Every Claude Code User NEEDS To Watch This

CChase AI
컴퓨터/소프트웨어창업/스타트업AI/미래기술

Transcript

00:00:00"You need to be tool agnostic or you're gonna get screwed."
00:00:03A perfect illustration of this just occurred a few days ago
00:00:06when Anthropic came out and said,
00:00:08"They're gonna change how they bill
00:00:10"programmatic cloud code usage."
00:00:12And if you're all in on a single vendor,
00:00:14changes like this can have significant negative effects
00:00:19to your wallet.
00:00:20So in order to prevent you from becoming a prisoner
00:00:23to a single tool, we need to learn how to diversify.
00:00:27And that's exactly what we're gonna do today.
00:00:29As I show you how to introduce codecs into your stack.
00:00:33And the best part about using codecs isn't even the fact
00:00:35that GPT 5.5 is really strong
00:00:37or that the desktop app is extremely intuitive.
00:00:40It's the fact that Claude code and codecs,
00:00:42the Venn diagram of these two tools is essentially a circle.
00:00:46Once you've mastered one of them,
00:00:47you've pretty much mastered the other.
00:00:49So if you're an experienced Claude code user,
00:00:51you're in a great position to start leveraging codecs.
00:00:54And in this video, I'm gonna give you a speed run
00:00:56of what you need to know
00:00:57so you can do it as efficiently as possible.
00:01:00So we've got two versions of codecs to think about.
00:01:02We have the codecs desktop app and we have the codec CLI.
00:01:06If you're someone who wants to use the codec CLI
00:01:08purely through the terminal, it's extremely easy to set up.
00:01:10We're just running a few lines of code.
00:01:12And then from there, I'm not really joking.
00:01:14You just treat it like you would Claude code.
00:01:16That whole Venn diagram being a circle totally applies here.
00:01:19Now there are some subtle differences.
00:01:22It's not completely the same,
00:01:23but in this video, we are going to be taking a look
00:01:25at those differences through the lens
00:01:27of the codecs desktop application.
00:01:30Now, why the desktop application?
00:01:32Because I like it.
00:01:33Super intuitive, it's nice to use,
00:01:34and it's easy to sort of explain all the differences
00:01:37through that lens.
00:01:38Also, I say that as someone who uses Claude code
00:01:41purely through the terminal.
00:01:43I like using codecs on the desktop app.
00:01:45I kind of just like mixing it up.
00:01:46And as you'll see, you can also use the terminal
00:01:49inside of the codecs application,
00:01:51which is a great way to use both Claude code and codecs.
00:01:53So understand everything we talk about,
00:01:55even though it's on the desktop app also applies to the CLI.
00:01:58Now, before we install and get it running,
00:02:00let's talk about codecs pricing very quick.
00:02:01'Cause this is a place where it wins versus Claude code.
00:02:05The open AI sort of philosophy on usage
00:02:08is much less draconian than anthropics.
00:02:11So you're going to want to be on the plus or pro.
00:02:15The difference is the $20 a month plan
00:02:17versus the $100 a month plan.
00:02:18And they also have a 20X pro plan similar to Claude max.
00:02:23The big difference is if I'm on the pro plan at $100
00:02:26versus the plus plan at 20, I get access to GPT 5.5 pro.
00:02:31So there's GPT 5.5 and there's GPT 5.5 pro.
00:02:355.5 pro is better, obviously.
00:02:38The base level 5.5 is still really good.
00:02:42So if you're someone who's already paying $200 a month
00:02:44to anthropic and you're like,
00:02:46I kind of just want to try codecs.
00:02:47I don't want to go nuts with it off the bat.
00:02:49Just do plus, just do the 20 bucks a month on 5.5.
00:02:52Your usage is still really good
00:02:53because the usage at $20 a month on open AI
00:02:57is better than $20 a month in anthropic.
00:02:59There's not like a great way to do like a one-for-one test,
00:03:02but 5.5 also just uses less tokens in general.
00:03:05So that's my take on it.
00:03:07You're going to get more usage, more bang for your buck.
00:03:09So to get it started,
00:03:10you're just going to go to openai.com/codex,
00:03:12download and install it.
00:03:14When you open up the codecs desktop app,
00:03:15it's going to look something like this.
00:03:17Very similar to chat GPT and also pretty similar
00:03:20to the cloud code desktop app.
00:03:22We have our prompt window.
00:03:23We have the ability to add photos and files,
00:03:26plugins, plan mode.
00:03:27We have sort of our permissions set up
00:03:29similar to how we set permissions inside of cloud code.
00:03:32I can set my model, the speed and the intelligence,
00:03:36and we're pretty much ready to go.
00:03:37From here I can also select where I want my project to live,
00:03:41what I'm working on.
00:03:42Now, before we demo it in action,
00:03:43let's take a look at some of the settings.
00:03:44So if I come down to the bottom left and I head to settings,
00:03:47so the first tab is general.
00:03:48You're going to want to be on coding
00:03:49to get more technical responses.
00:03:51And then it's going to ask you about permission.
00:03:53So this is similar to sort of the permissions with cloud code,
00:03:56what it can edit and what it can edit
00:03:57without actually asking you.
00:03:59For me, I enable all of these
00:04:01because I run on full access the entire time.
00:04:04Beyond that, it's sort of just asking you stuff
00:04:06about your environments
00:04:07and what sort of speed and all that you want.
00:04:09So for speed, understand if you go to fast,
00:04:12it's charging you 1.5 times the tokens.
00:04:15Another interesting thing is follow up behavior,
00:04:17queue versus steer.
00:04:18So you're going to want to leave it in queue for default.
00:04:21That's saying, hey, if Codex is in the middle
00:04:23of some sort of tool call, it's doing something
00:04:25and I give it another prompt.
00:04:27If it's in queue mode,
00:04:28it's going to wait to finish everything
00:04:29before it hits my next prompt.
00:04:31If it's on steer mode,
00:04:33then whenever I give it a follow on prompt
00:04:36while it's doing something,
00:04:38after it finishes its tool call in the middle of it,
00:04:41so maybe it's doing 20 tool calls.
00:04:43At tool call five, I add another prompt.
00:04:44It's then going to inject that prompt.
00:04:47So sometimes you want that
00:04:48and as you'll see, we can manually press steer
00:04:52whenever we give a follow on prompt,
00:04:53but that probably isn't how you want it to default.
00:04:55At least I don't.
00:04:56So that's what that is.
00:04:57Next up, we got appearance.
00:04:59So you can kind of adjust this how you want.
00:05:00You also have pets.
00:05:01Now I kind of like pets.
00:05:03Like I made my own like cool little custom pet.
00:05:04The way this works is if I'm working inside of Codex
00:05:07and I give it a prompt,
00:05:09this acts as like a little hook
00:05:12and it shows me what it's working on.
00:05:14It gives me a little like visual indicator
00:05:15that it's still working on something.
00:05:17So I can tab out, go do different things
00:05:19and this like little pet guy
00:05:20lets me know when it's done as well.
00:05:21So you'll see that in action.
00:05:23Some people don't care for it.
00:05:24I think it's kind of cool.
00:05:25Then we have configuration.
00:05:26So user config is similar to sort of the permissions
00:05:28we talked about earlier.
00:05:30When does it need approval to do things?
00:05:32This is you sort of setting the global values for that.
00:05:35You're also going to want to enable Codex dependencies
00:05:38and we have the user config.
00:05:40So if I open config.toml,
00:05:42what we're looking at here
00:05:42is just an internal Codex settings file
00:05:44that says some of the like plugins and things
00:05:47you have installed.
00:05:48One thing you probably will want to add to this
00:05:50is the goals feature.
00:05:52So at the very bottom,
00:05:52if you add a line and you manually put in feature
00:05:54and then goals equals true,
00:05:56that will enable the goals feature inside of Codex.
00:05:59I have a whole video on that.
00:06:00I'll link that above,
00:06:01but it's a nice way to use Codex
00:06:03as a long running agent to harness.
00:06:04Next, we have personalization.
00:06:05This isn't an agents.md.
00:06:07This isn't a claud.md thing.
00:06:09This is sort of just giving it a quote unquote personality.
00:06:11How do you want it to actually talk to you?
00:06:12And then you have memories as well.
00:06:14Memory inside of Codex,
00:06:15same way as memory inside of claud code.
00:06:17If I say, hey, every Tuesday I go to the gym,
00:06:21just gonna be like, oh yeah, let's write that down.
00:06:22So it's kind of just like, you know.
00:06:24I'm not a huge fan of these like goofy memory things,
00:06:27but if you want it, you can add it.
00:06:28Beyond that, the rest of this has to do
00:06:30with just like hooks and git and environment.
00:06:32It's relatively self-explanatory.
00:06:34Certain things like computer use,
00:06:37and you might have another thing called,
00:06:40I believe it's called like archiving or something
00:06:42to that effect, are Mac only.
00:06:44So there's certain things that you can only do
00:06:46when you are using the Mac OS.
00:06:48Now let's talk about actually using this thing.
00:06:50You kind of have two options.
00:06:51You can either be in a chat or you can be in a project.
00:06:53If I'm in a chat,
00:06:55it's kind of like I'm inside the claud desktop app
00:06:59and I'm just on like the claud chat bot section.
00:07:01Like this isn't really saving anything into a project.
00:07:03So I can say something like, hey,
00:07:05what's the difference between claud code and Codex?
00:07:11You know, it's gonna do its thing,
00:07:13but it isn't working inside a specific directory
00:07:16that I've set.
00:07:17So if you want to do that,
00:07:19let's say you're actually working on a real coding project.
00:07:21You're gonna wanna be in projects.
00:07:24So if I click on project,
00:07:25I can start from scratch or use an existing folder.
00:07:27I almost always do use an existing folder.
00:07:29So I have a little more, you know, granular control
00:07:32of where I'm putting this folder,
00:07:35but let's do a new folder and we'll call it CC to Codex.
00:07:43Select that folder and I now have a project over here.
00:07:46And you also see this as saying,
00:07:48hey, do you want to migrate some settings?
00:07:52So if I hit continue,
00:07:53what it's doing is it's importing all my settings
00:07:55from claud code into Codex.
00:07:57All my skills, all my plugins, all that stuff
00:07:59is now just like that inside of Codex, it's that easy.
00:08:02And so now if I'm here,
00:08:04what I can do is I'm now working inside my claud code
00:08:07to Codex project and this is what you wanna do
00:08:09if you're really doing anything of substance.
00:08:11Do you wanna be working in a project?
00:08:13So I can set the folder, I can set where I'm working,
00:08:16whether that's locally or in a new work tree
00:08:18or inside of Codex web,
00:08:20and then I can actually select the branch.
00:08:22So let's continue on this.
00:08:23Let's say, hey, can you create me an Excel document
00:08:28that lists out the top five differences
00:08:32between claud code and Codex?
00:08:34So we wanted to create this Excel document.
00:08:36Well, I wanted to also use the Excel plugin.
00:08:38So if I do the at symbol, you can see up here,
00:08:41I have a number of plugins and they're spreadsheets.
00:08:44So if I do that, it's going to get to work.
00:08:47Now, what are plugins?
00:08:48What are spreadsheets?
00:08:49What are all these things?
00:08:50So plugins are just like plugins inside of claud code.
00:08:53So if I come up here to the top left,
00:08:54I have a series of plugins
00:08:56and what's the difference between plugins and skills?
00:08:58Kind of a blurry line.
00:08:59Plugins tend to just be like kind of official skills
00:09:04that have a little more to them.
00:09:05So you can see some of the ones I've already enabled
00:09:08up here, things like spreadsheets
00:09:10and presentations and Chrome,
00:09:12but you can add a number of things.
00:09:13So I've added Vercel, I've added Supabase.
00:09:17And if you want to add any of these things,
00:09:18literally all you do is just hit the plus button
00:09:20and installs it.
00:09:21Usually you have to log in to that system.
00:09:23It's very, very simple.
00:09:24Up here, we also have skills.
00:09:25So here we are inside of skills.
00:09:27If I click on one of the skills that are installed,
00:09:29like the plugin creator,
00:09:30it brings up exactly what's going on inside there.
00:09:33I can enable it, disable it, try it, uninstall it, all that.
00:09:37I come up to the right, I can easily create a skill
00:09:40or plugin, so it brings up the skill creator skill.
00:09:43And so it's pretty intuitive to sort of control
00:09:45all these things in a visual format.
00:09:47Now, taking a look at our project,
00:09:50you can see our chat over here
00:09:51that we were working on earlier.
00:09:53I can pin it, I can rename it, delete it, all that,
00:09:56or I can click back into it.
00:09:58So I'm back inside our Cloud Code to Codex chat,
00:10:02and we can see it working right here.
00:10:03And while this is working, let's talk about automations.
00:10:06Automations, just like with skills, relatively easy to set up.
00:10:10It has a bunch of templates you can work with,
00:10:13but I can also just hit new automation up top.
00:10:15I can describe whatever automation I want to do,
00:10:20give it a title, auto one, put the work tree, the project,
00:10:25and when I want it to run.
00:10:26I can also set what model I want to use,
00:10:28as well as the reasoning.
00:10:30That easy.
00:10:31And like you kind of saw me do earlier,
00:10:33if I ever want to use a particular skill or plugin,
00:10:35Codex is smart enough to know when I say,
00:10:38"Hey, can you create me an Excel document
00:10:40to use the spreadsheets plugin?"
00:10:42I just explicitly called it.
00:10:44So if I want to explicitly call a skill or a plugin,
00:10:47I just use the at sign.
00:10:48So you can see all the plugins.
00:10:51I can also do forward slash stuff.
00:10:52So if I do like forward slash front end design,
00:10:57you know, you can see the front end design skill there.
00:10:59So again, very, very similar
00:11:01to how you would use Cloud Code in the terminal.
00:11:03So it finished up its Excel write-up,
00:11:05and I can actually view the Excel thing
00:11:07inside of Codex itself.
00:11:09And then if I actually want to see it inside the folder,
00:11:11I can just click right here, go to open in folder,
00:11:13and here's the actual Excel file opened up in Excel itself.
00:11:18So again, really intuitive.
00:11:20Then we can do something like,
00:11:21let's create a landing page for a website
00:11:24that talks about exactly what's in this Excel file.
00:11:28Go ahead and use the front end design tool
00:11:31to help you create that.
00:11:32So now it's going to use this Excel file
00:11:34as sort of its inspiration to create a landing page
00:11:36that talks about the differences
00:11:38between Cloud Code and Codex.
00:11:39And I will show you how we can also have it populate
00:11:42inside this sort of like sidebar,
00:11:44and we can leave comments and things like that.
00:11:46That again, makes it very easy to do things
00:11:49like front end design from the Codex desktop app.
00:11:52And one other thing like you see here,
00:11:54you can very quickly see what sort of changes it makes
00:11:57with the code.
00:11:58You can undo it, you can review it.
00:12:00So if you review it,
00:12:00it actually brings it up here in the side panel.
00:12:02So again, so easy to use.
00:12:04Now, remember how I also talked about steer versus queue.
00:12:07So if I give it a prompt while it's working saying,
00:12:09can we make this in a Neo Brutalist style?
00:12:12You see right now it's queued.
00:12:13So it's going to wait till it's done.
00:12:14If I hit steer right here,
00:12:16it now essentially injects that
00:12:19into the current working tool calls that Codex is using.
00:12:22So, you know, it's doing something.
00:12:24I pretty much tapped it on the shoulder and said,
00:12:25hey, by the way, before you finish, make sure you do this.
00:12:29So it's a nice little thing to have.
00:12:31Also, while we're here, other things to talk about context.
00:12:34So you can see right here, if I mouse over this window,
00:12:36you have the context window.
00:12:38It's only 258K tokens on 5.5 Pro versus 1 million on Opus.
00:12:44I would argue that's not actually a downside
00:12:47when we think about things like context rot.
00:12:49And because it's so small at 258, when it auto compacts,
00:12:53it's less likely to have sort of that auto compacting drift.
00:12:57Now you still can get that
00:12:58if we just let it auto compact again and again
00:13:00and again and again, it kind of starts to struggle, right?
00:13:05Like any system does.
00:13:07So, you know, if you're someone like me,
00:13:09who's very aggressive with their slash clears,
00:13:11there is no way to slash clear here.
00:13:13All you would do is you would just start a new chat
00:13:17inside the same project, right?
00:13:19That's the equivalent of doing a forward slash clear.
00:13:21And again, very easy because all the new chats
00:13:23will show up here inside of the project section.
00:13:25So overall, in terms of context management,
00:13:27I would argue it's actually easier than Clawcode.
00:13:30So here's a look at what it built for us.
00:13:31So I can go ahead and expand this panel as well.
00:13:34Looks like it took the Excel document,
00:13:38almost used that as like a background visual
00:13:39for the website itself.
00:13:41It's sort of this Neo Brutalist style
00:13:43and then goes through the top five differences
00:13:46between the two.
00:13:46So we can see Clawcode versus Codex, on and on and on.
00:13:50Now let's say there were certain things
00:13:52I wanted to change about this.
00:13:54So what I can do is I can like highlight something
00:13:58and then I can leave a comment,
00:13:59say something like can we change the font color to all black?
00:14:07And you can see over here, it's now an annotation.
00:14:09And I can continue to do that over and over and over again.
00:14:11So once you do one comment up here,
00:14:14this gets changed to annotating
00:14:15and I can go ahead and select any sort of component
00:14:20of this webpage and continue to leave more and more comments.
00:14:24Again, compare this to working purely inside the terminal,
00:14:27whether that's with the Codex CLI or with Clawcode,
00:14:29a little harder to do this kind of stuff.
00:14:31You can, but it requires other IDEs
00:14:34or additional skills and plugins.
00:14:36This is all out of the box.
00:14:37And another cool out of the box feature with Codex
00:14:39is because it's part of the OpenAI ecosystem,
00:14:41it has stuff like GPT images too.
00:14:45So it can create images for you.
00:14:47We don't have to hook up anything
00:14:48like a Higgsfield CLI like that.
00:14:49So if I give it a prompt like,
00:14:51hey, I want you to create images that kind of go along
00:14:54with all the top five differences
00:14:55and have that be like a backdrop,
00:14:57we could do that.
00:14:57So it's pretty flexible.
00:15:00But for the most part, everything we've covered,
00:15:02even though that was relatively quick,
00:15:03like that's kind of like 99% of it guys.
00:15:06Like everything else is just like, you know,
00:15:08AI coding, AI coding in general.
00:15:12But if you're someone coming from Clawcode moving to Codex,
00:15:14you just watched this video,
00:15:15you pretty much got like 95% of it.
00:15:17I'm sure I missed one or two little things,
00:15:19but like you see, like it's not a hard transition.
00:15:22And the one other thing I didn't really mention was like,
00:15:25hey, we can also add the terminal here.
00:15:26So over here on the top right, I can toggle the terminal
00:15:29and you know, I can also have Clawcode open in here as well.
00:15:34So like I talked about, you know, big picture,
00:15:37we don't want to be just pigeonholed into one tool.
00:15:39Like I don't want to just use Codex,
00:15:40just like I don't want to use Clawcode.
00:15:41It's really easy to use these two things in tandem
00:15:44with a setup like this,
00:15:45where I have like all the benefits of the Codex desktop
00:15:48with all the benefits of the Clawcode CLI.
00:15:51So highly suggest you guys try this out.
00:15:54Just do it on the $20 a month plan.
00:15:55Don't use, you need to use the hundred.
00:15:57Just give it a whirl.
00:15:58If you like it, you like it.
00:15:59If don't, hey, no harm, no foul.
00:16:00Like you saw, it's not some crazy lift
00:16:01where it's like you're learning a whole new coding language.
00:16:04Like you're not.
00:16:05It's pretty much the same thing.
00:16:06So this is where I'm going to leave you guys.
00:16:10Let me know what you thought about in the comments.
00:16:12I also have a link to Chase AI+ in the pin comment.
00:16:15That includes not only a Clawcode masterclass,
00:16:17but a Codex masterclass as well.
00:16:19So if you really want to dive deep into this material,
00:16:22I have stuff for that.
00:16:23So definitely check it out.
00:16:25Other than that, I'll see you around.

Key Takeaway

Integrating the OpenAI Codex desktop application or CLI into a development workflow prevents single-vendor dependency while providing superior cost efficiency and native visual design tools compared to Claude Code.

Highlights

  • Anthropic changed its billing structure for programmatic Claude Code usage, increasing the financial risk for developers relying on a single vendor.

  • The OpenAI Codex $20 per month Plus plan offers higher token usage limits and better general cost efficiency than Anthropic's equivalent $20 tier.

  • OpenAI Codex 5.5 Pro features a 258K token context window, which reduces context rot and auto-compacting drift compared to Claude's 1 million token window.

  • Setting 'speed' to 'fast' in the OpenAI Codex desktop application charges the user 1.5 times the standard token rate.

  • Adding the line 'feature.goals = true' to the internal config.toml file enables long-running agent behavior inside OpenAI Codex.

Timeline

The Financial Risks of Single-Vendor Dependency

  • Relying entirely on a single AI vendor creates severe vulnerability to sudden pricing changes.
  • Anthropic's billing adjustments for programmatic Claude Code usage directly impact developer operating costs.
  • Mastering either Claude Code or OpenAI Codex allows immediate proficiency in the other due to identical core concepts.

A recent adjustment in how Anthropic bills programmatic usage demonstrates the risk of developer platform lock-in. Diversifying the development stack by introducing Codex mitigates these sudden financial risks. Because the functionalities of Claude Code and Codex overlap almost entirely, transitioning requires no extensive retraining.

OpenAI Codex Subscription Tiers and Pricing Dynamics

  • OpenAI offers a $20 per month Plus plan, a $100 per month Pro plan, and a high-volume 20X Pro plan.
  • The $100 Pro plan provides exclusive access to the superior GPT 5.5 Pro model over the base GPT 5.5 model.
  • The $20 OpenAI tier yields more active development usage per dollar than Anthropic's competing plan because GPT 5.5 consumes fewer tokens overall.

The open philosophy on usage makes Codex highly competitive against Claude Code. Developers looking to test the platform can start on the $20 Plus tier without committing to the $100 Pro tier. GPT 5.5 base efficiency ensures that even the lowest paid tier outlasts Anthropic's limits by using fewer tokens per operation.

Desktop Application Interface and Core Permission Settings

  • The Codex desktop application includes a prompt window, file upload capabilities, plugin architecture, and plan modes.
  • Enabling full access permissions lets Codex modify local project files directly without prompting for manual approval every time.
  • The default Queue mode forces Codex to finish existing tool calls before processing newly submitted user prompts.

Configuring the application settings correctly optimizes the technical workflow. Setting the general profile to 'coding' forces more technical responses, while selecting standard execution speed avoids the 1.5x token surcharge triggered by the 'fast' setting. Setting the system to Queue mode prevents mid-operation confusion, though users can manually select Steer mode to inject a new prompt into a running sequence of tool calls.

Visual Status Indicators and Advanced Configuration Overrides

  • Customizable visual indicators provide a clear graphic hook to show when background tool calls are actively running.
  • Modifying the internal config.toml file allows developers to activate hidden features manually.
  • Advanced features like local computer use and archiving are restricted exclusively to macOS environments.

Visual status indicators let developers change windows or focus on other tasks while monitoring whether the AI model has finished processing. Advanced customization happens inside the config.toml settings file, where adding a dedicated line enables long-running agent workflows. Users must note that certain deep system integration features do not function on Windows or Linux environments.

Project Migration and Native Plugin Integration

  • Codex automatically migrates existing Claude Code environments, including custom skills and plugins, during project initialization.
  • The platform natively separates temporary conversational chats from directory-locked development projects.
  • Prefixing terms with the at symbol explicitly calls specific operational plugins like spreadsheets or web browsers.

Working inside a designated project folder grants granular control over branches and local environments. During the initial setup, Codex detects and imports previous Claude Code configurations to ensure a frictionless transition. Utilizing specialized plugins allows the model to generate and modify complex external files, such as Microsoft Excel spreadsheets, directly inside the workspace.

Context Window Optimization, Visual Annotations, and Dual-CLI Workflows

  • The GPT 5.5 Pro context window size is restricted to 258K tokens to minimize context rot and auto-compacting drift.
  • The visual sidebar allows developers to highlight rendered UI elements directly and leave precise code change annotations.
  • Toggling the integrated desktop terminal lets developers run the Claude Code CLI and Codex in a single unified interface.

A smaller context window of 258K tokens acts as a defensive mechanism against the information degradation common in 1 million token windows. For frontend design, the desktop application provides a visual panel where highlighting components generates automated code edits. Developers can toggle an embedded terminal on the top right to combine the specific strengths of both Claude Code and Codex simultaneously.

Community Posts

No posts yet. Be the first to write about this video!

Write about this video