Web News: Trying Claude Code for the First Time

All right, everybody, this is the web news and we are going to be doing something different.

I have no context for this, which is why I'm kind of hesitant in my intro because I don't use

cloud code at all. I do use AI to help me code here there, but I do not use cloud code. I don't

use any of this agentic workflow stuff at all. And so Mike said for this web news, just sort of

live. And I again have no context, other than maybe I get a YouTube video or two, he's going

to show me some stuff. So Mike, why don't you why don't you take it away? Yeah, absolutely. So

I think it'd be cool for us to just kind of experience this live together. I don't have anything

like super, super planned. I want us to stumble. I want us to get through this. I want you to ask

questions as you go. But since you've never touched cloud code, I think there's going to be a lot that

will present itself as maybe confusing or maybe interesting. So I figure this will be a good

experience. And I'm kind of glad you haven't touched it. I kind of glad that I get to be the first

one to show you what you can do with it. We're not going to do anything too crazy because we obviously

want to keep this web news long, like so 15, 20 minutes. But we are going to at least try some

things. And I'll show you kind of the agentic side of things. So I do want to show you how you can

like kind of launch a browser, ask it to do stuff in a browser, help you in your like day to day

life, as well as the programming side. Well, to give you any context, I have opened codex and

done one demo project, which is actually what the next episode we're going to record is going to be

about. And then I've watched a cloud code video on YouTube because I knew we were going to do this.

And I just at the very least wanted to understand generally what we're doing. But I have not I don't

have cloud code. I don't have any subscription. I never downloaded it or anything. So just to

give people like if people are like, Hey, you kind of sounded like you do that. It's like I've seen

a, I've seen a YouTube video and I've literally opened codex and for like five minutes. Yeah.

Well, that makes sense. That makes perfect sense. But yeah, hopefully, hopefully we'll get

somewhere or maybe we won't. Maybe this will be an epic failure because cloud is down at the moment

or something. It should be fine. I think I checked. Although we just tried to do a cloud code

thing with web flow and web flows down right now. So fantastic. Yeah. My first demo already failed.

This is attempt number two. Anyway, just jump it in right now. What we're going to do is like

to start with any project. We're going to actually create a directory for it. So that we're

going to call this McDermot one, right? And then CD into demo one and clear the screen. Everything

should be nice and cozy for us to begin. So right now, are we just on like a terminal? Like

we're just on a Mac term on Mac either, by the way. So this isn't this isn't just a Mac terminal.

This is called ghosty. It's just a terminal emulator that has a lot of extra functionality built in,

makes it faster. It's more like compatible with two ease, which are terminal user interfaces.

Oh, yeah. Ish term. But yeah. So it's just a better way to kind of interact with cloud code.

The terminal like the Mac terminal actually is pretty glitchy with cloud code.

Oh, that's thanks. And then again, just to give context, I do use terminal and command line.

Just I'd use that on Windows primarily. Yeah. So with this, all we have to do to start a

cloud code session is type in cloud. And that will give us, you know, it'll tell us what MCP

tools are called. Don't worry about this isn't really not important. That's more for my work.

And here, this is where actually in cloud code now. So this is still terminal. But this is,

again, that too we were. This is a terminal user interface. Yeah. So we don't have,

I can't perform traditional terminal commands in here. Like I can't do like a, you know,

a McDurren here. This is literally like a prompt box, like a text box where you can talk to the

the cloud, the cloud code interface. And this is your chat GPT like chat box kind of kind of

the only difference between this and chat GPT is the orchestration and the agentic side of it

is quite more interactive with your computer. Chat GPT is interactive with the web.

It can't really action on your computer. You can give it context. You can, you can paste in

some screenshots. But with this, it has access to the terminal. Right. So it has access to pretty

much anything that's on your computer. And depending on what kind of level of access you give it,

it has, it can do more things autonomously or less depending on if you allow it to like go free

rain. I haven't given it the the free rain yolo access because I like to least like for the

dangerous commands, like, especially with like in production work that I do, I don't let it like

push anything to a repo. I don't let it remove any, any files on its own. Like it has to ask me

for everything. But you can give it like an unsafe command essentially that will allow it to do

anything, including, including like literally delete your entire disk if it gets that crazy.

It has, you know, commands inside of it not to do that, but it's, it's happened before.

So this is why people use like a Mac mini, right? Like this is why, because although they mostly,

I guess use that for open claw is what I hear, but I mean, a separate Mac mini or just a separate

computer or bluntly said, that's correct. Open claw is essentially just an automated version of

cloud code. Like it has cloud code built in a lot of times to the open claw and it just allows

cloud code exactly what you're saying to be kind of sandboxed in, right? So that even if even if

it's running unsafely, the worst it can do it is wipe itself rather than access your, you know,

personal files and wipe them or whatever. That is exactly it. So have you said that?

Our first example, what we're going to do actually is we're going to take our HTML

of things website, which I love and it's, it works great. But I want to see what

Claude's design capabilities are. Like I'm not really that good at design and Claude is

apparently getting better. Opus is getting better at design. So I'm curious what it can do like

with our current website. So I'm going to tell it to go into our current website and look at all

the content. Well, I'm going to it's going to it's going to it should be able to gain context

of what it is. Like it should be a, you know, a podcast website with blogs. And I want it to

design a better, more catchy cover and I want like I want there to be a significant difference

than what we have right now. Well, to be fair, the thing I built was years ago. We did it

quickly and we haven't facelifted it. So I definitely think it could use a facelift.

Yeah. So maybe this will give us some ideas. I doubt it'll be able to replace the website

because we only have 15 minutes, but maybe this will give us some cool ideas at least, right?

So let's try let's do that. So in this, it would be just like chat GPT. So can you please,

I'm saying please to the bot, take a look at HTML, all the things.com. And I want to redesign the website

and the website to be a bit more modern and have a much more actionable cover portion of the site

and be clear about what we, what we are, which is, which is a web development podcast.

And which web development podcast and I don't know what else. What do you think? What else do I add

here? Like, I don't know how much context you have to give it like because we cover AI,

we talk about some small business stuff. I don't know if we need to do that. Yeah, let's do that.

We're going to a podcast. We cover AI web development web dev careers like dev jobs, careers and

freelancing. Do we want to limit it? Maybe like, should we, what should we say? Like maybe

three, let's do three AI web development and small business or we talk about dev jobs quite a bit.

Dev jobs. Okay, let's do that. And then, okay, so that's that. Then we could talk a little bit

about text actions. I'm a svelte person. Let's create a svelte kit site with tailwind. We don't

have to give it this. I want to be clear. It'll probably just make a either a regular HTML CSS

JavaScript site or a next JS site. If we don't give it any information about the text stack,

it's just my own preference to do this because I have more context into the code that it will write

if it does this. So that's literally just a preference thing. With tailwind, with tailwind,

you make sure to grab all the logos and all the logos and colors that match the logos.

Right. Right. From the site as well as all the content you can grab by scraping. It's

creating it. We can't. So my, my initial intention was actually to have it be connected through

an API key because you can do that. You can give it an API key. You can give it a .e and v file.

And it's able to then use that to actually connect to the web flow API and pull in all the API

information, which it would be able to do a lot better because there's a lot more structure in the

API information. Whereas right now, it's just going to have to go through the site map,

which, I mean, it's structured enough. It should be able to do most of what I'm asking for it.

Yeah. And just to be clear, we just can't generate an API thing because web flow is down currently,

unfortunately. Yeah. Exactly. Okay. And I'm going to put, don't make any mistakes with the

shits and giggles. And we're going to let it go. So now it's going to start thinking. It might

start asking us some questions and giving asking us if we can, for example, it'll probably ask if I

can even access HTML with things.com. That's the first thing it does. And I'm going to say,

yes, do whatever the heck you want with HTML with it. Improvising. It says it's improvising.

What the heck is it doing? Yeah. So this text right here is going to be just gobbledy good. It's

literally just kind of a joke. It's not actually, it doesn't actually mean anything that it's doing.

This text up here is the real like thinking and reasoning path.

This is the mind. It's the Minecraft menu yellow text. Pretty much. Yeah. That is correct. So you

can see what it's doing. You can see where where the tokens are going. You can also see how much

it costs, by the way. I am on an enterprise plan for my job. So thank you, Cypherin, for letting

us burn some tokens here. But overall, it's, it's quite expensive. Enterprise plans as well. The

catch there is that they don't allow the typical cloud code subscriptions, which are $200 a month.

Already a lot. They just give you tokens. Like they'll, it'll be $200 of the tokens, which is

way less than the $200 you get with a subscription from the non enterprise plan. So it's just 22 cents

there. You've used 22 cents of computing. Yep. Is that on top of your plan? No. So it's like the

plan itself, I believe, is like $200 per developer. And then so I'm using those that money from

straight from the plan. And then as soon as I go above that amount, it'll start just be on top.

Now, now there's a question. Does it stop and tell you? No. It does give the administrator

a warning. It does give the, you know, like the administrator can set limits. Okay. Yeah.

The administrator can set limits. Yeah. Like I, yeah. So it's, it's controlled by the admin of,

of the enterprise. Yeah. So how long do we think this is going to, because so it's been going

for a minute and 33 seconds, you know, again, counting, it could actually take quite a long time

depending on what you're doing. I'm hoping it will do it in phases. We'll see. And I actually

I already screwed this up, to be honest, and not really screwed up on, not really screwed up on,

not on purpose, but to save time, I skipped a planning phase, which is bad. And that might,

that might result in a worse result. But the planning phase actually takes quite a bit of time,

and then the implementation would take time. Well, what's, like, like, lay that out for me,

what is the planning phase? Like, is it, I understand you're planning, but are you planning

and the bot is doing anything? Is it just gaining context? Is it taking tokens? Like, is it using

tokens to plan? Absolutely. Yeah, it's taking tokens. Yeah. Planning phase is more or less the most

intense phase, because of the fact that it has to go in and research everything. Like, it does a

lot more research. There's a lot more thinking. It'll, it'll almost like think against itself. So

it'll get, it'll make an idea, and then there'll be a reasoning, a reasoning bot that looks at the

ideas, like, no, that's a bad idea. And then it'll, like, it'll, it'll have a lot of back and forth

internally as it, as it's making ideas. And I can show you that while this is going, by the way,

we can launch another one of these. We can go cd.dot and create another maker demo too.

We can launch another cloud instance here so we can have as many cloud instances open as we want.

And here, let's, let's actually do a plan while the, while the other thing is going.

Here, what I want to do is actually a little bit different. I don't want it to make a website or

anything that what I wanted to do is help me with my taxes, right? I did this already. This is

something that I've done. I think I did this in cursor, but I'm, I'm assuming it'll work in cloud

code as well. I need some help with my bookkeeping. I want you to be able to fire up a browser or

me to log into my open AI account. And then for you to take over and download all of my invoices,

so I can match them to their transactions and book. Now we're on plan mode now. So right now,

what it's doing is just considering this. It's not executing anything. It's just going to come

back with a plan like, Hey, this is how I think I'm going to do this. Are you okay with that?

And then you have an option to accept the plan or you have an option to actually like ask it

more follow up questions, do some different things. It could also itself ask you questions,

right? So it's not doing that right now. I was just asking for permission right now, but it can

itself ask you some questions in turn and be like, Hey, how do you want to access it? How do you

want to do this? And then based on those questions, it'll generate a better plan. Let me ask you

a question here then. So like, this is not a question that is out of the realm of what I would

potentially type into chat GPT is what is the reason, I suppose, for you to do it here in

cloud code and not in like a chat GPT. So how would chat GPT do this? You would have to do the

browser extension thing like you, I think there's a browser that chat GPT can control. It's very

similar in this case to that. I would have to I would have to go into my open AI account and

actually like grab context and numbers and things. Yeah, I see that you're prompt, you're literally

telling it to fire up the browser and to do stuff. So this is this is mostly actionable. I suppose

exactly. And we're going to go back to this. As you can see in the terminal here, you can see a

little star. That means this is stopped. So if we go back here, you can see that it's already

asking us a question here. So we can ask it to go on. There's a little thing there that says 23

percent. Is that your tokens? So that's my context. It's a little bit it's different than tokens.

So that context limit is how much it can keep in memory before the model degrades in performance

quite a bit. Oh, and then you have to open up a new chat. No. So it actually auto compacts. So

by the time it hits about 80 percent of that context, it'll ask you, Hey, do you want to

compact this chat, which essentially means it'll summarize the most important parts of the chat

into like a markdown file and leave that mark down file in a temporary space. And then it will

continue on and reference that markdown file as like a memory. Okay. Yeah. Because I so in the

YouTube video, I watched somebody was doing this and they were using cloud code plugin for VS code.

And they had switched. Like I said, we're still in the same project. I might be describing this

correctly. They were still in the same project, but they were going and opening like new chats.

And one of the reasons when they did that at one point, they mentioned that it was because it

would degrade like you're saying. So is this better? Like does the VS code not do that? Or is

that a new thing that cloud code will compact and summarize and then start a new chat?

So if someone's using the cloud code in VS code extension is pretty much exactly this.

When it first came out, it was a little bit behind the CLI and I never really adapted it as much

as I should have probably, but the reality is is that right now it's probably cut up like 95%

of the way. So there's just I think there's a couple of things that it still can't do the same way

that the CLI can do it. But it's essentially the same thing as you're seeing here. Just in your

interface rather than the CLI. Okay. So we can see here that it's finished something. So we can we

can start to like we can hear. I'll I'll actually open up my browser in another window

and see if I can get anywhere with this. Oops. That is. Oh, I see. So you have to actually write run

npm dev. Okay. So it's telling us what to do. You can either do it or you can actually just allow

it like to do it itself, which I will do right now. Run it yourself and open it in a browser or

you don't actually even need to follow the commands that it gives you. The web dev podcast you've

been missing. Man. So just for the listener out there, it's made a website. We're looking at it.

It's got like a big hero section, multiple buttons, some badges and they're 458 episodes

and counting it says. And then it has a course on content boxes and other things so we can see.

And it'll mention stuff as interviews or whatever. It definitely messed up the interviews.

Yes. You can see that it has the web news, which I think that's accurate when clients

ignore it. Yep. That's a act. But then if for some reason, it there's them all all the other

ones interviews, which is not unless you and I are interviewing each other each week. Yeah. I mean,

it could it could assume that. I mean, what do you think? Like what other than the small nitpick

there? What do you think about like the minimal amount of prompting that we did to what we got?

At least the click. I mean, it's it's pretty impressive. I mean, I will say, you know, it's got a

Spotify button, Apple podcast button. It's got some other badges there to say. These buttons should

work. Yeah. The buttons work. Yeah. You clicked on Apple podcasts. If you click on browse

episode, it has an anchor that takes you down the page. Like it's interesting. It's something that

you know, would take one of us a day. Maybe something, I don't know, to like lay out the HTML and CSS.

Obviously, you mentioned there was no planning phase with code with cloud itself. There was

obviously no planning phases in a lot of a website is, you know, especially within a web ad agency

is going back and forth with the client on designs and what they need and all those other things

depending on their project. Of course, do those other buttons work? Can you see the, I see there's

a nav bar that articles there. It looks like everything just kind of anchors you down and it looks

like it did not build out each individual page, but we can do that. I will ask it to build out each

individual. Wait, so what do you mean by that? So if you click, oh, I see. So this has become like a hub

and then it links out to I see. Yeah. So the first step of this was just building out the

cover page, which I mean, it is a reasonable step. And I'm kind of glad it did it that way,

because again, it would have, it would have taken a forever if it did build out each individual

page. What I like about this is it did grab our logo. It didn't quite nail the colors. Like,

I would say the colors are colors at all. They're incorrect. Yeah, they're incorrect. And I did

ask it to use the colors based on the logo. So it wasn't able to do that. Something we can

reprompt so that that's something. What's that? Listen, now button do in the top bar because you

have like other buttons for spotifying that. What is listen now? I think it's the same as the podcast

button. Oh, latest episode looks like the same button. Oh, hang on. Can you scroll down?

So there's latest episodes. Is there like latest articles as well? Oh, no, there is. Yeah,

there is. Well, I was wondering if if that listen now was actually like a like the latest.

Like, I see some websites will have like that page that says like the latest and I'll have the

latest like five episodes, the latest five blog posts, the latest five newsletter. So okay,

well, like I mean, I don't know if it's going to take a while. Why don't we try a correction

live? Why don't we try to get it to fix the colors? That's something that's like very visual,

very obvious. That's that's incorrect and shouldn't take you know, a long prompt. I don't think

right? It's like, what would you write here? Are you writing something? Sorry. Yeah, I'm writing

right now. So I'm writing. Can you fix the colors to be more represented of the colors in

our logo? And I think that should be enough personally. That's what I would say to a designer.

I would I have a question here. So you you can kind of have like that. I always hear

the the official name of it. Let's kind of like the design the brand guidelines, the one sheet,

if you will, where some brands will have that where it has, you know, these are our primary,

secondary, tertiary color, our font sizes, our font families, you know, the list goes on and on.

I've done that for people in the past on like a web page. So I assume could you give it that

file? We don't have that. But if we if we did, that would give it more context. So I recently

created a presentation using this using cloud code. And I literally fed it the cipher and design

guide. And it created a like a perfect designed presentation based on the cipher design guide.

So it looks like it finished. We can probably go and check that out. It's done. And I would say,

where is it? It's definitely better. And where is this living? Is the question? Like, is this

something that's just in a file? Yeah, local host. Okay. It's in local host. It's living. It's living in

that folder that I created for it. That demo one. Yeah. That demo one folder. It's a svelte kit

application. So there's a bunch of code that you can open up and curse or whatever and you can

change yourself. So it's it's living. I wasn't sure if that was in the cloud or not. Because you

can do cloud. I remember when I was kind of like poking around Claude or codex. Man, there's two

many services when I was poking around codex. It there was some stuff where it was like, hey,

you can keep working on this on the go or on the web or whatever. If you like allow like some sort

of cloud or web project. Yes. That's correct. So like we can see, I just hopped into my other one

that's still doing the plan for for the automation tasks that I wanted to create. Oh, it's still

a wow. It's still doing well to be fair. It had a question that I had to answer like it. It was

asking for questions about something. So let me let me ask you a question here. What is the point

of plan mode in terms of in terms of okay. So the bot is doing all of our stuff. It's not like

I'm asking Claude, you know, make a plan and then it gives me a PDF and then me the human goes

and does the plan. So what is the point of plan mode? Is it not running? So for example, if we have

we had used plan mode for the website, would it not build the website? It would let you know I'm

going to build it. I need NPM. I need it would like say all that stuff. It's a more generic thing.

So I can we're running a little bit at a time here, but I can do a quick plan mode here probably.

Can you land out a page a page for the blog and podcasts? Right. So so that we aren't just linking it

to the to the current site. So now you you type that out. It yeah, because you said plan

it's plan mode or did you have to do something like hotkey on it? Yes, it's a very good question.

A shift tab activates plan mode. You can see plan mode is on down here.

And so I did shift tab. I forgot to say that. That's my bad. But having said that, if you do just

say can you plan out, it should activate plan mode on. So that isn't it's not a requirement to do

the shift tab. Okay. Just like it. When you ask it to plan, it does plan. So we can see that this

is asking me something. So let's see what it's asking. Yeah, okay. Just one time. So it's asking me

a bunch of questions. The first question was do you want to check for API or just chat GPT plus?

I said chat GPT plus. The second question there was do you want reusability like you want to be

able to create a script that can do this multiple times? I said no. And here it's asking me what language

to do it in. And I'm going to say type script because that's the language that I usually like.

So it's just going to write a script. That's what it's trying to do. It can continue. So that

that was that's part of plan mode plan mode. We'll ask you questions that it comes up with.

If you don't do plan mode, it has a very small chance of actually asking you those questions.

So loosely, I mean, we get like content from clients when we like that would be plan mode.

So those initial meetings of like what design, what color, what's your brand guidelines? Do you have

that? What's your slogan? Are you going to have a blog? Are you going to have any commerce store?

That's kind of what Claude is doing. That is exactly that is exactly it. But you'll you'll get the

plan. So the thing that we haven't gotten to yet that I wish we did is the actual plan. Right?

It's still asking us for permissions here. Again, it's a little bit annoying with permission

asking, but I'd rather have it asked than just do. Well, especially when it's driven through tokens.

Well, I mean, again, the tokens are too, too bad right now.

Well, see that so this is something and maybe this would be a more

not for this web news. But I've been seeing people that use Claude code or other things. And

they're like, yeah, I spent $200 today on tokens. And I was thinking to myself, I think I maybe

even posted this on X is it's like, people were complaining about the $200 a month.

Chatchy BT plan that came out about a year ago, I think. Now people are spending $200 a day and it's

sort of like, okay, but the $200 a month is not. It's just a little interesting. Now I understand

that this is quote unquote, replacing or augmenting a software engineer. And so maybe there's a

certain amount of value you can affiliate with that. But for the like the indie hacker, I mean,

I don't anticipate like I have friends who worry about $5. I don't anticipate them being like,

I have a great app ID. I'm going to open up Claude code. I'm going to get a $200 a month plan on

them. I'm going to start coding with Claude code and then, you know, paying $200 a day while I

build an app for a week. Oh, the app didn't do well. You know, it's so the indie hackers kind of

still left behind a little bit here. I agree with you. Like the indie hacker that's cost conscious

is definitely left behind because these things are only going up in price, right? Like there are some

cheaper alternatives, but they're also worse. And worse right now is actually a pretty significant

thing. Like only recently did the top tier models become so good that we're really relying on them

to do most of the code. That's very recent like last three months thing experience. So if you can

talking like four months ago, like yeah, we have we have cheaper models that maybe four months ago

would have been top tier quote unquote that you can do in a fairly economic way. And you can still

get a lot done with them, but it's not going to be this like the top top level. And that does

matter quite a bit. So currently, yes, price conscious people are priced out. That will change,

in my opinion, because we're seeing the cheaper models catch up fairly quickly. So I think in three

to six months time, the cheapest model is going to be almost at this level that we're seeing right now.

And therefore will be a very viable solution to for everyone, right? Where you could get get away

with like a $10 subscription actually built something quite good, but we're not there quite yet.

There are again, there are some cheaper models. If you're an engineer and you can play around with

the cheaper models, you're you can still get a lot for with them, but it's not it's not going to be

like as hands off, I should say, it's going to be a little bit more like reviewing a little bit more

reviewing is going to be required a little bit more back and forth with the plan. So right now,

we can see the plan is actually complete. There's a multiple sections to this plan. There's a summary

add four new routes podcast podcast slug blog blog slug. So it's giving you a summary of what it

wants to add. It's going to extract reusable cart components, expand data files. I guess it's

going to extract and scrape more content, which makes sense. It's going to update navigation.

It's going to, you know, this is going to generate a bunch of new files. It tells you which files

it wants to generate. It's going to create new components tells you which components is going to

generate. It tells you some of the data that it's going to select, right? The navigation changes

are going to be slightly slightly different. The navigation is going to be slightly different. The

footage is going to be slightly different. And there's going to be new homepage sections. So

there's a bunch of things and then it talks about design as well. Like what's the podcast design

going to be? And this is where you can again, this is the where you can go back and forth with it,

right? Like if you don't like this idea of all, you know, all, all episodes, search, input,

pill style, like you can tell it to do something different. At this point, before it's implemented

anything, saving you some money. Like this, the plan mode is going to cost a lot less than a full

implementation. Right. So you can, you can have a better back and forth without actually spending

as many tokens. And that's a really keeping here. And then the other really, really big advantage

in the main advantage is the plan mode, this plan will be fed into the model and it'll have a much

more structured to-do list of what it needs to do. So look, but I have to go soon in a couple minutes

actually. But if I click on, yes, an auto accept edits or clear context and auto accept edits,

it's going to start ripping through it. And you're going to see it's actually like doing a to-do list

of things that it got from the plan. And it's going to go step by step through the to-do list

and actually be much more like structured in what it changes. So usually you're going to get

much better output out of these models. If they have a plan and they are able to execute that

plan with a to-do list versus a model that's just going to eat it. And I mean, we saw, like,

this isn't bad. I want to be clear. Like the, the site that it generated without a plan is

probably fine. Like, it's not the greatest website in the world. It's not revolutionary,

but it's fine. Like, it's okay. It definitely looks like an AI-made website, but it is modern.

I mean, it is definitely modern. It looks better when we have because ours is dated.

Yeah. Like, sure. But I guarantee you, if we spent like 20, 30 minutes back and forth going

through it during the plan mode and getting exactly what we wanted out of it, it would be 10 times

better. Almost a guarantee. Do you have a rough estimate as how much that would cost?

It wouldn't, it wouldn't be like maybe a dollar more or something like that. It wouldn't

actually be that much because during the plan mode, it's not that expensive. But I mean,

like the whole project, if you did the plan mode, like, roughly, if you did the plan mode,

if you did the development, and then we were going to deploy this, well, you could easily be covered

for us. It would be easily covered under the $200 subscription. It might even be enough to do

like the $20 subscription on Codex. Okay. Yeah. It wouldn't be that much for this specific project.

It would not be that much. There are some projects that could be much more complex that would need more.

But for the most part, I haven't run into any issues using the Codex subscription. I use that

personally for a lot of things. And the Cloud Code $200 subscription is much more generous than

there. Again, their enterprise level stuff. So it is, it is going to be enough. But yeah, you can see

here, this is this is the to do list that I was talking about. This is really key to these models.

Once it has a to do list, it's much better at keeping on track and following instructions,

obviously, because it can just check these off. And you'll see it as it goes. Once it finishes

something, it'll just like check these off, which is really cool to follow. And it can work on

multiple at once, as long as it knows that there's no dependency. So internally, this to do list

looks like a kind of a to do list with dependencies, right? Like it knows that you can't verify and

spell check until all of this is done. But maybe a couple of these you can run in parallel and

is doing that. Yeah, like maybe spinning up the blog pages and the podcast pages because they're

different design. Yep. That's exactly it. Interesting. Yeah. We're kind of out of time on this part as

well. But suffice to say, like, yeah, we're going to get to a plan. It's going to tell us how it

wants to execute. And then it'll open up the browser similar to how you're seeing it in here.

And it'll ask it'll ask me to sign in. And then it'll be able to take control of that browser.

Impressive and concerning. Yeah. Because why am I here? Yeah. Well, you're here to control it for

now. For now. Yeah.

I don't know exactly. But yeah, hopefully, again, let's go back to the site. Hopefully, this was

information. At least you got to see some of the moving parts. You got to understand it. And I

think we're going to have more of these kinds of episodes where we go through stuff like this

together. Because again, a lot, I think a lot of people haven't touched this stuff quite yet.

You know, even if it is in the zeitgeist and it'll be important for people to see it in action

to understand what it can and can't do as we move forward. And hopefully, like, again, it can do a

lot. It can't do everything. It can do a lot. And we need to be aware of that. Because if you're

kind of putting your head in the sand out there, if you're not using it as much, it could bite you

in the ass. If you're not, if someone else is using it and is able to do stuff faster than you.

Right. I'm not talking directly at you, Matt. I'm just talking directly at like the audience

looking for jobs and stuff like that. It's not really something you can ignore anymore.

Yeah. Like, this is definitely before we used to advertise like a page builder to be quick. And

that was like very quick. Like, that's even faster than a page builder. Yeah. Exactly. And you

speak to it in English, right? Like, you speak to it like a regular, you would speak to a designer.

That's the difference here, too, is like with a page builder, you have to learn the interface.

Here is there's no learning really. There's some like adapting to why? Like, there's some

adapting to the LMS and how to talk to them properly, maybe. But even that's kind of going away.

There's less and less needed as they get better. It becomes more and more natural of how you speak

to them. But unfortunately, I've got to go. Maybe we'll post what we get here as like a screenshot

somewhere on our on our socials. So look out for that. And that's up to you because you're you're

I'm just watching. Correct. Yeah. For sure. All all all posted. But yeah, hopefully this was at

least informational for you, Matt. We didn't do anything crazy, but we kind of scratched the surface

of this a little bit. Well, no, it's good. It's good context because I actually understand the

context of our website. So it's a good place to start for sure. Yeah. That's what I was thinking too.

It would have been cooler with the the API key, I think, who would have been a little bit more

demonstrable, but either way. Yeah. I mean, it worked good even without leaving with that hurdle.

But I think that's it. It can. It can. It can. Sorry. All good. But that's it.

I think that's it. That's it for the web news. Thanks for hanging out. If you liked it. If you

liked this episode, slash if you liked this is sort of a new video for us. Please let us know in

the comment section on various places. If you're on the audio version, there's video version should

be on Spotify, but definitely on YouTube. And we are signing off. Goodbye.