3 Things Claude Code SUCKS at so I Use Codex (Astra Fixes Them) transcript

Now, I don't know about you guys, but I am sick of hearing that Astra is AGI. It's definitely not. However, I did end up canceling one of my Cloud Code subscriptions and switched [music] over to a $200 per month CodeEx plan. I did not want to make a video about this until I had some actual use cases that were going to make you some money that nobody else on YouTube was talking about. Now, in this video, I'm going to bring you the three things that Astra and Codeex is actually way superior to Claude Kodak.

By the end of this [music] video, you're going to have something to actually walk away from and start implementing inside of Codeex with Astra [music] that Cloud Code is currently not doing for you. And this is actually going to make you money. I guarantee it. Cuz when you click on one of my videos, you're not just going to learn about benchmarks or [music] the latest news. You're going to learn what real business owners, I have over 80 employees, my companies do over a million dollars per month, are using AI [music] for right now.

So, if it's okay with you, let's go over to my screen and let me show you the three use [music] cases I'm now using Astra and Codeex for that are completely changing the game for me. All right, so right now we're inside of the Codeex app. Now, when I'm using Cloud Code, I'm typically using it like this. I'm in the terminal. This is a CMOX app. However, when I'm using Codeex now, I'm pretty much exclusively using it inside of their app.

They crush the app and it has some capabilities that Cloud Code just can't do. Now, I'm going to show you something right now that I showed my community a couple days ago on one of our insiders calls and it just blew their minds. We spent the entire time just talking about this and how insane this is. Open services in the top navigation. Choose all services. Select AM. For cloud code access, choose policies under access management.

In the upper right corner, choose create policy. On specify permissions, choose JSON. Copy the policy. So, what are you watching right now? Well, as a as an educator, you know, I've got school communities. I'm making content right now. There's a lot of times that I need to teach you something that's just too boring or redundant to do on a YouTube video. I am not going to walk you through the Amazon AWS platform right now and click through and tell you how to copy and paste JSON.

That's just stuff that is not going to work very well on YouTube. And also, it's not something that I really want to do over and over again for my community. This skill was created inside of Codeex by Astra and can only be done by Astra. And I'm going to tell you why. in addition to a couple other really insane skills that I'm doing inside of Astra explicitly right now because it just crushes every other model. So, what this is is called my demo video skill and it's one of a few skills that I've built with Astra for a couple of important reasons.

Let me show you what those are. So, this is how my demo video skill works. And the most important part, the reason Astra is the one involved here is the browser control. So, in order to make one of these demo videos, like I said, that wasn't me. I wasn't the one doing the clicking. I wasn't the one even speaking or teaching. That video was created through a screen recording with Astra actually coming up with the plan of like where to click, how to click, going through recording the screen share.

So, it did this recording, built that proof clip, built the uh screen recording walkthrough, locked the script, and then finally once it had the screen recording, actually going through and clicking through AWS correctly, it then built my Hey Gen avatar using Avatar 3, assembled it all with the the script and the voice and made sure it lined up with the clicking and then did a quality assurance bundle to make sure it was actually correct.

Now, this is only possible because of how insanely good Astra is at identifying where to click on a browser and using the browser. And that becomes super important for these next skills that I'm going to show you. And by the way, guys, as I'm going through here and I'm showing you, okay, like this skill looks sick. I really want this. All of these skills look sick. What are you even using to generate diagrams like this?

These are all skills that I'm building and releasing for my community all the time. So, if you're ever curious, head to the AI Automation Insiders. It's going to be down in the description. We've got a free version as well that gets you a lot of free resources. So once we realize just how good Codeex is and Astra is at doing demos, actually going into an app, going into a website, figuring out where to click and clicking [snorts] around, where can we expand that to?

What else can we do with that? This might not be helpful for you. Maybe you're not a content creator, but I guarantee the next skills that I show you that have to do with the same concept are going to be. And the next one I want to show you is called walkthrough. So, when I'm in my Cloud Code sessions and I switch between different paid Cloud Code accounts cuz I burn through tokens like crazy. If you're like me, this is going to be even more important for you.

But what I've realized is that Cloud Code just isn't that good at opening up a browser and figuring out where the UX is wrong. Not the UI, the design. and it can do okay when it comes to design, but if you're building an actual app or you need to improve something when it comes to user experience, where things are overlapping, where buttons should be appearing, cloud code just doesn't really do that well. However, if you give that same walkthrough code to Codeex and ask Codeex to click through and debug things and make sure that something is user or customer ready, Codeex does a much better job, especially when using Astra because it's so aware of where to click of user experience and it's able to actually give live feedback and make live changes.

And that's why anytime I'm doing anything that requires actually opening a browser, clicking around and figuring out if it looks perfect, if it functions perfectly, is everything where it needs to be. This is a great example. I gave this uh handoff to Codeex. It actually clicked around and improved this go highle page for me. And here's another example where it it fixed the page. This is what my skills pages look like.

And this whole thing here was messed up cuz Cloud Code did it. However, Codeex came in and fixed it in a single pass because it's really good at opening up and seeing the user experience. In fact, I've got a skill for this too. It's called live test. So I'll run /walkthrough or slash test and it'll actually hand it over to Codeex so that Codex can take it over, open up the browser, click around and improve things. So so far we've got demo videos, we've got user experience really opening up and and fixing the user experience.

The third thing is a little bit advanced for a lot of people. I'm not sure everybody's going to get the same value out of this next one, but for those nerds out there like me, this thing is insane. So let's open up Codeex again and let me show you what I'm currently doing with this. I'm going to open my social media tool. Expand school CLI coverage. This is something I'm actually using this for in real life and real time and I do this all the time.

So, a lot of apps will publish a public API. The things that you can ping their API with with your API key and that will do some sort of action in the app. The problem is a lot of apps don't give you access to all of the actions that you possibly can do. For anyone that's been following me for a while, this is very well documented in my gohigh level system that's called fully auto go high level where what we did here is we deconstructed their internal APIs.

When you click around go high level, it fires off all kinds of API calls that are not exposed for us to use and we reverse engineered it. And now in goh high level we can build funnels, we can build workflows, we can manage our sub accounts, all the things that their native API does not allow us to do. But that doesn't restrict you to only doing this with gohigh level. You can do this with every single application. One that I really wanted to build an API for, but the API just doesn't exist, is school, where I host my communities, where I host my courses.

I would love the ability to post in there, to comment in there, to respond to comments, to post lessons. I want my cloud code to be able to do that so it can help me with those things. However, they've got no API at all. How does Codeex get involved with that? What is the skill? How does it work? Well, as you can see, when I started here, it was to expand school CLI coverage. and I created a handoff to add the tooling to our school CLI.

The way that this works is it will open up an actual browser and because the browser is open and Astra is so good at reading it, it'll be able to click through, figure out where to click and figure out what API is triggered in the browser based on where it clicked. Saves the API endpoint so that we can deconstruct it. And once all of those different API endpoints are mapped, now all of a sudden we can give it over to cloud code, give it over to codeex, and we can do just about anything that we want inside of school.

We did the same thing here for YouTube. So, I wanted my cloud code to be able to publish community posts on YouTube. There's no API endpoint for this. There's no way to do this. I actually had it build the same internal API so that we can figure out how to publish community posts on YouTube just through Cloud Code or through Codeex. And it figured it out in 2 seconds. And now we're able to make community posts inside of YouTube just by me telling Cloud Code to do it.

We've built CLIs for every cold email tool. We're building it for all of my course tools. And these are also going to be available for members of my program. Now, there is a GitHub repo that's free that can help you with this. Now, this is free and it's very popular. It's got about 50,000 stars. It's called CLI Anything and it's kind of useless just in cloud code cuz it's not very good at this browser control. However, you give this over to Codeex and there's just about nothing that it can't create your own CLI for.

So, if you have a software or a tool that just doesn't have great API and there's stuff that you wish that you can do, hand this GitHub repo to Codeex and ask it to build you an internal API for that tool. And now with Astra and with browser control, it's never been easier. All right. Now, believe it or not, that's only the first insane use case, which is how good it is at using the browser. There's two other use cases that I'm currently using Codeex and Astra for right now that are even wilder.

Literally even wilder. So, if you thought any of that was fun, nerdy, helpful, just wait until you hear these next two things. And just before I go on, quick ask from you. If you appreciate me making this free and actually giving you real value here on YouTube instead of just updates and news, it would mean a lot to me if you like this video and subscribe to the channel and you'll be doing yourself a big favor. So, let's go into the second crazy thing that we're using Astra and Codeex for, and that's what I like to call the Codeex Bridge.

Now, this is the Codex Bridge, and it's completely free. If you want to go to the link, it's going to be linked down below. so that you can install it and start using it. Now, what this skill is is a way for you to get the most out of both Claude Code and Codeex. Now, I told you earlier that I canceled Claude Code to switch to Codeex. That's not entirely true. So, I used to have three Cloud Code fully maxed out subscriptions and I would just switch between them because I was burning through credits so fast, especially with Fable, which is the only really good model right now.

And now I actually have two fully maxed out Cloud Code subscriptions and two maxed out codec subscriptions. And I switched between those two. And what I've realized is Codex is actually a lot more generous with credits, but two other really important things. It makes an unbelievable reviewer and coding companion. So what this Codex bridge does is it gives you the maximum power of both Fable and Astra. But most importantly, when you're actually writing code, you don't want Fable to write that code cuz you're just going to be burning through tokens.

The best way to actually do it is to use Codeex's lower tier models to write the code cuz it's way more accurate and it's way cheaper and it's better than Opus. So, if you're using Soul 5.6 or even Teral Luna to write some of the cheaper code, you're going to save so many credits from Claude Code by doing this and using this codeex bridge to review, it's using Fable to talk to both Astra and 5.6 Soul to come to a consensus agreement about a plan.

And that is going to save you more tokens than anything because the more refined your plan is, the less work that you're going to have to do debugging and going back and forth and back and forth trying to get code that actually works. So we use this codeex bridge both for consensus planning. So you can see uh just briefly kind of how it works. Codex review iterates a plan to consensus. Then GPT6 Astra ratifies risky approvals.

So it'll go back and forth with Soul 5.6 until it reaches a consensus that gets approved by Astra. If not, it goes back into consensus mode. The more time and energy that you spend planning, the less time and energy that you're going to waste, not just time, energy, but tokens. waste coding and reiterating because the code was broken or you miss something in the planning phase. And then we run codeex execute which routes items across Luna, Terra, soul and it ladders by risk and runs them in parallel.

So what does that mean? It's going to spawn multiple codec sub aents and it's going to code faster and more accurately and you're going to be saving on cloud code tokens. Why are you saving on tokens? The reason for this is this terra model is way better than sonnet and soul is better than opus when it comes to writing accurate code. So, if you're using codeex, which is way more liberal with credits, it's actually harder to run out on codecs, this is going to save you a ton of tokens in the long run, and you get the best out of both worlds when doing this.

So, let me just come into my actual terminal window and kind of show you what this looks like in practice, cuz I use this for just about every workflow. So, if I'm coming to Consulty, Consulty is my software application, and it's where my most complex code lives. So, I'm just going to come in here and open up another window and open up another terminal session. And when I'm starting a new session, as you can see, I'm I've got Opus open here.

If it's a really tough session, I'll change the model to Fable, meaning I'm about to do something really complex. And then I'll start by saying codeex review. And what this is going to do is it's going to enter that consensus planning mode. So whatever task that I give it, it's going to use codeex soul 5.6 to go back and forth right now with Fable to come to a consensus agreement, get it ratified by Astra, and then once that plan is done, we're going to use codeex execute.

And they're both obviously connected because it's the same skill. So when I use codeex review, it suggests to use codeex execute after. And when I'm using Fable, it knows to use Fable only as the orchestrator and reviewer. That way I'm only using those expensive Fable credits to spawn sub aents, coordinate those sub aents, and review their code. Make sure that it's accurate and correct. Now I don't need three fully loaded Cloud Max subscriptions.

I can survive with two, and I barely ever burn those two out. Now, if you're interested in this skill, this Codex bridge skill, it's going to be free. It's linked down below. Hope you enjoyed it. And just a last note here on this codeex bridge. I told you that my consulty stuff is by far the most complicated stuff that I'm doing inside of my business. It is. In fact, right now I'm building a fully autonomous AI SDR. It's a really complex feature.

This would be impossible without the codeex fable superbrain that I've now built with this codeex bridge. It's able to figure things out that Opus could have never figured out before this new generation of models was born. So take advantage of this Fable plus Astra. It will literally solve problems that you could never have solved without them. All right, now the third and final use case that I want to share with you.

And this one is also pretty damn wild and it will save you a ton of money because in the past we were using something called foul.ai for just about all of our image generation because I just plug cloud code into this and now it has access to all of the latest image generation models. And our old way actually worked really well. You know, we would use the latest image generation model inside of cloud code. We would give it access to foul AI so it could access all of these generative image models and it would go through and generate amazing looking thumbnails like this.

However, each time that it generated a thumbnail, there would be a QA round, meaning it would review the thumbnail, decide if it was great. If not, it would keep generating new images, new images, new images. So, every time it generated a new thumbnail, it would be like $5 cuz we're using the latest image models. And each thumbnail takes 5 to 10 generations. Gets very expensive. Now, if you're running Facebook ads, if you're a content creator, if you're doing any sort of graphic design at all, this next thing that I'm about to show you is an ultimate unlock and ultimate hack.

In fact, I'm a little bit worried to share this with you because if this gets overabused, I think Sam Alman might pull the plug and change some rules. But until then, this is a loophole. What I'm about to show you is a freaking glitch. So, if I was to use GPT image 2.5's latest model, this just came out. Let's just say image editing flare. I actually don't know what the most expensive one is, but it is not that cheap.

Look at how expensive these tokens and image generations are. It is it's not nothing. Not nothing at all. So, if you're like me and you're trying to save money while producing a massive amount of images and not having to look at the images every 2 seconds and say it's wrong because of this, it's wrong because of this. You want your AI to do that for you. Run that quality assurance loop and only show you the final finished product.

Then what I'm about to show you is nuts. And what that is is a loophole inside of the chat GPT app. And this exists inside of the web browser app as well. But inside of the web browser, they don't give you the same capability. So I call this codeex thumbnails. And what this is is a self-reviewing loop inside of codeex that has native access to the GPT latest image models. So we're not using it through foul AI. We're not using it through any kind of harness.

We're using chat GPT's native image generation just like we're doing it through the browser. And you know what the limit is? Do you know how expensive it is to use it inside of the Codex app? $0. You get basically unlimited access to their latest image models and it will keep generating and reviewing inside of the Codeex app without you having to pay for any of those generations cuz you're already paying for the Codeex subscription.

It's included in your subscription. However, if you're to use the API for OpenAI or to for FAI and you're going to grab an OpenAI API key to generate images that way and give that API key to cloud code, doesn't matter what subscription you're on, it's going to charge you. It's going to charge your API key. However, if you do it in here, you don't need an API key. It's got the tools natively built in. So, literally all I told it here, use the latest skill to produce three new thumbnails for this video.

Use our approved thumbnail templates to find good matches for concepts and ideas that would be good for us to use. do research and on some other popular thumbnails and find good headlines and concepts to use. It went for 13 minutes. It did all of this Q&A. I found three strong matches, found the headlines. Reviewer caught a real issue in the whiteboard draft and it redid it. Review loop caught facial texture introduced by the whiteboard edit.

So, I'm rebuilding it. All of those, if I were to use the API, would cost money. All of those images that came out bad and had to be redone using this QA loop, this auto improvement loop would have probably cost me several dollars just to get this one. Actually, I got more than one thumbnail here. If you want to see all of the ones that it generated, they're all pretty damn good. After two failed repairs on the same defa defect, it actually stopped.

So, now I'm going to do something cool. I'm going to run autoheal or auto improve. So, anytime any of my skills work through something broken, I like to run self-improvement loops on it so it never deals with that again. And I've got a few of those accessible in my leadj.com/skills database. But I just wanted to show you how it does all of this basically for free because image generation is included inside of codeex and codecs can spawn all of these sub aents.

They cannot do that inside of the chat GPT app. So, what that means is instead of you having to look at and give feedback on every single image that it generates in the web browser, now it can do those rounds for you and it can self-improve and fix the skills and fix the workflows all without you having to pay for graphic design. Now, with all of that being said, does it mean you should cancel your Cloud Code subscription?

No. It's still my favorite harness to use when I'm doing the heavy lifting and the hard work. I am still doing it inside of my terminal window. This is the CMUX app that I use to basically coordinate all of my different projects and apps that I'm using to run my business. That's all cloud code. However, Codeex is carving out a bigger and bigger space for itself. Before Astra came out, I never had this Codeex app open on my computer ever.

Now, I always have it open. I've got my Codeex on the left and my Cloud Code on the right. And I only see these models working closer and closer together and more and more responsibility being handed over to Codeex. As each of these companies kind of finds their specialties, finds their niche, and gets better and better at more specific things, I think this is going to become more and more true. I did a video recently about Grock and how their latest model was actually really good.

Before Astra came out, I was primarily using Grock as my coding companion inside of Cloud Code. I would run /grock and have Grock help with the consensus review and execution cuz it was really good at writing code. I'm using that less and less now, now that we have the newest codeex model. You just don't need all of it. But all these models are going to have a place. And now Codeex is one of my most used applications.

And before I get done here, there's one other little hack that I wanted to share with you that just came up. If you're not already doing this, you need to check this. And that's come down below and come into usage. They actually give you these usage limit resets. So if you haven't already done this, or maybe you're not even close to hitting your limits, but if you are, I've already reset my limit three times this month.

Cloud Code really doesn't let you do that the way that Codeex does. It's almost like unlimited credits. How the heck do you run out of credits and codecs if they let you reset their their their limits multiple times in a single month? This is crazy. And we're at a point right now where a lot of our use of these large language models is subsidized. I have a strong feeling that I'm getting a lot more than $200 per month of value out of codecs that I'm paying them.

I'm probably using $1,000 a month worth of credits and they're charging me 200 per month because it's subsidized. It's investors money. Same thing with claude code. So, you're in a golden age right now to use as much AI as you possibly can. And to do that and to stay upto-date with what you should be doing and the best practices and how you should be doing it and all of the latest skills and AI systems, you've got to be parts of communities.

You got to be following channels like this. And I actually have a community that you can join right now completely for free. It's called the AI Automation Insiders. We've got over 5,000 members all learning and growing and building AI together. It's going to be linked down below in the description if you want to join. It's completely free and there's going to be free resources pinned for you. Anything mentioned in this video is going to be pinned inside of this community.

So, head over there, grab it, and I've never been more excited about AI in my life. Hope you found this helpful. If you did, please drop a like, leave a comment if you agree, disagree, or want to talk about anything that I mentioned.