My Notion Runs Claude Code While I Sleep

You're at lunch, you pull out your phone and tell Claude to build a landing page for a new product launch. And by the time you're back at your desk, it's done. This is possible because last week Anthropic shipped Cowork Dispatch, but this now works with Claude code too, so you can fully control Claude code from your phone. We took this a step further and built a system where we can assign work to Claude Code directly from Notion. That changes Notion from being our task manager to our task executor. It literally does all the work itself. Now we'll also show you Claude's new computer use feature where it takes over your keyboard and mouse and a skill we built that one shots entire websites. I'm joined by Gail Breton, my co founder, authority hacker, where we separate what works for business owners and knowledge workers from what's just noise in the AI space. How's it going, Gail?

0:00

I like the new catchphrase. At the beginning, it's like, did AI write this?

0:49

It did, but I kind of directed it. It sounds like something around this space.

0:54

It sounds like a local news station's catchphrase a little bit, which is like, it's both good and bad, you know?

0:59

But there's maybe a reason why all news stations have things like this. And one of the problems we had with our podcast is like, new people come and they're like, who is this for? Who are you guys? What is this? It's fine if you've been listening for 10 years, but not everyone has. So for the benefit of those people, we hope you're in the right place now.

1:06

And we got a bunch of new listeners last week, actually. So welcome to everyone.

1:25

Yeah, thanks for tuning in this week as well. If you've just found us last week, today we're going to talk about controlling Claude remotely in various ways. Some very basic and some a bit more advanced. So, Gail, let's start off, if we can, with Cowork. It was originally called Cowork Dispatch, but my understanding is that now works with Claude code. So it's just. Yeah, just Dispatch.

1:29

Yeah.

1:53

I've told a lot of people about this. I was at an event last week and. And the first question everyone said was, oh, can it control my VS code? Or how is it working? And it can't, but it kind of can do the same thing. Anyway, so you want to kind of fill us in on exactly what's going on here.

1:53

I mean, there's two things, right? There's the dispatch that connects cloud code, but it connects to cloud code on your desktop app. And it's kind of a weird one, right? Because when you use Dispatch, you basically have this one feed with everything and it just never clears. It's just one feedback and then that acts as an orchestrator that then starts either cowork sessions or code sessions on your desktop app, which with whatever setting you have on your desktop app. So if you haven't connected your MCPS on your desktop app or whatever, it's not going to work. But if you have, it will work actually. And it kind of decides what to do. Like, you don't really have too much control.

2:09

You can replicate VS code essentially by setting the folder it's working on to be the same folder that your VS

2:43

code works on, but not really. So let's imagine, for example, we have these meta ads that we showed a few episodes ago that creates images, right? To run that skill, you need a Gemini API key because it uses nanobana, right? It's not a skill, it's just an API key that it calls when it calls the API. And as a result, it's like you cannot store secrets in a cloud app, for example, right now. So you could not run that skill, for example. It wouldn't work. So it's like there are still limitations and it's still not perfect. By this time next week, it will probably be solved, given how fast these things are being done, et cetera. And it works very well if you. You have your code on GitHub, but then there's still the secret management that's still a little bit janky. You can make environments, but it's not very clear. Entropic's throwing a lot of shit at the wall right now. A lot of it kind of overlaps and it's like 80% done every time they release it. So it's like, it's kind of good, but it's kind of also not that good because it's all vibe coded. And so my guess is they're just throwing everything they can at the wall. They see what people use and they consolidate. I can imagine that eventually they will focus on the things that people use and abandon the things that people don't use and appreciate them.

2:52

So right now, how are you using it? Or how would a business owner be using this?

3:59

Yeah, I'm not using it because I built a better system on notion that we talk about later. Okay, so I am.

4:04

That aside, like, what would you be using it for?

4:11

I don't think I would use it for cloud code. I think the best Way to use it right now is to actually focus on Cowork. Connect your mcps, connect it to your folders, and then have Skill loaded on Co Op. And then you can run all this stuff. It's pretty good on Cowork. It will not do everything you can do on VS code, if you're on VS code, but it can do 70% of it. It doesn't have a terminal, but that's how I would set it up. And so I've set it up a little bit with Cowork and it works kind of okay, but I really don't like the way it works because it's basically like. It works like a chat, but it's not really like a chat. It's an orchestrator. You talk to the orchestrator and then it just spawns a session of Cowork, but you see nothing. So you're on your phone and it's just kind of like idle for 10 minutes while it's running the Co Op session. And it's like, boom, it's done. Or it failed, or it just kind of hangs because it's failed silently or something. What's good is it does have the approval thing, so if it needs access to a folder or something, it will just pop you. I think you can see it on the Twitter image here. It asks you for access and so that's decent. It works, but it still feels like you don't really know. It's kind of this weird hybrid where there's a chat happening in the background, but you don't see the chat. So you don't see what it's doing. So you don't know if it's going well or not well, you cannot interact interrupt, like cloud code. And so that's why it's like, in my opinion. Well, it's not just that. It's like, I want to read this file. And you're like, why? It's like, why are you doing this? And it's like, you don't know. And so that's why. Actually, the notion system I built, I like it better because it's fully asynchronous. So you kind of assume it's just going to go do its task, but you set it as a task and it just comes back to it. But you're not like hanging on the chat. It's just doing it basically, just before

4:13

we jump onto the notion thing, because I can see you're eager to get there.

5:47

Not that it's just like, I don't use this dispatch.

5:51

Basically one of the things that I hate a lot about the AI industry is the weak use cases. You know, they'll release this new feature and like, yeah, you can use your voice to send an email or like, you know, look up the weather or you open your calculator or something like. And it's like, that's stupid. That's a stupid use case. Are there any realistic, practical use cases that business owners or knowledge workers would use dispatch for today?

5:53

Yep. If you have any MCPS that has valuable data. Let's say again, lots of SEOs listening to us, right? So you're on a meeting with a client and you need to pull the latest data from their site and you have a small phone like I have. It's kind of complicated. You could open ahrefs, but honestly, like navigating ahrefs with all the filters sounds like a nightmare. If you have the Ahrefs MCP connected to Cowork, you can just drop your query and it will just operate the whole thing for you. And it can make you a PowerPoint or whatever if you want, for example, pretty handy if you're like on the go or you're in the car or whatever. And it can use your tools that are connected via MCP or skill that don't require API keys, just mcps.

6:21

So still pretty good. I'm thinking as well, if you're at a conference or an event and you're speaking to someone, you could pull it up there and then. And in a couple of minutes during the conversation, you could have something meaningful to show them too.

7:01

With the new cloud code integration, we're going to show the site building skill. I built the skill to start new sites. You could literally talk to someone and just prompt on your phone, it would start a cloud code. With that skill, I would spin up prototypes and then as you're having that conversation, it's building. I do that quite often on calls. I just launch these things and I'm like, I know that's what they want to do and I show them I can achieve 80% of it by the end of the call. So you could do that like a site redesign or a landing page concept or a marketing campaign you talked about. Be like, hey, you can prototype this, you can do all of that. It can do all of that.

7:14

So it's good for people who are meeting others, meeting clients, meeting prospects, like in the field, face to face, going to their office for a meeting or an event or a conference. Then there's probably quite a lot of use cases here, but for you And I, who are scared to leave our bedrooms, maybe not quite so much.

7:47

Yeah, I mean, I think it's really the idea of you can take a lot of clothes capabilities outside of having a big screen, having multiple windows, juggling and kind does it for you. And it's a first attempt at it. I personally like the idea of moving to asynchronous things where it's like you don't necessarily have to hang on a chat window for things to happen. It's getting there and that's what's happening. People are kind of working more with these tools right now, more hours and they kind of feel like, oh my God, I have all these limits left. I need to use it right now. Or I could just keep it going. It's not super difficult. I need to go do groceries, but I kind of want to keep my thread going, et cetera. You can start doing these things now as this is going. This is also a big dig at OpenCloud, obviously. Like this is basically how OpenCloud was working. So now we're like 80, 90%. OpenCloud is built into the Claude app already and there's not really much use case for OpenCloud anymore.

8:10

The obvious limitation which we have talked about in previous episodes is that your computer, where you're running Claude code needs to be on. Yeah, needs to be on and connected to the Internet. You know, you can't use it, it's not full server side processing at the

9:04

moment, but you can. There's plenty of tools to keep your computer awake. But the screen goes off, etc. It doesn't destroy your Mac battery. It's all on Mac, by the way. So it's like if you have a Windows computer, bad luck. But yeah, it's not too bad. It's just. Yeah. In the use case when you're traveling, your laptop is in your backpack and it's not plugged and everything, then it's not going to work.

9:17

Yeah, see, you weren't a huge fan of openclaw in the beginning anyway. Is it all but dead now? Like has just copied the best features and then. Or is there any use case for it?

9:38

And actually it's better because you can see like this window shows it like the asking for access to things means that it can't just go rogue and just mess stuff up basically. And that's kind of the thing is like if you want to run a more secure setup, then you can basically. Okay. So yeah, it's better for 99% of people. It's probably like A bit worse for some people, but overall, I don't think there's a point if you haven't picked it up. And if you don't have your whole Setup yet on OpenCloud, I don't think there's a point doing it anymore. You might as well just start on Claude and in two weeks it will cover all use cases.

9:49

Okay, sounds good. So on Sunday morning you sent me a message saying that you had created or you've set up Headless Claude. I said headless Claude code in Notion. And I've come across this term headless before. It's like a. I don't quite understand exactly what it means, but we now have a box in our Notion setup. It says delegate to Claude code and if you tick it, Claude code somehow, and I'm hoping you're going to explain that to me right now, goes and executes the task. So what have you built here? What have you unleashed?

10:25

I mean, first of all, it's all vive coded, so I hope things will go okay. You can see my notion right now, actually. So this is kind of experimental still. So this is like a different task database, it's remote from the rest, et cetera. But if you look at the activity, you can see that now Cloud Code is a coworker basically, and it's just tackling tasks. And so the idea is I can create a task like I would on Notion. I can give it a name. So I don't know, let's say just test task and I can just say hello in the comments. That's it. And then if I set it to todo and I click delegate to Cloud Code, it's actually going to, through an automation on Notion, it's going to pick it up. Just give it a minute. You can see that I have this automation and this is a demo. But the idea is when these conditions are gathered, then it sends a webhook to a URL I'm not going to share on this podcast, and it sends the content of the task. And what that does is it starts a terminal headlessly, which means my screen did not open a terminal right now. But I have a little thing that says that the server started and you can see that our task just moved to in progress as I was showing you that, for example. And so it's going to run my Cloak code work folder. So the same folder I work on on VS Code on the terminal or something is the same one with all the skills, with everything. And then as I update it, it will also update the cloud Code that works on Notion. It just feels like the same person kind of. And the point is that it's using the Notion cli, which is like, it's a thing they release. It allows Claude to use Notions through the terminal and it's just much more token efficient, which is what has unlocked this workflow. And the point is. Yeah, I just basically.

10:59

So can you just explain that bit? It allows Claude to use Notion more efficiently through the terminal.

12:42

It can basically get the content of a page in markdown format. So it's very token efficient. And then it just can drop comments, it can create new pages, et cetera.

12:48

It could do that before using just.

12:59

It's more efficient, it's more fluid, it's better using it. And so the idea is actually just before this podcast, I just wrote that task. I said, hey, I need to explain to Mark how this works. So I said, hey, I need to make a presentation how it works, blah, blah, blah, please make me a presentation. There's a presentation mode on Notion. So I was like, make me a presentation I can do to Mark, basically. And then it just ran for a bit and then it just dropped a comment that said, hey, I made a 12 slide presentation. It's here. And it just kind of gives me details on the run and just ran the task, basically. And so if we open it, you can see this is the presentation mode. So let me see if I can actually put it. And let's see what Claude actually prepared as a presentation. So if I click on present. Oh, this is big. Can you see? Yeah, you can see. So claudetasks, what is it, your notion to do this? Just learn how to execute itself. The problem? You have ideas faster than you execute them. Your Notion task board is a graveyard of good intentions. What if checking your box could actually get the work done? Not a va. Not a freelancer by Claude running headlessly on your Mac 24:7.

13:01

Okay, so I mean, like if you're listening on the audio version, it's basically a design or anything, but there's bullet points, there's like simple explanations of what's happening here.

14:07

And then it also made you an infographic explaining how this works, basically. So, yeah, that's the idea.

14:17

And so did Claude use nanobanana to do that or how did it make the infographic?

14:25

It used nanobanana. Yeah.

14:31

Okay, cool. So that was like one of the limitations you were saying before with dispatch is that it wouldn't be able to call external APIs.

14:33

This is a cool.

14:41

Because you're doing it this way. It can.

14:41

And so you can see, for example, my test task is actually marked as done. And you can see say hello, that's it. And so just checking as request, task received, executed and done, basically. And you can see it's just. So the point is like, why do you do that? Why is this interesting, right? And the point is when you're in VS code or when you're in a chat interface, the problem is that you kind of are in the single chat and you don't think of your projects and how you don't think of the high level of the big project that you're doing. You're working on a single task at a time. When you use a tool like notion, you can use things like project templates. So you can have, you know, 10 tasks being created in one click. And then you can also. It's a multiplayer workspace, which means that if you want to comment on that task and review what CLAUDE did or that presentation he did for me, you can jump in and drop a comment

14:43

right when I set this up, it's using your CLAUDE code to execute on it. It's a way to.

15:34

I could set it up for you too. I would need to set it up for you. But yeah, the idea as well is I've built some protection in that. So for example, it only will execute if the task was created by me. So if you go in and tick that box, but it's obviously tied to your account, there's a created by property, then it will actually comment and tag me and be like, is it okay if I run this task? And then if I reply yes, it will run, but if I don't reply or I say no, it will not run. So that's a way to protect my computer, basically. And any other integration could start creating tasks otherwise, for example, it's a little bit dangerous. So again, this way only my account can create tasks. So as long as I don't give access to my account for something else to create tasks, in principle I'm the only one that is in charge. But you can absolutely go ahead and comment on that presentation and then I can put it back to do the status and say, go ahead, do it. And it will read your comment and take that into consideration in the way it's going to revise the task. So for example, like I actually worked on building a newsletter skill, industry news, et cetera, and usually I do that on VS code, but I was like, oh, what if I run this like a project, right? I make tasks, et cetera, and just does it in this case, actually I just ask it to research. I kind of prepared the whole outline for the skill and everything and then it ran an eval on all of it. And then I'm in the process now where I'm reviewing it. So it's like actually attached all the test newsletters directly in Notion. I can highlight text and I can drop comments, for example, this headline is quite weak. Or like I say, never use EM dashes, for example, et cetera. So I can kind of review it like I would review a Google Doc and then I can put the test back and I'd be like, hey, just you update the scale based on my comments on the document, for example, and you update all that. It's kind of like nicer than working in VS code, to be honest.

15:40

Yeah, I can see that for like editing a document or commenting on a document. You know, the Google Docs style commenting within Notion is really good.

17:34

But also you can review it, you can drop commands. I can make a task for you to review to a human and I can make a task for Claude to be like one. Mark is done reviewing, you update the skill. And so you can see how this become a much better team work tool actually than just staying in VS code. And so that's why I'm just starting. There's also another thing that this has unlocked that is massively powerful, is like if you want to run a cloud code on your VS code, you need to be at your computer and type in. You can kind of schedule your own desktop app, but again that's the desktop app, that's not your VS code, blah, blah, blah. If I set a due date here and a due time on any task, it will run it at the time and date I set. Which means that.

17:42

And you can set recurring tasks in Notion as well.

18:28

Exactly.

18:30

Automatically do that.

18:30

Which means it's like. And again, project templates and all of that, blocked tasks. So I could be like, oh, Mark reviews. And then when that is unblocked, then Claude reviews it and automatically does that. So you can see how you can start building these much more collaborative hybrid workflows where multiple people can actually be part of it and Claude just becomes one of the coworkers. And that still runs on the setup that I've highly customized for my use. Basically.

18:32

Yeah, because this is like one of the biggest pushbacks or questions we get from members in the AI accelerator is like, you know, how do I get my team to use this? Or how do we use this across the team? But like, I'm Starting to see how, you know, especially for like non technical people that don't want to open VS code using something like this. Especially because everybody's already using Notion or ClickUp or whatever your system is, you know, to. To build this on like it's not a huge leap forward to do that, but it unlocks so much.

18:57

If there is someone not technical in the team, they can essentially piggyback, write my cloud code so they can set the task. I will still have to approve it, right? But it's just like I can be

19:27

on my phone, but you could set it so you didn't have to approve.

19:36

I could set it so it's just a security thing for me. So it's like again, it's always a trade off between security and access. I could set it so it could piggyback, right? You could have a Mac mini running and everyone in your team uses the same cloud code, for example, and it can run like each task is a different thread, so each task is a different one. But if I comment back on this task, for example, it will reopen the same thread with all the history, so it will not have to rebuild everything. So one task equals one thread. And so you get 1 million contexts now, which is plenty for a task at this point. So are you now convinced? Because I know before this podcast you were like, I don't really see it.

19:39

It's like, how did you figure this out? Because I haven't seen many people talk about this.

20:16

I just came up with it. I just kind of came up with it. I was like, is it possible to do this? Is it possible to do that? And I kind of came up with the edge cases, like the publish dates, like the dates in the future, the security, all of that as edge cases.

20:21

You can't hack it. You can't change the created by field on notion.

20:34

Is that fixed by who? Actually that's fixed by your account. So that's kind of a way for me to protect it. You cannot change that property once a page is created. So it's not perfect. I'm sure there's holes in it.

20:38

So for example, if you create a task, could I not just then go and edit the task? It's still created by you.

20:51

Yeah, you could. I think you could actually. So I mean, in principle, inside notion, you're always friends, right?

20:57

Yeah, of course.

21:02

You know what I mean. It's not public on the Internet, so it's like I don't feel like I need to build insane security. But yeah, there's another thing that I've done is actually Claude connected me to a tool called swiftbot I had no idea about. I just came up with it and was like oh, I can actually make you a little widget on your taskbar. It shows you which tasks are running right now on your computer so you can kind of keep an eye. So it kind of like lights up on my taskbar on the right. It's like it's grayed out now because it's not working. But then when it's working it turns green and then I can click on it and I can see all the tasks that are running. And then in my VS code I've also created an audit skill so it can kind of like look at what happened and you know I can, I have an emergency button basically so I could literally click. I could even create a task in notion to just stop everything and it would just stop.

21:03

Basically you've built that. That's not like you built those safeguards.

21:52

This is all custom built basically. Yeah, it's just me talking to AI for Sunday morning.

21:56

Super interesting actually. Yeah.

22:04

So okay, I'm happy you're excited because it's like I know before you were

22:05

like, I mean I was coming from this perspective of like there's two of us in our team, like we're trying to grow using AI, not humans. And you know, I could see how this would be super useful like if you had a team that needed to use stuff in your cloud code. But yeah, like the kind of feedback commenting thing like I can see instantly

22:08

and the comments and back and forth like hey, check this out. You know I'm working on this with cloud. Check this out Mark, whatever. This is kind of handy. You share it with someone else, you do all of that and the thing is you can go further. It's like my next step now is prompting the notion AI to write better tasks. So I talk to it and I write the tasks and as the tasks are better laid out I'm hoping to get a better result out of cloud code. So I'm going to use Notion AI as a way to administer the task list basically and all the automations that you can plug into notion as well. It's flexible enough as a system that you can really do a lot.

22:28

It really is like it's a total sandbox notion. Like if you're not used to constructing things in that way. We came off of I think todoist before this and it was like asana. Asana. Sorry. Yeah, we used to do this before that it was asana is very fixed in the way it works. So you kind of have to have an idea of what you want to build with notion, which I think puts a lot of people off of it because it's not like the task system is not out of the box, so to speak.

23:06

But right now, literally, this is a sandbox, this task list. It's just me kind of. I will expand it. I think the test is successful so far. As I said, I built an entire new skill, et cetera. It's actually good. It's like it built newsletters. I will probably release that skill. It's good. Yeah, it's good.

23:35

Let's talk about on the subject of skills, though, which is what we give away inside the AI accelerator, which is the thing we sell. But we want to talk today about building websites with Claude code, because it's changed a lot recently. We have made podcast episodes last year on some ways to do this where you would, you know, pick and choose different components and have AI assemble it all. But it's gotten so good right now that it can kind of one shot a full site. So do you want to tell us about how the website creation skill works?

23:54

Yeah, I mean, it's like a whole course I made, but I made a skill to kind of like bootstrap the websites. Whether that's doing a redesign or. Or starting a new site from scratch, you kind of do both. This is kind of the case study site that we built for the course on Vibe, coding your site. So I think compare that to most people's WordPress sites. That's live on the Internet as well. You can go check it out. It's not perfect. I didn't want to fix all the imperfections because I wanted realistic expectations from people. But I mean, this is a good site. This is your wife's website, right? I mean, it's your wife's business that I used as an inspiration for this. And it has service pages, et cetera. That's a local business, I would say

24:28

as well, if anyone's ever worked with a web developer, like when they give it to you, like, there's always a bunch of stuff that's. That's wrong. Like, you know, you notice or you want to change.

25:05

So, like, yeah, this was built in two and a half hours, I would say, of actual work, like I did more with the shooting, et cetera, when I did the course, but that's two and a half hours. And that includes hosting it. That includes, you know, hosting the images, which is something you need to do with custom sites. Et cetera. You see, there's still like some issues here. For example, I didn't want to remove them. I wanted to be honest with how I did it.

25:14

What is the secret sauce then in this skill? And like, why is it so good? Is the models better or like, what else?

25:35

The models got a lot better. The thing with AI is kind of like every time you do a design, you're kind of rolling a dice, you know what I mean? It just kind of. And so the idea was like, if you roll the dice once, you might win or you might lose, but if you roll the dice like four or five times, you'll probably have one role that you're probably happy with. So it's like I went after 99designs principle, which is like, it's a marketplace where you submit a design brief for a website or logo or something, and then many designers come up with their version of your branding and you pick the one that wins and you pay that one. Basically that's kind of the idea, right? So I did that with AI and that's what the skill does. So it's like this is actually I took one of our members website in the last mastermind called like a very old website called miracledata.com and I was like, yeah, he could use a redesign. I think we can do better. The idea is I literally just put his URL in the skill and I told it to not use the images. So it would use images. Like in a site that I did for the dog training, there was images. In this case, this was stock images. So I was like, let's not even bother with it. And I told it to just kind of reimagine this business branding from scratch. And that's what it did. It kind of like came up with like four different designs that have four directions and you can see like dark, monumental, light and warm, et cetera. And so the idea is like if you click on each one, it's a completely different direction. This first one was this kind of like dark mode city, art, sourcing, etc. Not perfect. You can see there's some issues with the animations here, but. But overall you get the vibe, right? It's a vibe shot here and not a big fan of that vibe, to be honest. But you can't say it's a bad design. I think.

25:41

I think with all first designs, you kind of have to imagine them with like images as well, because that's really what likes when the site kind of starts to take shape.

27:16

That's why this one had no images. But I even tried to make some animations, but I would probably put an image here instead. For example, like, the idea, I mean, it looks.

27:25

It looks super clean. Like, there's just like. I mean, we've made many websites, we've worked with many companies and many web designers, developers over the years, and there's just. There's something about, like, it. When the quality hits a certain point, it's just like, oh, yeah, they get it. It's. It's crisp, it's clean, they get it. And this one, when I saw this example here, it's like, yeah, okay, that's. That's at that threshold for me.

27:32

Yeah. And like, I think this one is also pretty nice, actually.

27:56

Different style, more traditional, but yeah, it's still so decent. It's nicely built.

28:00

There's a bit of an issue on this box here. You can see these were one shot, right? Don't judge too much. You can fix these things very easily. And so, yeah, that's the idea. The system is like, it creates multiple shots at your potential branding. Then you can kind of vibe code a bit and maybe change the dark green to dark blue or something like that. Then once you've picked a direction, I've made another skill that transforms that direction into a design system. So if we go back on the DOC training site, for example, if I go on design system, which is no index, you can see that the model has transformed that into a set of rules to reuse everywhere. And so the idea is, this idea is the CSS behind the scenes and all of that. It writes itself a cloud MD that tells it how to use things and so on. And then the idea is after that. So in the lesson, I actually just built a design system and I'm like, okay, now let me show you how this works. And I just. I have kind of the homepage like this. I kind of edited it a bit since then, but not too much. And I'm like, okay, now you make the about page. Just use the design system, right? And this was one shot. I didn't touch anything after this. Just came out as the about page. Now, is it like an incredible about page? No, but it's okay. This was literally not touched.

28:06

I mean, it's better than what we have at the moment. But yeah, yeah.

29:14

And it's like you don't have one actually, so it's like, it's better. And. Yeah, that's the idea. The idea is like, it's able to build more pages, any page you want. Same with the contact page. I was like, you make a page in the theme. I just made this page. And even the form is like, you know, it doesn't look like a shitty gravity form that you embedded quickly or something. Like it actually looks nice. And it's like, yeah, it's in theme. So that's the idea. And then after that you can kind of vibe code. We talk about many ways of editing your site and I know a lot of people are like, oh, but this is code, it's difficult, etc. Actually not that difficult. I'm going to share the Claude app, but you can literally edit these sites on the Claude app now. So if you just open it in the code and then you just tell it to start the preview site, you literally have this window here and you can just select an element. Like for example, I could say maybe services instead of what we do. So I click on the element I want to edit and I say change copy to services instead.

29:17

You still have to prompt it. You can't just change the what we do text services text.

30:17

You would need to go in the code for that, but you can literally just select the text and say change it to that basically. And after that I'm not going to do it. But you click two buttons and it just updates your website. So that's how simple it is for at least the base changes and it's much easier now.

30:22

There's obviously a lot more considerations when publishing a website. Things like hosting, SEO. You covered all that stuff as well.

30:38

In the course it's all covered. I know people have these things and I'm actually going to shoot lessons on how to connect a WordPress install. Even your old WordPress site, if you wanted to manage a blog on the site, so you don't even have to migrate anything, you can then move your WordPress site to A$5 per month hosting. It doesn't receive any traffic, it's just used as a cms. But I know a lot of people are like, oh, how do I upload my blog posts? I think you should just have cloud code do it for you. But if you really want to do it manually, then it's possible.

30:46

Interesting as well. And you also said that the hosting for these types of sites, because you're doing it on cloudflare, it's more or less free at the volume.

31:13

Actually, if you set it as a static site, you could get close to actually a million visits per month and it's still free. So yeah, it's very good. And that's the thing. Same with the form for example, right. Again, we talked about Gravity Forms. If you had to pay for Gravity Forms for this, it's like 60 bucks a year or something like this. And that's the problem with WordPress. You end up with five to 10 plugins that you pay 30 to $60 per year, plus you're hosting, let's say 30 to $99 per month WordPress. And then if you want any change or custom development, you pay a dev because it's not super easy to wipe code it.

31:22

And so you just don't update it because it's a pain in the ass.

31:57

Right. And so that's kind of the thing. It's like now you can literally just connect your code base into Claude and you Vibe code your changes, you click the bottom you. And then I made a video that we're going to email to our subscribers. So subscribe if you want to see more details.

32:00

Authorityhacker.com subscribe.

32:14

And then, yeah, it just goes live on your site and it's like you don't even have to open the code editors anymore. So I think that I've talked for a long time to people like, stop using WordPress. It doesn't make sense anymore. And to make matters worse, the creator of WordPress SEO, Yoast, just moved his site to this framework that I just added to the course, actually. So the guy who was one of the Most invested in WordPress just literally started moved his site completely and he's like, oh, that's way easier.

32:16

There's going to be a long lag here because, you know, WordPress is such a. I don't even know what percentage of the Internet it is these days, but like a significant percentage. And, you know, just people that like, aren't going to be involved in this or Vibe coding. But there's also, I think people that have client sites and the clients want to be able to go in and just use, you know, they're used to updating WordPress and the CMS there, but. But you're saying that they will still be able to do that to upload blog posts?

32:47

For the blog? Yeah. Not for the course pages, but for the blog, yes. But to be honest, you should probably set up your clients on cloud and just have them, if they have the cloud desktop app, just set it up for them once, then they just have to go to the code tab and just start a new session and push a change. And it's that more difficult than using a cms? Sometimes it's easier. And the models know the tech. So in a way there's kind of a guardrail that doesn't exist when you edit yourself and you're not technical. So I don't know, it's like to me, I think it's time to change, to open your mind a bit because we are there.

33:13

Okay, well, and if you are interested in changing your mind and launching a new site or landing page using this, then you can head ON over to authorityhacker.com AIaccelerator and you can join the AI accelerator there where you get this course, all of the skills or everything you need to execute on this. Like how long does it take for someone to spin something like this up?

33:47

I mean to do the site, maybe like three, four hours. I think we had someone who's really not technical did do one in five to six hours.

34:10

And a lot of that is kind of giving feedback and saying the angle you want to take, who your teams are, those types of.

34:18

If you start a clock right now and you make a one pager site live, how fast can you go? I would say 15 to 20 minutes for me, something like that.

34:25

I mean that's pretty impressive. So yeah, you can get that. Plus our whole plug and play library with all our skills, automations, workflows, prompts that are actually useful for business owners and knowledge workers. Everything we build in there is tested by us. We spend a lot of time making sure it's actually usable and it's stuff that we would actually use ourselves. So if you are afraid of being left behind by AI, fear not. You can join the AI accelerator at authorityhacker.com AIAcccelerator but speaking of fears, something which is scaring a lot of people has happened to the limits in Claude and we talked about this last week on this show where they had this double usage promo that was going away at the end of the month, but we're not even at the end of the month yet. And something kind of underhanded seems to have happened. There are a lot of reports now over the last 24 hours on Reddit and on other places. Someone's even consolidated all the reports into to this one Reddit thread of people essentially getting like a quarter of the use usage. They were, you know, not even including like the double usage promo. So like it seems to have fallen off a cliff somewhere. I think people are suspecting that maybe Anthropic is trying to kind of like secretly dial it down behind the scenes. There's rumors of that. We have nothing to confirm that. But Anthropic also haven't released A statement or explained what's happening here.

34:36

They're not answering to anyone acknowledged it. Yeah, usually when there's a problem, they were like, oh, we have a problem, we're going to look at this. And then they reset the limits. That's kind of like the way it works. Codex has done that so many times. It's kind of a meme at this point how many times they reset the limits. But now they're saying nothing. And the best time to hide reduction in limits is like during these two times promosing because it's kind of blurry. Right. Your usage changes all the time. Like usage limits changes all the time. It doesn't feel like you don't really understand what the baseline is because it's always different.

36:05

So that when the promo does end, you're like, oh, it was because the promo ended rather than because they dialed it down behind the scenes.

36:36

Okay, so my guess, I mean that starts to confirm what we said in last week's episode. So I kind of feel good about that. Don't feel so good looking at my usage limits because yesterday I was not even using it that much and it just hit the limits really, really quickly. So yeah, as we said, you're probably going to have to end up paying more for what you used to have. And it starts step one, basically. And then in one week, this 2x usage limit ends. Now what's interesting with the 2x usage limits from Entropic is that they do different times which they have essentially peak time and not peak time, which works a little bit like electricity. And if you think of their data centers, it's a little bit similar. They have the time where everyone's coding at the same time. They need all the GPUs that they have. And then during the off peak time they have a lot of their GPUs that are not even used. Right. And so I'm wondering if they are not thinking of introducing this peak time, off peak time usage limits and if these things could not become something that stays, maybe not as much usage, etc. But it would make sense logistically.

36:44

It's a good analysis. It's a good analogy looking at the power grid because all of the problem with power generation is peak usage. Right. And how do you address that? If you can spread it out among other to other times, it's so much easier to manage. And maybe that's why they're pushing a lot of these tools like scheduled tasks like dispatch, you know, so you're able to actually do stuff which you don't actually need to do right now in this second, but just because you're at your computer, you can do it. You can set that to run overnight during off peak hours when it's cheaper and you're not going to hit your limits as fast. So yeah, I'm starting to see like a strategy evolving here amongst everything that they're throwing out there.

37:52

Yeah. And if you think also there's a Sam Altman interview recently where he actually used the same analogy of like, oh, we think that people are going to be buying metered intelligence from us the same way you buy emitted electricity. I think because they have to think of the logistics of the delivery, which we don't think about as a user. But for them it's like servers, energy usage, GPUs, et cetera. They think a lot in these terms. And how do we maximize the usage? So my guess is maybe at the end of this promo we see the introduction of peak time and you get maybe the usage limits that you used to have at the off peak time and then you get reduced usage limits during the peak times and then that becomes semi acceptable to people.

38:34

Yeah, because that's like if they instead of taking away, we know this is getting taken away, but if they then in April introduce off peak time and give you more back, that's seen as a positive thing. That's how I would do it maybe.

39:17

Exactly. It's like knowing how companies have to manage resources. It's not a given, it's just a hunch. But so far the hunch has been correct. Right.

39:31

I think it's all worth putting into context as well. It's still massively subsidized. They're losing money on all the processing that they're are selling us through. Our.

39:41

You have to get closer to the baseline eventually. Right. It has to happen. And it's a mix of gains in efficiency that are invisible to you. So they manage to run the model for cheaper, but they don't increase your limits. And on the other side just reducing your limits.

39:50

Yeah, sorry to go off track a little bit. We hadn't sort of prepared for this, but I'm hearing a lot of news about them trying to build, not necessarily anthropic, but Elon Musk trying to build like AI data centers in space. Have you been following that? Is that an energy thing as well or what's going on there?

40:05

There's two things that are bottleneck with these things, right. It's like energy and when you're in space you can capture like way More sun energy. There's like not the atmosphere. And second, cooling is a big problem for these servers. And guess what it's called outer space and it's like infinite cooling. And so it wouldn't be hard to literally just cool things in space. So for these two reasons, it kind of makes sense. Google is also working on it. They said that they are already doing tests this year. They will launch some stuff in space

40:23

this year, just as you said that. I was like, yeah, of course it's really cold in space, but you also have 24 hour sunlight, right?

40:53

Yeah, exactly. So they can have the satellite. Yeah, exactly. So it kind of makes sense. They can harvest more energy and the cooling is also sorted. The obvious problem is something gets down. How do you reboot? Where's the reboot button?

41:00

Send an engineer up there.

41:15

Yeah, I mean that's exactly the idea. But they imagine that the demand is going to be so big that it's going to be worth the costs of having it so remote, basically. So it's kind of a moonshot at this point. It sounds a little bit crazy, but if we 1000x our demand. And also the US has a big problem, which is the power grid cannot sustain the upcoming demand. This power grid sucks in the US and so they have that problem if the administration wants them to run this in the US for security reasons, for patriotic reasons or whatever. And at the same time they don't have the energy. So it's like, well, space sounds not too bad. Diesel by an island, I guess. But yeah.

41:17

Okay, let's move on then. Let's talk about OpenAI's alleged or proposed super app that we're hearing about. So last week we talked about that leaked memo where they basically said, hey, we need to focus on enterprise and on coding because that's where the money is and we need to make money. We're hearing now that they're Planning on merging ChatGPT Desktop App Codex and even the Atlas browser all together into one like ChatGPT OpenAI super app. So what have you heard about this and is it a good idea?

42:01

I mean, we haven't heard much, but there's a lot of devs on X from OpenAI that are hinting at this, like, oh, it's all coming together and so on. I think it's a good idea. Like the ChatGPT desktop app is basically abandoned. Well, at this point it's missing most of the new features that have been coming out. There hasn't been many coming out to ChatGPT, but even it's missing that stuff. It's bad. The Atlas browser is like two guys running it on X as well. There's literally, there's no money behind it and it's so slow at updating that it's not great. The only thing that has momentum right now is Codex and Codecs is great, but again, it's very code oriented and they're missing this cowork type thing like this. How do you use something like Codecs for everything? And so I think they're kind of like the idea would be like the chat of ChatGPT could become this cowork thing probably. Or maybe this app becomes, as they focus on professionals, cowork plus codex and then the ChatGPT app stays for consumers, basically.

42:35

I disagree. I think centralizing it all in one app, it gives them access to their distribution that they have through the ChatGPT name and stuff. But I would say that 95% of ChatGPT users, they've never even heard of Codex or Atlas. They're free users, but this will give them access to those in the same way. Like Claude introduced Cowork in the Claude app. So everyone that uses Claude is now aware of Cowork and now of Claude code, because that's in there as well. So it's kind of like, you know, bringing people up that value chain that latter those who are willing to take it. But ChatGPT just has such wide distribution that that's going to be, you would think, a big advantage for them in getting their paid products to wider audience.

43:33

Yeah, I think it's again, more formal from Claude. Right. Because Claude has been consolidating everything in one app and the Claude app was like laughingly slow for a long time. But they've actually optimized it a lot recently and it's acceptable. I wouldn't say it's great, but it's acceptable again because they're vibe coding it all. So now they can have loops of Claude just optimizing the app on its own and it's decent. And I think, yeah, they see that the app usage on Claude is high. The app usage like you should not use the desktop app on Mac for ChatGPT is completely outdated at this point and they're missing out.

44:13

Yeah, it's interesting, this notion of a super app because I think that originated in China with. Is it WeChat? WeChat, yeah. And there was another one as well which basically does like, it's like messaging, blogging, food delivery, your email.

44:44

It's like grab entirely storage.

45:01

Like it's like it does everything you need. You could actually only use that app and like do most of the things you would need to do throughout your day. That doesn't exist so much in Europe in the US because you know, Uber is separate from WhatsApp, is separate from Google Docs, Google Drive, you know, like it's.

45:03

I don't think that's where they're going though. I don't think they're starting a social network or something like that. I just can't imagine it. So yeah, I don't think it's going all the way to a super app, but it's more like a super app that has every service that OpenAI offers.

45:22

Yeah, but you know, at the same time you hear the kind of longer term vision of these AI companies just eating, you know, smaller businesses, especially SaaS tools, things like that. If OpenAI is bringing Atlas, which is a browser into the ChatGPT app, suddenly like the possibilities for what you can do in that one app become pretty interesting. So yeah, I mean we were a while away from seeing this come into

45:37

play, but we'll definitely slowly OpenAI ships is like it's going to take a while because like, you know, Claude delivers an update every day. And when was the last time you saw something new on ChatGPT? It's been a while.

46:05

So they haven't hit that flow state with their releases yet.

46:16

They must be in full panic state right now. Like when they see the speed at which Entropic is shipping and then how slow they're getting. Like Codex is fine, like the Codex team is fine. But everything else in OpenAI is sucks and the models are good too. But yeah, the product is not there.

46:20

So speaking about anthropic shipping stuff fast couple of other things that they've released this week. So they released Claude Computer use. This lets you use your computer, your keyboard and mouse from Claude. You have to update your app to the latest version to be able to use this. What's your first impression? Scale.

46:33

It sounds very cool. Basically it's a user of your computer. It just can open a window, you can click on things, it can do things. The problem is how much of your life, apart from your VS code setup, how much of your life is still locally on your desktop with something that you could not use in the cloud. To me, not that much to be frank. And so sounds very cool. In practice it's pretty handy if there's a file that you forgot to download on your downloads folder, for example, and you need it right now and you're on your phone, it can do that. And I think what it's good for is all the apps that don't have an mcp so they show Photoshop, for example. That's kind of interesting, maybe for that it's okay.

46:58

But what's actually happening, just to explain, is that it's taking a screenshot of your screen, it's analyzing that screenshot, it's thinking and it's reasoning with itself, and then it's executing the mouse or the keyboard press.

47:41

Yeah.

47:54

So although the keyboard press and the mouse click, it is very fast, it can enter data quickly, takes time taking that screenshot and thinking and reasoning is not fast at all. So, you know, it's not doing any tasks that you can do yourself really much faster, at least at a basic level. I'm sure there you could kind of build some combination things there that would work. But yeah, like the use cases I've seen other people share with this have been like super weak. And I just, yeah, I haven't been impressed with it. And it's like, I'm not really that motivated to use this right now.

47:55

One thing that's also kind of like limiting is that the vision capabilities of Anthropic, they're not amazing. So I've seen it when I do website design, you know, I have this loop where I connect the Chrome MCP and it kind of like opens the website, starts taking screenshots and looking at things, and it misses a lot of small issues that I have to point myself. Hopefully this improves. But obviously for this kind of work, like Photoshop, obviously it's not going to be very good at it. And so that's the problem is that Gemini, for example, is a lot better at vision than Entropic, sucks at everything else, much more, but it's better at vision. And so again, this Photoshop example, in my opinion, the chances of you getting a good image out of the model using Photoshop for you are fairly low, I would say. And you need to think as well that the terminal use allows it to do so much stuff already, like, you know, resizing images and cropping and everything. You know, our thumbnails, they are made by like adding our photos on top of a background generated by nanobanana that's not made with computer use. It's not like taking the mouse and moving the thing. It's literally just using the terminal to just superpose an image without using a mouse. And it's arguably a better user of the terminal than it is at using your mouse. And keyboard. So it's almost like because people don't really understand everything you can do with terminal commands, it kind of bridges that cognitive gap of what can AI do. It does open some more, as I said, on the apps that cannot be used with the terminal. But I don't think it's as big of a leap as people think it is. But it's a cool edge case. Like it's fine for daily release. I just don't think it's a revolution.

48:26

Yeah, it feels like using the model to decide what to do and to take screenshots, it's just a bit kind of a janky process. If there's some deeper integration into your operating system, it could be a bit faster and it's almost like streaming it rather than taking screenshots every time it does something like they would just do it a lot faster. So yeah, we're not there yet. I'm sure it'll come, but we're not there yet. Where we are though with again another Claude feature that they released in the last week is visual and interactive content. So this transforms Claude Chatbot from instead of just giving you text back all the time, it now starts generating these visual and or interactive elements so you can trigger. Here's an example of what's the weather today and it showed you like a moving cloud in the background, you know, not that impressive where everything comes into its own is if you ask it to, for example, show me how inflation works instead of like how does inflation work. But if you use the word show me or something similar, it went and without any other prompting just built this interactive inflation calculator that you can, you know, increase the inflation rate, the number of years starting saving point and like it shows you how to inflation eats into your savings over time and then how much milk or rent is likely to increase.

50:11

And we have a graphic on how tariffs work for some people that don't understand. Can you do that?

51:35

It could absolutely do that. Whether or not you will believe that that's how they work might depend on how you vote. But we won't get into that today. It's interesting its own right. You said you were using it for like shopping and like understanding like difference in products, which is something that a lot of our older listeners will be familiar with in terms of creating, you know, like roundup reviews, reviews of products on websites, something we used to do a lot back in our SEO days. But this is really starting to eat into what you would see on some of these, these resources. So for me this kind of Feels like at least one part of the browsing experience from a shopping or an information gathering perspective is being replicated inside the chatbot here. So again, another reason why we don't need to use our browser to find certain types of information.

51:43

And that's the thing. It's like people are like, oh, but if you put all this effort in making these custom graphs and everything that AI can do because it's just text, et cetera, and well, here we are,

52:35

it can now do all of it. You can, by the way, you can share this, but it'll just share an image of it. But in Claude, you can save it as an artifact and then share that. You can publish that artifact and then share that and other people can use that as well. So again, pretty cool, I would say.

52:46

Yeah, I like this. I actually use the chat app quite a lot on my phone. I like to use the chat app and I almost like using the chat back now. Like, I have this crazy VS code setup, et cetera. But sometimes you kind of want to go back. When you think in principles and high level ideas, it's kind of nice to go back to a context, free discussion and talk about ideas and based on what comes up there. Then you go back to your big VS code setup with all your stuff and then you just actually execute on that.

53:04

So it's like saying one of the benefits of chatbots is that it doesn't have context. Yeah, okay.

53:31

It's not biased.

53:38

It kind of comes full circle in a way.

53:39

Yeah, you do. It's like sometimes it's nice to kind of like walk in an unbiased environment and you do want it to be biased when you walk on tasks and have all the context, et cetera. But it's almost like fresh perspectives. Right. It's like instead of being completely ingrained in what you do and knowing exactly how you do things, it actually takes a step back and gives you different perspectives and that broadens your horizon. So it's like I actually do use the chat quite a bit for that and quite often. And then work is done on my VS code setup still and now starting to be on notion as well.

53:40

Okay, so that's been everything we're going to go through this week. Any final words of wisdom, Gale?

54:11

No, just again, things are moving. I think the website stuff is probably the one that is the most important for people on this episode. It really is not that difficult. You don't need to know code to make websites, like making custom sass with complex API keys, et cetera. Sure, Maybe you shouldn't start there. But simple HTML websites, now, I'm 99% sure they look better than your sites, I'm 99% sure they're faster than your site, I'm 99% sure the code is cleaner than your site, and I'm 99% sure it's easier to administrate than your site.

54:17

Just to interject there, it's not just for redesigning your website. Right. If you are in the marketing department and you want to run a campaign for a new product that you're launching or something like that, you can just launch a landing page for that. It kind of frees you in a way of having to go through procedure and red tape to get a new page deployed. And that might be quite restrictive for certain small businesses before.

54:53

Yeah, you can just put it on a subdomain, right. If you want to do a PPC campaign or something like that, it's super easy. And you don't need to redesign your website. Actually, the scale I did can lift off your branding. So it's like it can go on your existing website, lift your branding, your fonts, your color codes, et cetera, and build landing pages that use that branding. So it's like it's not going to be one to one perfect, but it's going to feel branded and then you can steer it from there.

55:17

And as a treat for those who've listened to the end of the podcast, I think we can reveal that we're actually about to redeploy or release the new version of our website on authorityhacker.com as built exactly like that.

55:43

Yeah, yeah. So it's like we do what we say.

55:56

By the time this goes out, is it going to be up or later maybe.

55:59

Maybe it's like just. I don't want to be saying for sure because I don't know when this goes up and if there's any issues, but hopefully depends on your review. I'm waiting for you to review it either way.

56:04

Something I think we could probably talk a little bit about next week on the podcast. So thanks for listening to this episode of the Authority Hacker Podcast. Do us a favor and leave us a review on your favorite podcast app. So if you're using Apple Podcasts, Spotify, Amazon Music Podcast Index, iHeartRadio, Podcast Addict, Podchaser, Pocket Cast Listen Notes or Player FM or any of the other ones out there, all of them have this section where you can just go and review the podcast. Leave us five stars if you think we deserve it. That is what drives reach and engagement on these platforms. It's really hard for us to grow on these platforms otherwise, so we would love it if you could do that. If you enjoyed the show, really appreciate it and hope to see you next week for another episode. Thanks.

56:15