that just works. And again, perfect meme for pro open claw people. And just like other people again triggers emotional reaction. 140 likes, 7.6k reach. And yeah, I can get quite a bit of traffic using that. It just works repeatedly, as you can see. And I mix. The one thing that's quite important is if all my content is kind of like AI generated from the scale my engagement is going to go down. I need to mix these kind of hot takes that are not AI generated. Just me shit posting basically. And that I feel adds some kind of level of authenticity to what you do and that gets you a lot more reach. So that's what I was saying earlier, which is like if you try to go all AI, it goes down the toilet. You need to have some level of effort. But you can reduce the level of effort you have significantly by having AI help you package things for you.

18:17

So can you kind of talk us through or walk us through start to finish? Like an idea comes to your head and then you're creating a social media post. Like, what are all the steps? What's automated and what's you.

19:14

Yeah, so I mean, let's just open VS Code and just show you. So this is VS Code I've shown it before.

19:28

This is where you're running plot code. Yeah.

19:36

Yep. So I'm running in bypass function. It's just faster. And I just do like Social Manager and I'll just say something. So I'll say let's make a post about how this new remote cloud allows me to finally vibe code outside. And my best ideas are when I walk, for example. So I'm just going to do that. So I would just dictate, I'll be like. So Cloud Code just released a new feature where you can connect to your terminal on your computer from the Claude app. And it's very nice because I get my best ideas when I walk my dog and I'm outside and I'm not staring at my screen. And so I would like to kind of convey that while sharing that this new feature exists, I want to focus on how I feel about it and how useful this is, even though it's a small thing. And then I want to kind of create a bit of duality in saying that this is why most people installed OpenClow and now the gap is closing. So it's basically OpenClow for grown up and serious people. So I'm just gonna say that. I'm just gonna make sure like the transcription for OpenClow might be. Oh, it actually did. Okay. And it just says cloud code every time. So that's the only thing I need to fix. So I will just fix that and I'll just do it once. And what is going to.

19:38

What software are you using there?

20:49

I'm using MacWhisper. It's like WhisperFlow without the subscription. So I hate subscriptions.

20:51

So the one off price. Yeah.

20:56

So if we go actually in the Social Manager, you will see that it has a bunch of stuff. It has like references.

20:59

This is a skill file that you created before based on stuff that works.

21:05

Yeah. So it has formulas of hooks, for example, I scraped some competitors and then as it learns, it just updates its documentation. You can see it has this learnings document that I was mentioning. Right. Like high bookmark equals reference value. Poster with the highest bookmark ratio. Practical SEO workflow, common 2.3x average generational urgency. Millionaire entrepreneur. It just takes notes and it references to that. So now it gave me three angles to cover this. So it could be a decision.

21:10

You programmed it to give you three options because it doesn't always hit the mark the first time.

21:40

Yeah. So I can be surprised that there's this new feature. We start with the contrarian, essentially start with the Open Claw angle. The number one reason people installed OpenCloud just became obsolete. I think that's a very good one. That will trigger all the open cloud people. And then cloud code just added something small that changes everything for how I use it. So if I wanted to be a bit more genuine, I would do that. If I want the highest performance 100%, that's going to be the two, actually. So let's do that.

21:45

You told it to do number two and then it's going to go and write the full post based on it.

22:12

Yeah. And so one thing that I discovered when I created this scale is that when I actually asked the main thread of cloth code to write, if I ask it to write multiple posts, sometimes I do multiple posts, let's say from the podcast, they kind of all look the same. If I ask it to write in the same thread, like it learns from the previous ones. So I've actually taught it to use subagents. So it starts like a new fresh context clothed subagent for each post. So you can see it made a test.

22:17

So they don't influence each other, you mean?

22:40

Exactly. So otherwise I would try to do some long ones, some short ones and so on. And then they would all end up being the same lens. And that would be.

22:42

This is another reason why if you're doing like one long chat thread in ChatGPT to write all of your social media content, it eventually all sounds the same.

22:49

Yeah. So I just wrote me a draft, for example, it put cloud cloud or something. I don't know. The number one reason people install cloud cloud was to use cloud code from open cloud. Sorry. Was to use cloud code from their phone. Entropy just shift that natively. Quietly. Was walking my dog this morning, had an idea for workflow, pulled out my phone and ran it in cloud code right there on the sidewalk. No OpenCloud, no tunneling, no setup. It just walked. I've been saying this for a while. Plug code keeps quietly absorbing the features. People build entire products around openclose. Cool. I get the hype. But if you're installing third party tool mostly to get mobile access, that's something Anthropic is actively building. So it's like okay, but could be better. So I'd say the kind of stuff that I like to do is make it more authentic, for example, less perfect. And the point is you don't want to sound like AI. And so I almost have it introduced clunky language sometimes.

22:58

So when you say less perfect, is that basically saying don't write like AI? Like make some huge mistakes or don't write.

23:53

Yeah. You see how I speak. I have a lot of speech patterns and so on that AI would never use because that would be better. And so it gave me a version here. Was walking my dog this morning, had an idea for a workflow, pulled out my phone and just ran it right there from the sidewalk. Didn't even think about it until after because it didn't update this one. So I can see that's the thing. So yeah, you can see it's a little bit more raw and that tends to be better. So if I like this one, I actually have it connected typefully. So I'd be like, cool. Queue it up in typefully and it's going to.

23:59

Typefully is the scheduler that you use.

24:34

Yeah. And then it will just basically send it through the skill, it's going to load the skill and it's going to send the draft. And I could tweak it if I wanted that. But yeah, that's. It just helps me transform a thought into an okay post. And maybe I would attach the video of the announcement or attach maybe a screenshot of my terminal again to make it feel authentic and then that's it.

24:36

And just to challenge you, then that post you have created right there, how would you rate that on a 1 to 10 scale relative to your.

24:57

I think it needs a little bit of publishing still, but I think the hook would get a lot of reactions

25:06

because is that like a 6 out of 10, 5 out of 10 or needs.

25:12

The rest of the content is probably like a 6 out of 10 and the hook is like an 8 out of 10, which is everything. And the reason why is because what's going to Happen is the OpenClow people will be like, oh, I didn't just install OpenClow for talking to my phone. It does all these other things that close code can do, etc. Reaction, distribution. And again, yeah. And then people who are. For people who are against. So really that's the thing. It's like the hook is 80% of your post. Like the rest of what you write. Like the hook and the visual are everything. This is just filler. But at the same time you're sharing an update and we shared it in a way that, as I said, was a little bit emotional and would have engagement from some people. Now this is slightly confrontational. Not even that much, to be honest. It's fair and it's all about. You can do these things while not being a complete asshole on social media. It's all about gauging how far you can go and how far you're happy to go.

25:16

Okay, let's talk about replies now, specifically auto replies because that's been in the news this week. Twitter or X has got this operation kill the bots thing going on. Like what is the fundamental problem here? What are you seeing on social media here?

26:10

I mean this is the head of the product of X. That said like in 90 days all channels will be full of spam basically. And the problem is that people, people who are very lazy, they auto reply on threads. Complete useless stuff as well. Like I'll open one of my threads and it works on LinkedIn as well et and these are okay, I hate

26:25

it on LinkedIn because they pre populate replies and you can just see everyone's put zero effort into the replies because they didn't.

26:44

It's fully automated. It's like this is an automated reply. It just repeats what I said in the post and say, oh my God, this is great. And again, this is this fantasy of like, oh, if I get rich, if people see my stuff, like he got 68 impressions from this, I'm going to be rich. But who the fuck is going to follow this guy because he responded something that literally has no value. This is also.

26:52

How many followers does he have?

27:16

63. And again it's like you see he has his DCTAS here and what he's doing is like if I go and check his replies, I'm sure there's like a gazillion replies everywhere. You can see he's a reply boy. That's what these people do, they just have these things running and it just ruins it. So it's like, yeah, sure, I have 200 comments, probably 100 plus aya in here. There are some real ones, but there's a lot of AI going on and that's the thing. Same this one, I mean this one's super obvious AI image, AI everything. And it's like automated reply.

27:19

And what are they trying to sell?

27:51

Some they just try to get residual traffic basically and get people to click on their profile and then they will try to sell you something on their profile or yeah, this one is not even that good but they probably warm them up and then when they reach enough reach they just add some CTAs but that's a big problem and don't do that, it does nothing for your business.

27:53

So they've been trying to stop this though because they recently in this Operation kill the bots they restricted in the API. If I'm Right. Your ability to programmatically reply where you're not. Not mentioned on all but the enterprise or above plans, which I think are going to start at like 50k a month. So that should stop people doing this still happening.

28:17

I mean, no, you can still do it through browser automation. Use a very cheap AI model that operates a browser and it's done. It's less efficient and so on. But if you use a cheap enough model, people will still do it. And that's why these responses are shit. And just repeat what you say because they are the cheapest models.

28:41

Okay, great. Let's talk about meta ads now, because we have, I would say, somewhat successfully managed to automate. I know we're talking about not automating social media, but like automate ads on social media profitably. So you want to tell us, you want to tell everyone how we did that?

28:59

Yeah, I mean, again, I think it's different to automate ads and to automate

29:17

social media, they want you to make more ads and pay for them. Right. So they're kind of not trying to block you in quite the same way, at least.

29:22

And also, your interest is aligned with the platform, which is like, you want to create ads that engage people and make money. If not, you're losing money, so why would you bother? So I think it's a little bit different. And an ad is kind of like a billboard. Right. It doesn't need to always be extremely deep thought or something like that. It just needs to have one message and send people to your landing page. And so it's a lot less bad to do that. And so what I've done is I've created a skill. And actually before this podcast, I actually decided to run it on a product that many people know. I ran it on Headspace, the meditation app, just to be like, on a decent product. What can we get out of it? And what I really like is these are the ads they generated. Actually, this one is pretty good. For example, what I really like is it actually lifts off the branding. So Headspace brand color is orange. It always lifts up the branding and reuses it while keeping the templates. Now, the way I built this ad skill, so let's actually go and check it out. It's called metacampaign. This one here is in the assets. I created a list of templates and

29:31

it's basically all the good ads that you use for inspiration.

30:42

So I just got all of these before, after, et cetera, and I trained the model to understand what these formats are and how they work and you can add your own if you want. You can just create a folder and add them. And so it does that. But what's really cool is when you run it, it actually role plays as the user of the app and it tries to understand what their problems are. So you see for Headspace for example, which is a meditation app, the Persona was A Stress Professional 2045 at the burnout point. And what specific problem does this solve for me? I can't turn my brain off. The simplest way to do it. I lie in bed at 1am and my mind is still running through tomorrow's to do. Replaying conversations from a meeting that went sideways, worrying about the deadline that I know I'm going to miss. I wake up at six already exhausted. It actually is pretty good. Imagine it's like a therapy session from your ideal customers.

30:46

I think that concept of role playing there is probably applicable in many other applications.

31:35

It's one of my best marketing tricks is to have AI role play as the customer and then that gives you keywords, that gives you ad angles, that gives you ways to solve the problem and so on. So it does all of that. It interviews me as well. You see, it asks me a bunch of questions. So these are like who's the target audience? Who's the competitive landscape? I didn't go too deep because I don't run this app. It did competitive analysis. So it actually found the current ads that Headspace is running and learned from that as well. Found all their social proof, urgency tactic, try for $0 always available free trial. It just. Just mind stuff for me. It also downloaded some visuals like their logo and then this stuff to do that and added some market research because it has all these formats. It just combines it all together and it made three ad sets so you can feel like yourself was like the first one. It wrote the copy sleep tonight, focus tomorrow. You know, we talked about the customer filling sleepless, etc. Like that's pretty good. Calm in five minutes, not five months. Your brain deserves five minutes feel like you again. That's decent. And then it's made to be coupled with this one. Not the best to be honest, but this one seems pretty good. My partner told me I seemed less intense. That was after two weeks of headspace 4.7 star. Again, probably a real problem that came out through the 14 days free feel like yourself again. Try Headspace Free like simple ad, but it works. Did this chat one as well? Like, you seem way less stressed lately. What changed? Lol. Honestly, Headspace, the meditation app Really? I know, I know. But five minutes before bed and I'm actually sleeping through the night for the first time in months. Okay, send me a link right now. And I love that. For example, it keeps the capital letters and so on. It feels like a real text.

31:43

These are these types of ad formats that just work because they don't feel like ads. So people pay attention to them more. Right. And kind of stops the scroll.

33:26

So some of them are more generic like this. And some of them, it tests that. That's the point. That's the idea. It's like create it test. It uses different concepts. We have this one that didn't fail. So it's like you didn't. Ah. Because the angle was like, oh, I tried meditation apps before and didn't work, basically. So you didn't fail at meditation. Meditation isn't what you think. Three minutes, no change, no bs. Try again. Different this time. And then. Yeah, you didn't fail meditation. Meditation failed. You has a little headspace one. Meditation isn't what you think. Missed. Like, you know, 30 minutes, 3 minutes, etc. Handwriting. Since meditation isn't clearing your mind. Chanting 30 minutes. Woo woo. What it is 3 minutes of noticing your thoughts on your couch. Actually backed by science. Headspace. Try free this and that. So yeah, I think we've kind of nailed it. And it's really nice because you can just keep running it with different prompts and different angles that you have through your mind. You can talk to the model. Just combine all this data, this research, and it works like an agency. It replaces an agency for me because you can just talk to it about your high level thing. You give it your landing page. It just does it for you.

33:35

And the images are actually Nano Banana Pro. Yeah, driving that.

34:36

Yeah. They cost 17 cents per image. So like this run was like two and a half, $3 plus.

34:40

Yeah. But I mean like if you hired an ad agency or something to do that, you'd be. That was.

34:48

I challenge you to find a cheaper ad agency. Yeah. And we've been running ads like this. Obviously this version of this is actually a better version than the one that we ran and we still managed to be profitable on it actually.

34:52

Yeah, like quite significant return on ad spend for a Martin campaign we did last month. If you're listening to the audio version of this, go over to the YouTube version and look at this because they are actually very, very impressive. And if you want this skill right now, it's actually available inside the AI Accelerator, which is our course community program that helps business Owners and marketers automate stuff in a high quality way. So we've been releasing a ton of of skills over the last month. There's stuff in there for building out topical maps. You just did one today actually or yesterday for creating like long form SEO content.

35:04

Yeah. People.

35:41

What else is in there?

35:42

There's one that's connecting to your Google search console as well and essentially will fix your site and fix your issues. We have one coming up on social media. We have a bunch basically. So it's like I will actually make a version of my skill for social media that works for most people that's going to come out. I have one that connects to analytics that's coming out as well. I've been playing with carousels as well. It's not perfect yet, but it's almost there. So we'll show it in the next episode. I want to show everything now.

35:43

Okay, awesome. But if you want that right now, it's available so you can go and literally sign up on the standard plan or if you want to apply for the plus tier. Gal and I actually run a weekly group call with our plus tier members as well. And so you can actually like chat to us and get early access to a bunch of this stuff, see what we're building and just kind of get the latest as it comes out about what's happening in the world of AI and automation. So that's an option as well. Hope to see some of you in there. Let's get back though, and talk about the remote control feature in Claude code, because you sent me a message this morning at like 6am like, hey, this is available. You need to set this up now. And I was lying in bed and now I'm thinking, were you out walking your dog at that time?

36:07

No, I wasn't.

36:53

Or was this okay?

36:55

Yeah, because I'm one hour ahead. So it's like, yeah, maybe actually, what does it do?

36:57

And is this just, as you said before, like a response to OpenClaw?

37:01

I don't think it's just a direct response to OpenClaw. I think they have better plans. But the point is Claude had Claude code on the mobile already, but it would connect to GitHub. And the problem is like, for example, if you have MCPS or custom skills set up, they may not be working through that cloud environment. But now what you can do is you can create a direct connection between the cloud app on your phone and the terminal on your computer that has all your work set up. So you just run a command on your terminal, which is a remote control and it just stays on. Like you need your computer to stay on. So even if you're on a MacBook, there's a way for it to not sleep as long as it's plugged in. And then when I write on my phone, it creates this new thread on the cloud app and I can talk to it and just essentially talk to my terminal from there. So it's like, I don't know if I can show you, but like I have this. I don't think you can see. No, you cannot see. No. But there's this remote control session and then I can just talk to it on my phone. It will just talk to my computer. So yeah, I can walk my dog.

37:06

You have to leave your computer on. So it like doesn't sleep for that to work.

38:02

There's an option on Max, if you go into battery settings on Advanced, you can say keep it on even if the screen is off. And if your MacBook is plugged, then it doesn't use your battery at all. It's not like your battery is going to degrade from this. It's just running on power directly. So you can do that. Or if you have a Mac Mini, it's the same thing. And that's why I said it's a response to OpenCloud, because people run it on Mac Minis and always on computers. And you kind of need the same setup here to do the same thing.

38:07

Oh, let's move on now. Talk about another anthropic story. So they released lots of entropy sonnet 4.6. So this is their mid tier model. Slightly less powerful than Opus, but also quite a bit cheaper. So it comes in in API costs $3 in, $15 out, versus opus, which is $5 in, $25 out. And it's also available on the free plan, which is quite interesting.

38:32

Yeah, and it's good actually. Like it's almost at the level, like I put the benchmarks here, but it's almost at the level of Opus 4.5, which was everyone's favorite model. Over Christmas I had a look at

38:58

arena lm, which is the new name for LLM arena, which basically ranks these models. And it was actually the third best model for coding on Arena LLM. Yeah, it's still good above Opus 4.5 and above GPT 5.2, only below Opus 4.6.

39:10

Yeah, it's a good model. It's kind of like for the high level planning and so on. So if you go to, for example, novel problem solving, you can see it's quite a bit below Opus. So you can see we go from 58.3 with Sonet to 68.8 on Opus. So it's a pretty big difference. But on Office tasks, for example, it performs pretty much on par. So it's like when I use it in cloud code, there's a mode called Opus Plan, so if you type opusplan, it will use Opus for planning when you're in Plan mode and Sonet for everything else. And the good thing is like, okay,

39:29

API price, but that's to stop you running out of tokens in your window.

40:03

That's what I was going to go. It uses your limits much slower if you use a cloud plan, and so you'll just get more usage. And it's a good model.

40:08

Anyone can just type Opus Plan and it does that. Or is that a special thing you've set up?

40:16

No, you just type it. It's just a hidden command on cloud code. And it makes sense when Sonet is close enough to Opus. Like before, like, Opus would be quite far ahead of Sonet, so you'd feel a Dr. Drop in quality. But now that they're close enough, it's kind of nice to use the best model to plan your task still. And then when the plan is well detailed, a model like Sonnet 4.6 can follow it quite well. And then you'll get like 20 to 30% more usage out of your cloth plan.

40:21

So is Sonnet then just a distilled version of Opus? Is that how they're creating this model?

40:47

That's what it is, yeah. Actually, I had it on artificial analysis. Right. You can actually see the new Gemini 3.1 is very smart. On paper, it's horrible to use. We didn't talk about that, but it's not good. But yeah, it is the third smartest model while still being cheap enough. What's quite interesting though is it's quite expensive to run if you use it on the API. So if we actually scroll down all the way to price, which is going to be here, you will see that actually the price difference between Opus and Sonnet, despite the price on API, if you use the API is only like 15, 20%, even though it's much cheaper. And the reason why is because it burns a ton of tokens. If you are in the process, pitch

40:53

thinking and reasoning basically takes longer to get to the same result.

41:31

It uses more tokens than Opus. So even though the tokens on Opus are more expensive because it uses a larger number of tokens, it's not as far you can see, like the same. Even Opus 4.5 was cheaper to run than Sonnet 4.6 for the same task. So if I was paying for API, for example, Opus 4.5 might be my peak over Sonnet for some tasks that require high reasoning. So if you work on APIs on N8N, et cetera, that kind of stuff. Something to consider as well is that the sticker price on the models gives you no idea of how much it's actually going to cost anymore because they use such a varying range of reasoning tokens. But still, in the chat app, I love it. It's my new default chat model on the go, or for quick questions and so on. It's good enough, it uses the tools perfectly, it's quite smart, and it doesn't destroy your limits.

41:34

Okay, and another anthropic news item this week. They really are going all out with I told you every day at the moment. So they had a big enterprise announcement presentation. So inside Cowork, which is the, let's say, normies version of CLAUDE code, inside the CLAUDE app, they released plugins, but they are now allowing enterprise users to create internal plugins for their teams. So you can essentially build a plugin, a skill, as you will, and have your team or employees in an enterprise environment use that. What's your reaction to this?

42:23

That was a common complaint from a lot of accelerator members who have teams. Like, they want to create these plugins and skill and they want to distribute them to people. And now they finally introduced a way to do that easily within a company, which is super nice. It's going to make people use it even more. And I think just to explain something

43:01

here for those who aren't. So the skill is literally just a file or a folder with some text files in it, or so maybe some Python scripts. Sometimes, yeah, you can just like, you know, copy, paste that, send it to someone, you know, we give them away to people in the accelerator. So that's not the problem. It's when you do that updating, you have. You have two different versions of it, so when you update yours, it doesn't update elsewhere. And there are some workarounds kind of with like GitHub that you can do, but it's quite complex to set that up. So they basically made this really easy to do.

43:19

Yeah, it's like, I can see, like, to be honest, like, I would imagine companies contacting me, for example, and be like, can you set up skills for our whole team now? It's super easy to deploy to people and you can build workflows that are deployable. You can control who has access to it. If you update it updates for everyone seamlessly. They don't think about it. So the person in legal that's using that workflow or that marketer or whatever, I update my meta ad skill. It's just like they don't need to think, they just come on Monday, it's been updated. They call the skill it's there and so it's really powerful.

43:49

The nice thing is they have read access to that, not necessarily write so they can't accidentally update it the wrong way and affect other users as well.

44:19

Well yeah, I can imagine. I can see consultancy for this is going to explode. Like you want to make money with AI, do that setup for companies and then just mix training and building workflows for people through the whole cloth Consultant is going to be a thing now I think they're winning enterprise just with this. It's simple enough but it's powerful enough that it's really good actually. And co op got quite a cool

44:26

example in their presentation that they did. There's a video on their site as well where there's like an enterprise company that's like losing using sales and then they need to kind of react to that, analyze all the. They use a skill to plug in rather to analyze all the data from like their salesforce. So email the contacts across their company and then present make this into a spreadsheet and then make that spreadsheet into a PowerPoint deck for management and then it goes to legal and there's a whole like discussion framework there about the negotiations and it creates tasks to send out to their or to action for their potential clients. And yeah, everyone sort of living happy happy after that. I mean I think that's good marketing. It's good marketing. It certainly seems like a cool example. I think it might take a little while for bigger organizations to actually implement all that. That's not a simple thing for slow moving.

44:50

They're going to spend big money on that. Do like I think that like there's, there's a lot of consulting gigs like for that like you can become like an enterprise skill consultant, cloud deployment consultant plus training. And it's like I think you can charge big, big bucks for that. It's like now it's like the gain for companies is massive. They don't know how to use it and cowork. I used it with a customer that I had a consulting call with this week and it's better than I thought actually. It's like, what I really liked is, what we really like in cloud code is you create the scale, but then you can tell it to update its own skill as it works. For your personal skills, at least. You can do that in Cowork too, too. And it has the agentic vibe of cloud code. But for the people who will never open VS code because that's just too complicated for them and they work in an office and it's for devs, Cowork is, in my opinion, by the end of this year, is going to be used by a lot of people. And I think that's the next wave. I can see Entropics like, oh, we already won coding. Everyone's running after Entropic on coding. And now they're like, we need another massive market to justify our valuation, which is knowledge work. And they're really kind of ahead on that right now. You can do it on codecs, you could do it also on Terminal for other clis and stuff, but they're still way ahead and it's way easier. And I'm afraid, actually, that the competition is so blindsided with coding and catching up that they're not realizing that Entropic is taking that whole other market right now and there's no competition. ChatGPT cannot do that. Not even close. Google. No, it's not. Gemini 3.0 is not a good model to start with.

45:40

It feels like Google's approach really is like they just assume or they want to pretend they know everything about you. Whereas Claude Anthropic's very much like, it

47:16

discovers as it works.

47:26

You create the thing. Yeah.

47:28

And it's like, yeah. Google's products are just too fragmented. Nothing talks to each other. Whereas Claude is like, within the app, you have the chat, the cowork and you have code and you can do all of it in one place, which is really, really quite powerful. So I think that's the way, to be honest. They're winning. And I'm scared for the competition right now. I'm scared Antropic is going to become very expensive because the competition is doing such a bad job.

47:29

It's having a look at the Salesforce share price and it's down like 27% or something this year. And a lot of these big traditional enterprise software tools, the valuations are down. I think there is real risk that. But, you know, they're just not after some of this. Yeah, yeah.

47:54

It happened in a legal, like a cloud released, like a skill, a legal skill that, like, reviews legal documents. And it was so good that the investors got scared and it wiped 285 billion of the stock market for all these kind of like old software legal companies that do most of that work, but they don't do it as good as like a markdown file of 200 lines. Lines. And that's pretty much it, right? They just imagine decades of development replaced by a smart model and a 200 line instruction file.

48:14

It's not the 200 lines, it's the model that's really driving that. But yeah, I take your point

48:46

is now it's kind of the joke on Twitter right now is that Entropic is destroying entire industries daily with their updates and they're winning. It's like we might get GPT 5.3 tomorrow at the time at which we're recording. So probably at the day we release this podcast. I'm hoping it's good because otherwise I have little hope for the competition.

48:53

And we actually have another Claude story because that apparently wasn't enough. We're not sponsored, I promise you. Claude Desktop. You reckon it is now the best vibe coding app?

49:11

Yeah, it's actually decent. It's quite buggy still. You can see they're vibe coding the app, but this is what it looks like. I kind of like made it make a website, like in one shot. It's missing the images, etc. But the point is this looks a lot like, you know, lovable or something like this. You can just talk to it and then most importantly, you can click on the website. So let's say I want to change this headline. For example, I can click on this button here and I can start selecting elements. I click it adds the thing and I'm like, change text to hello Mark. And in principle it should be able to go and edit the code for me and do that. So it's like, like it's still kind of like local code and you still need to connect GitHub and it's not perfect and so on, but now you can see that it just did it. Now the point is you can actually visually develop on. This is the cloud desktop app, right? This is like my chat and if I click here, it's just the chat and so on. So it's quite powerful. I think it's really quite good. And you'll be able to vibe code using the cloud desktop. Now I will say I've had quite a few bugs with the preview and so on. You can see I clicked around, it kind of disappeared and so on. Maybe if I click here. Yeah, you can see there's some bugs. If I click, it disappears. Sometimes I need to tell you to fix it. I think once they iron the bugs, there's no point for a tool like Lovable or something like this because it's literally the same interface with your subscription, with your same cloud subscription that you're using. And you can just dev push directly to GitHub open in VS code if you want or whatever and you have your plan mode, you have everything. So, yeah, code, code. Best of that.

49:23

Is this kind of the reality of what we're seeing when we hear about this discussion where everyone's saying, like, you know, all the model companies, the anthropic, yeah, they're going to eat Gemini, they eat everything else. Is that what we're seeing with our own eyes here?

50:56

Yeah. And because now they hit a point and you can see it's a little bit buggy. So you can see it was wipe coded quickly. But yeah, now that development cycle is so fast because AI does most of the work and not the best ones at using it, that if you have an AI tool startup idea, if it becomes too good, it will take us a week. It will take them a week. Sorry to copy you and take it from you. And because of their distribution and because when you use Entropic Claude within their surfaces, within their app and so on, you pay a lot less. If you pay for API prices, it's a lot more expensive. You can't compete. They essentially have their model at an 80% discount and they eat all your margin. And there's no way you can compete unless you have a model that is competitive against their model. And that's kind of the scary part. One of the reasons we're not going and building a SaaS right now is because, I mean, that's what I believe. I believe that it's everyone's idea. The same way as when we started having text models. People were like, we're going to make 10,000 blog posts and we're going to get so much money and so much traffic. Where are these people? I don't see them. There are a few people who got some traffic temporarily, but nothing that holds. And that's kind of the idea. If you get some traction, you get someone like, you get a company with deeper pockets, better infrastructure, just take what you have.

51:11

And it feels like the whole Amazon thing where they just looked at everyone making money selling products on their platform and then just made a cheaper version and out optimized them on their own platform. Except there's no physical barriers to Doing this so they could just in days or even hours recreate many businesses here.

52:26

But it's not even that. They can even have agents that monitor the market and mine this and just build them and then they just literally just review it once it's built. They don't pay for the. I mean they pay for the model, but they pay the cost of running it, which is very low. Like when you buy API, you pay for the R and D and so on, but it doesn't cost that much to run. A ton, A ton, A ton of tokens to observe the Internet, observe what you do, come up with ideas, brief this feature and build it. And I'm sure they started doing that already. That's why we're getting daily releases and we're getting. It's almost like the very early days of self improvement loops where the model builds the model and the product. I started seeing that with the skills I build as well. My skills now self improve over multiple iterations for like an hour or two and then I go check them out for the first time. Basically, yeah, we're hitting very, very early self improvement loops that are doable with these models. It's still not perfect. It will still kind of go to shit if you keep running it fore and lose value. Like marginal value of an extra run is very low eventually, but it's happening.

52:47

It's interesting just thinking back to the start of this episode, we were talking about like, you know, creating social media content and authenticity and all that stuff. Do you think they'll have similar problems when like replicating business instead of making social media posts, it's like replicating a business. Like how do we, how do we make it good, how do we make it authentic? You know, are people going to go for it kind of thing?

53:56

I think that's going to be the problem because you have like a thousand of everything. There'll be so much content that you'll need AI to sort through the content and just kind of present the cur. Same as what we had with Google when it came out. Right. There were so many websites. We needed to curate the websites and show us the best websites. Well, imagine like an infinity of underlying content and AI agents picking the best answer for you. Yeah, it's going to be a problem but at the same time it's like picking your business model. Right now it's so difficult because you don't know how far this is going to go. And if you're too good, if you're aiming too high, they will come for

54:17

your lunch okay, last story. And this is not a cloud one, but Notion have released custom agents, which again, I'm getting vibes of like Open Claw response to me.

54:51

Yeah, Open Claw cowork. I think it's kind of a bit of a bit of the way they.

55:03

I think they phrased it on their site, like agents that work while you sleep inside Notion. And that to me is like an open Claw response. I think everyone's trying to capitalize at least maybe in their marketing, if not their features directly.

55:07

I agree.

55:20

Talk to us about custom agents. What do they do and how are they different from the old agent feature?

55:21

I'll show you, actually, and I'll just do that. So this is what they look like. You have these new agents in beta on the sidebar here. So for example, I have this one that's like a social media post drafter I was testing. I'm not using it actively, but you have quite a lot of settings in these where you can chat with it to make edits, as it says. So it has like a page of instructions. Think of it like your Skill MD or your system prompt that tells it what to do. And it's just a Notion page, right? So you could just write and edit if you want. And then you have. What's really cool is you have tools and access. So in here doesn't have a lot of tools. It has Notion, it has web access. But if you click on connections, you can see they have a lot of tools. But they also have add custom MCP and you can connect, connect any MCP here. So I could connect my Ahrefs MCP and have it do keyword research, for example. I could, you know, let's say I could. And then, yeah, any data source that has an MCP you can connect here,

55:26

you can make anything like an N8N automation, have an MCP that you can connect to, that you can trigger an nh, like you can create your own servers.

56:16

Imagination is the limit at this point. It's like whatever you imagine you can do. But what's really powerful is like, okay, you have all these tools, you can choose which model you want as well. Like, you know, they have opus, they have, have all of that. Like all of the models are here and you can tell it which URLs it can go to. But what's really interesting is the triggers as well. So, you know, cloud code, you kind of have to type on your computer unless you create a cron job. Whereas here you can do triggers on Slack. So, you know, I was testing it the other day where I was posting on Slack and the model would just respond directly in Slack actually, which is kind of cool. You can also do that on a comment added to a page. So let's say you're working on something and you're like, hey, hey, social agent, I think there's some cool stuff here. Read this and prepare some social posts and we'll just do it and then do whatever it's told. Like create pages on your notion, Create anything you want. A page added to a database. Again, you could imagine that you have your database connected via a webhook to your contact form, for example, and someone fills the contact form on your site that creates a page on the database. Well, you could have an agent that qualifies that lead whenever they opt in, for example, and updates, let's say if you have have connects to your email tool, connects to whatever via MCP and does things for you. So connects to your calendar, connects to your email, anything you can think of. So yeah, you can build these automations and obviously on the schedule you can say like every Monday morning, every whatever. So you can build pretty advanced automations here that trigger whenever you want or when something happens in your business. A bit like. So it's a bit of a hybrid between cloud code and then a 10

56:26

in a way that trigger feature functionality is kind of missing from CLAUDE code at the moment, at least in an easily accessible way.

57:55

I agree. I think they will bring it because codecs, they have it now.

58:03

Codecs is OpenAI.

58:08

OpenAI is cloud code, let's just say, which is not as good.

58:11

Do we have a name for these like CLAUDE code and Codex coding agents, name coding agents. But doesn't that accurately represent what it can do?

58:14

Because we use it differently. But codecs is very much for coding. I would not use it for something else, which is why we don't talk too much about it. And so yeah, you can do that and then you can trigger these workflows. Now the catch is that they say it's in beta, right? So for now I can run as many as I want for how much I want. However, coming May, I think they will work on the credit system and the credit system is not cheap. So you're not going to run all your agents on Opus, because if you wanted a morning brief agent that runs on Opus, it would cost you $20 to $30 per month. And so if you start having a lot of these agents doing a lot of things, you could end up with thousands of dollars of invoice to notion that's where the model creators win again. Pay API price.

58:24

So are Notion. Is this like a revenue thing for Notion? Like are they trying to make margin on.

59:07

I think it's a bit of both. Right. It's a way to kind of like make you pay more than $30 per seat and at the same time it costs them money, they pay the API. Right. The OPS API is expensive.

59:11

It must cost them a fortune at the moment because they're not charging anything for that.

59:23

Yeah, they're in gross mode, so they're paying a lot of money. But the point is that it wouldn't be surprising to me that someone who builds a ton of these agents on Notion ends up paying thousands of dollars of overage fees every month. And so we go back to our debate on the model makers always win because when cloud code. But imagine the Force tab on the cloud app that is called Automations and all these triggers exist and then you pay your $200 per month subscription and you get $2,000 of API from that, which is what happens right now. Then Notion's product is kind of in trouble because you can do. Because there's a Notion mcp. So I can have my cloth code interact with Notion as well. I don't need to run the agent on Notion. And so it's a cool feature, but I'm afraid they will not be successful. I mean, they will be successful with companies who have a ton of money and really not tech savvy people. I think that's what they're going for. But the reality is within three months, I believe cloud code or the cloud desktop app will have some kind of automation feature that does the same thing.

59:27

Okay, we're going to wrap up there, so thanks everyone for listening or watching if you're on YouTube, if you want to ask us any questions about this stuff, head on over to the YouTube version of this. Leave a comment there. We'll do our best to respond to all that them or if you want all of these automations and skill files and things like that that we've been sharing and a lot more, head on over to authorityhacker.com AIaccelerator and join our program there. Otherwise, make sure you subscribe so you don't miss next week's episode and we'll see you next time. Bye.

1:00:30