Today, Anthropic made an announcement that they are opening it up so that you can use Claude code with your Clawude Mac subscription. We're going to get into all that here in a second, but the first I'm just going to say, well played. I think this just makes too much business sense for them and it's something that honestly probably should have happened in other companies sooner.
I'm talking about Cursor, Windsurf, Augment Code. All of them should have a $200 a month plan that gives you near unlimited access. Let's jump into this news in particular.
Catwomb on Twitter posted this and said, "Cloud Code is now included in the CloudMax subscription. " Now, I missed this when it first came through. I want to thank my Discord community for actually calling this to my attention because I'm a big fan of Cloud Code.
I actually really enjoyed using it, but it just burned so many tokens it was kind of expensive to run on a daily basis. And this is coming from someone that spends a fortune on AI and I've been trying to minimize that as much as possible. What is the pricing for this?
So, there is the $100 a month plan and the $200 a month plan. I because of this have upgraded to the $200 a month plan to see if I can use that to offset a lot of my cost. But I'm worried about that.
I'll talk more about that here in a second. And if you aren't familiar what Cloud Code is, it is a coding assistant that is in your CLI. Now, it only works on Linux and Mac, I believe.
I think it believe it works on Mac. I haven't tried it there. But on Linux, which is where I'm running it now, I guess technically I'm running it in WSL using Iuntu.
Um, this is what it looks like. I have auto accept on, so it's just going to keep going here. There are a bunch of different commands that you can fire off in here.
One of the main things you have to do, which I actually kind of like, is you have to kind of control your own context window. Uh, you can just hit clear. It clears the entire context, and then you can go again.
I also like that they have the compact where you can basically squash your uh conversation into a smaller context and then continue on. There's other things that are in here. So, you can connect MCPs.
There's uh ways to do like to-do list, which I don't know if you noticed that, but it kind of built its own to-do list and work through it. So, it's sort of got the I want to say orchestrator mode, but it's not really that. This is what cloud code is.
And you're getting nearly unlimited usage of this. I'll talk about the limits on that here in a second. But this is the post that I put out on X uh on April 9th.
I basically said, "Awesome. You launched the Max plan. Now let us use it via the API.
" I had a sneaking suspicion at that time that it was going to be a cloud code thing, but I really was hoping that they would open it up so that we had an API key. We could authenticate it with our Cloud Max account. and we just get a really healthy amount of API usage.
That way I could use it wherever I wanted to. Now, I think they did what I was just makes the most business sense for them is they're trying to lock you into their ecosystem. And there are pros and cons to that.
I mean, a lot of pros for them. Um, I guess the cons all kind of fall on us as users. And if you don't know, I did test this out in the past.
This is the video I did on it. Uh, and the reason I quit using it was simply because of the expense on it. I was burning 20 million tokens up a day.
It was just a ridiculous amount. And I bet you they've optimized a bunch now. Uh, and I think their caching and stuff is probably better.
But anyway, the the premise of this is I really liked using Cloud Code. It felt natural. It felt good, but it was just too expensive.
And this $200 a month plan kind of forced me to have to go sign up for the Claude called Max today and give it at least a one month try to see how it decreases my cost. Now this is about my breakdown. I am terrible uh because I keep adding more cost rather than cutting them.
I keep doing things. So don't be like me um because this is just getting outrageous in my mind. Uh, so for example, anytime that I think about cutting Gemini, they put out something like, hey, their image edit or their video creation, I don't really need that stuff, but man, I just like playing around with it.
So, they've kind of got me all hooked. And OpenAI is the same way. Uh, Copilot's one I bought for a year.
I probably cancel that. It's kind of become useless to me. Augment Code right now is one of the best value agent coding systems out there.
And honestly, I think they need to hurry up and launch a $200 a month plan or heck, undercut it, make it 190 and make it basically unlimited access and they are going to just be rocking it. And then I play paper. But API cost when there are free models.
When Google Pro experimental was out, I was using that like crazy and my costs were near zero. It was freaking amazing. But when those go away and I start hammering uh costly APIs again, you know, I get to around $100 a week.
Now, this isn't a typical $40 or 40hour a day job. Like I'm doing a startup, so I'm working a lot more hours. I'm working weekends and nights.
I'm doing all kinds of stuff. So, for example, on requesty, uh last seven was $46. I looked at that about $35 of that was for me uh in particular using Clawude.
Uh and then on Open Router, you can kind of see that you can see most of the day-by-day breakdown, but in general, it ended up being close to $60 uh for the overall last 5 days of me using that. And then the nice thing is is the month of April for Gemini, I only spent $12 because most of that I was very good at using the free versions, the experimental versions whenever possible. but they're so rate limited now I just can't I can't do that.
Uh in fact if you look on April 27th on open router I had a day where it was like $22 a day. So if you you know if you extrapolate the $22 and $14 let's just say I'm averaging $12 a day um 6 days a week or something on API cost and that's not counting the other things I'm using like Gemini Direct and Requesty. You know the price gets very close to about $400 a month which is just insane.
So yeah, I did end up upgrading. My hope is that I can offset a lot of my coding cost with cloud code. And then the sad thing to me though would mean that I have to kind of hold on using R code much right now uh for the sake of cost savings.
That probably won't I probably just won't hold with me to be honest because I have that configured just so much how I like it. And even using cloud code for the last hour, I'm going to show you some examples how that was kind of painful. So I just want to talk about this a little bit.
It's a brilliant move, but really we get locked in heavily to Anthropic and to Claude. And then I worry about this a lot because I feel like I don't know if you guys feel this way, but we keep getting bait and switched. Cursor, huge fan.
I bought paid for a year of cursor, then they nerfed it. And I know I know there are a lot of you that Cursor fans and you're going to tell me I'm too hard or I'm whatever, but they nerfed it. If you're really power using cursor as a professional developer, you can tell a difference between a 200k context window and like a 60 or whatever they have it at.
And and I don't know, that is my opinion. We're entitled to have our own opinion. I feel bait and switched by cursor.
It's one of the reasons why I have a pretty negative uh perception of it, even though I was super positive of of it at first. I also feel the same way about GitHub Copilot. GitHub Copilot, I did a video on GitHub Copilot about how it actually became relevant.
It's amazing. But you know what? They've nerfed it.
So, they caught me uh because it's this amazing thing. When it was on the insiders edition, agent was awesome. Now, it's so slow.
It gets very limited. I I feel like I'm moving at a snail's pace. I feel like I'm actually slower than fingertyping on the keyboard.
You know, you know, there's people that actually peck on the keyboard. That's what I feel like when I'm using GitHub Copa. I know that's a little unfair, but in AI, in the world of AI, that's how I feel.
Um, when it edits files, holy crap, it feels like it's like one line, one line, one line. I I don't know how those of you that use Copilot um bear with that to be honest with you. And the final thing I'll say is just we lose so much customization.
There is some that you can do with claw code and I was going through some of the updates that they've done on it so that you can do some rules and you could do some guidelines and stuff for it. Um but you just lose so much compared to what I've just come to love in Root Code. But honestly, I really need to figure out a way to keep my API cost under control cuz it is it's just crazy for me to be spending as much money as I'm spending right now on AI.
Now, this is something that drives me nuts about cloud code. And, you know, you go back months ago when I I don't even know how long ago it was. Let me see.
It was we go back to February 25th when I did my first Cloud Code video. Uh, looks like maybe the February 24th. And you know it's come a long way.
AI coding in general has come an incredibly long way. So when you think about that um my perspective then was very different because agentic coding was it feels like a lifetime ago. But now I've got my setup so configured like look at this question that I pose.
I say all right when an image is generating if we refresh the page and lose its generation is there anything you can think of to make it so that the image keeps generating and we have it when we come back. You can think about this like I'm implementing uh OpenAI's image generation. I've got a really cool workflow around that.
I kick off generation but those things are so freaking slow. If you refresh the page I don't want to lose it. So in my head, I know exactly how I would code that up if I was doing it, right?
I would basically build it. I'd have a job in the back end that kind of runs, waits for that to get done, it persists to the database, and then I'd be able to actually query that database on the front end to be able to pull the image. And I already have a lot of the data model set up.
That's the thing. I have the data model in there that connects an image and where it's actually going to be presented on the UI. I literally just need to have kind of a placeholder job attached to it.
So instead of talking through it with me, it literally jumps into coding immediately. And this is frustrating because I I have come to appreciate more and more when I want some when I working with an ajunctctor tool, I don't want it to always just code immediately. And that is one of the reasons why in particular um I started customizing rue code so much that and cost.
But then look at this. I ask it because it did a horrible like absolutely horrible solution to this. It put it in local storage after it generates.
If I refresh the page, it is not even going to see it generate. So then I ask it, but will they even get to the local storage if I refresh? And then it'll kind of tell me, no, it won't.
Like yeah, uh it'll work. you still have to sit on the page, but if you refresh it uh before you save it, that'll do it. But I already handled that.
Like that's not even new functionality. This is all handled if you stay on the page. So very frustrating.
Now take that into consideration with this pure programmer mode that I made in Root Code. The same prompt, what it ends up doing is it ends up start giving me options where I can actually go through and look at it. It did actually come up with the same local storage option.
To be fair, I am currently pointed at Claude in this one, so I would assume it's pretty similar. But then we start getting into option two, which is a better solution here. Uh, it's the one that we should 100% be going down and we should not do option three.
We don't need websockets or anything like that. So, we go through option two and now I can actually get a list of questions. You can see that I can say, "All right, cool.
We're going to implement that and go forward. " So going back to this, this is going to be tough for me because I've built this like really healthy kind of ideiation mode that I really appreciate. I tell it it it forces it to always give me more than one option because a lot of times the normal like coding mode in augment code and even uh root code will just if you if you give a general thing it'll typically do 50% okay 50% terrible solution.
This is happen to be a 50% terrible solution. So going to the next p going to the next uh slide here. I do want to say I have hope that cloud code is going to upgrade a ton.
I'm doubtful they're going to add the level of com uh configuration that things like Root Code currently have, but you never know. They've been pumping out things left and right. And this is just over the last like from April 24th to the 30th.
I was looking at they're just pumping out updates. You should follow Catwoo if you want to keep up with that um on X. And then I'm also very interested in this ability where you can actually run cloud code in headless mode because I thought about this for hooking this into my GitHub repository and then having it potentially be able to go in and fix things for me automatically and then and commit them uh and generate a PR or something for that could be reviewed.
So I am very interested in this. I'm not sure how this is going to work with my Mac subscription. most likely if I'm in headless mode, it's probably going to need to be um usage based just based on how I'm thinking about how that would be set up.
So yeah, uh my final thoughts, this one's tough for me because I really do want to get my cost under control for AI. Uh, and you know, maybe I'm one of the only ones like harping on this, but you know, developers are getting like really it's getting really expensive to be a developer because our output needs to go up. A lot of us need to pay our own AI cost.
Uh, so I I do think companies are going to have to start having like AI budgets for engineer and I know some of them probably are doing that already. Um, but it is tough on me because it makes a lot of sense for them to do it. But I lose so much of what I really love about working with some of the other tools.
So, I'm going to give it a run for a month. And you know, it might be one of those things too where like I I went with Windsor for a while and love their IDE, love the way their tools worked, but it was just like one of those things where I felt like they were preying on the way they were doing pricing. Uh, which they've changed now, which I actually have thought about giving Windsour another try to be honest with you.
The final thought I have is I really just hate being locked into an ecosystem and if a new model comes out that I want to test, you know, I'm going to do it. You know, I'm going to pay for the API cost in it. You know, if I can't run it locally, um the claw code is very locked down to just clawed.
You know, maybe that'll change over time, but it's very doubtful. I feel like they're finally tuning it to their ecosystem, which just makes a lot of business sense. I mean, you want people to be locked in.
It's like uh I don't know if you guys remember in the crypto days where everybody was like, "Oh, things in games should be inapp purchases, you know, it should be uh tokens or whatever that you can then trade with people outside and then go to other games. " Let me tell you something. If you own a business and you own a game, you do not want things in your ecosystem to go to another ecosystem.
it is not in Claw's best interest to allow us to go to other ecosystems. Uh they kind of do it now because they have that API access so it works out. They make a ton of money from that but ultimately they want that sweet sweet recurring SAS revenue.
They're guaranteed now to have that sweet monthly subscription plan that they probably can tune the usage on quite a bit. And the final thing that I'll just touch on real quick that I think I forgot to talk about this earlier is that ultimately you are rate limited with each of the plans. You can send about 225 messages every 5 hours for the $100 a month or about 900 messages with Claude every 5 hours which means about 40 to 80 coding tasks every uh every five hours.
This seems incredible to me and I have I've been using it for probably three hours now right before I started making this video. Didn't have any major issues with it other than just fighting it wanting to code things that I didn't want it to actually code. Uh so it's just not good at ideulating.
Anyway, I'm going to wrap it up there. I appreciate all of you. Let me know your thoughts on Cloud Max and how it integrates with Cloud Code.
And if you are going to sign up for CloudMax, I'd be curious how many people are actually going to do that because of, you know, I have to imagine a lot of you also have pretty high a uh AI cost on a monthly basis due to like just the expense of the API calls. All right, take care everyone. If you wouldn't mind hitting the like button or consider subscribing if I've earned it.
Till next time everyone. Peace out.
