Did Claude Code Just Kill ChatGPT Codex?
By corbin
Summary
## Key takeaways - **Claude Code: Code from Anywhere**: Claude Code allows you to initiate coding sessions without needing to open your terminal. You can connect to your GitHub repositories, describe your needs, and Claude will handle the implementation using cloud-based services. [00:24] - **GitHub Security and Access Levels**: When authorizing Claude Code with GitHub, choose security levels like 'trusted network access' for package downloads or 'custom access'/'no network access' for company code to mitigate risks. [01:06] - **Granular Repo Access with GitHub App**: To grant access to specific private repositories, install the Claude GitHub app, which allows you to granularly select which repositories Claude Code can access. [01:56] - **Parallel Tasking and PR Creation**: Claude Code supports parallel tasking, creating separate sandbox environments for each task. Upon completion, it can create a pull request for each task, which can then be merged. [04:12] - **Codex vs. Claude Code Features**: While Claude Code is powerful, Codex offers features like 'versions' (providing multiple code variations) and 'code review' (AI checking PRs for bugs), which Claude Code currently lacks. [07:46] - **Diversified Tooling for Development**: The best approach to using AI coding tools is diversified; use the tool that best fits the specific stage of your software development pipeline, potentially using Claude Code for mobile, Cursor IDE locally, and Codex for code review. [08:26]
Topics Covered
- Cloud Code: Code Anywhere, But Mind Your Security.
- Master GitHub: A Fundamental Skill for the AI Coding Age.
- Secure Your Code: Navigating Cloud Code's Network Access Options.
- Parallel Tasks in Cloud Code: Streamlining Development with Separate PRs.
- Diversify Your Tools: Choose AI Coding Assistants by Pipeline Stage.
Full Transcript
Clawed Code on the web just got
released. First major question is that
is this any better than Codeex or Cursor
AI? In addition, I'll show you how to
set this up and make sure you get the
most value out of it. So, if that all
sounds good, make sure to leave a like
right off the bat and let's jump in.
Welcome back, y'all. In this video,
we're going to be checking out Cloud
Code. If you want to get updates like
this as fast as possible, make sure to
check me out on X. Already did a whole
post on it, but I'm interested to see
this. Essentially, TLDDR, what's the use
case of this Corbin? You can code
anywhere. Idea being that you don't
necessarily have to have the code in
your IDE or your local computer. You
could in theory just connect to the
internet, put in a little request, and
then it's going to use cloud-based
services in order to push the relevant
code changes, which is cool. But
obviously, if anyone accesses your
Claude account, and they have access to
your repo, then your code's cooked. But
let's check it out. So, like any of
these AI coding platforms, we're going
to connect with GitHub. I'll make sure
to also leave a 40-minute guide of
everything you need to know about GitHub
in the description down below. As
obviously with all these little AI
coding bot things, you're going to need
to know how to use GitHub. It's a
fundamental skill you got to have in
this new age. In the vibe coding age,
authorized claude. So, first it's going
to prompt us with the level of security.
I'm going to go with trusted network
access. This is going to allow us to
download specific packages in order to
run applications. Although if you're
working with a real software's codes
repository, I would encourage you to
either choose custom access or no
network access as you are putting your
code at risk in theory by giving it the
ability to access the network. But in
reality, it might not be that deep in
your situation. I'm just saying from
experience, specifically when working
with real company code, there is usually
VPNs involved. Usually, you can only
access that code through a specific
network. But if none of that really
matters to you, then don't worry, just
go recommended. trusted network access.
Next, we're going to select our
repository. So, we got a bunch of
different options here. And in the short
term, it seems like it will only have
access to public repositories that are
currently associated with your account.
So, things like this public repo where I
built out an entire cloud skill, it has
direct access to, but I want to give
access to specific private repos. So,
I'm going to do install GitHub app. Now,
when installing it this way, now we have
the ability to granularize by the type
of repository or do all. For now, I'll
do all install and authorize. So, what
it does is when I do that, it takes me
back to the settings page, but let me
show you how to revoke access if needed.
So, going back over GitHub here, let's
just say you've used it and you're like,
Corbin, I absolutely hate cloud code.
Then all you need to do is come up here
to your settings. Also, make sure to
throw me a follow resource code every
single week. Just download my code
#freecode Corbin. Go to your settings
here, scroll down to where it says
applications, and you can see all the
relevant applications that are currently
integrated into my GitHub portfolio. Hit
configure. And then right here is simply
where we can change its specific access
or alternatively get out of my repo
cloud and uninstall it or suspend it.
Let's check it out though. We are going
to use test app here. Uh test app simply
is a repository that is a React app.
Literally just did one command of just
install a React front end and ended up
with this. And to be honest with you, I
don't even know why this is private. So
let me actually make this public real
quick. So I guess you'll learn something
real quick in GitHub. to make a
repository public. Go to settings,
scroll all the way down, change
visibility, change to public. I want to
make it public. And the reason I made it
public is you can simply come to my
profile here, description down below,
steal my code, find the repository that
says test app. You can download it
yourself and then do these little tests
that I'm about to show you. So once I
have it selected, here we have the
environment. The environment is
specifically where we set high-risk
variables. This right here is when your
application leverages different secret
variables or high-risisk variables like
API keys. For example, OpenAI's API key.
Maybe you have like a hybrid app that
uses Anthropic and OpenAI. You would
simply put the API key here. If in your
relevant code directory, you reference
it in your EMV, which I know I don't
have one here, but it's a test app, but
you know, or not. EMV.v,
you know, if you have any type of
high-risk variables there, you will
simply put there. If that only makes
sense, check out my courses here, and
I'll show you how to build out a
software from scratch. And when I say
courses, not like there's a payw wall.
YouTube has this new feature. If you go
to my profile and there's like a tab
called courses now, all of it's free and
it's lesson by lesson. I know courses
has like a really bad negative stigma
around it. So let's just call them learn
AI with me AI fun. So let's show off one
of its features here such as parallel
task. So I'll go and show this in the
test app here. I'm going to first
request dark UI. Second, I'm going to
request this counter go up by two and
down by two every time I press this
button. So boom. And then I'll be like,
okay, this is going off. make UI dark.
Boom. Again, so what this is essentially
doing is it allows for parallel task.
And then what each task will do is try
to create its own PR. Obviously, we can
enable notifications as well. So if you
want to just be scrolling through X or
check me out on X, you can enable this.
So I'll enable it. Essentially, what's
going on here is we are creating two
separate sandbox environments. What I
mean by sandbox environment is it's
installing all the relevant dependencies
required to run your application. And
when I say dependencies, I'm referring
to your package.json here. Everything
that's required to actually run and
render your application, it's creating a
little bubble for it to run it in just
for safety reasons and so that it
doesn't get a little crazy and it can
run this in a cloud environment. So once
this is going, we have two separate
parallel tasks going. Obviously, I
wouldn't set this up so that when you
use this, these parallel tasks are
counterintuitive. And what I mean by
that is, for example, this task was
create dark UI and then we put another
task that says create light UI. You
might as well pair that up into one type
of task. So your prompt is more meaty up
here rather than just two separate. So
once a task is complete here, you can
notice it as create PR. So I'm click
that. And because it has access to a
specific repository here, I can simply
say create pull request. I know some of
you are like Corbin, what about your
comparison with codeex and cursor? I'm
going to get there. I'm going to get
there. Let me just show you the
fundamentals. We get a nice little
description here. modify counter button
increase decrease by two. Nice create
pull request. No conflicts and I can
merge it for merge. I'm going delete the
branch and then fundamentally we just
got our first commit ever done with
cloud code and merged into our main
little app here. Obviously to pull this
into our local code I'll simply do get
most recent code from GitHub or you do
like a git pull origin main. But it
seems like nowadays we can literally
just talk to the AI and it'll do it all
for us. We don't need to know git
commands anymore. It's fun. Or are we
getting dumber? I don't know. So then
next we have this still going. So we got
the UI was a create PR right here. It's
outlining everything that changed here.
Nice. Create pull request. And this kind
of gets into the second thing I wanted
to talk about which is this can be quite
excessive for small changes to do a PR
every single time. So I'm going merge
this real quick again. Delete branch.
That's fine. We're going to see the
changes pretty soon here. But notate
that. Understand that when using cloud
code here, okay, yeah, if it's going to
be a small bug fix, fix the bug. Fix it.
But if you're doing something bigger,
then I would leverage more meaty prompts
and jive down a bigger rabbit hole when
leveraging this little chat feature
right here. So obviously I don't have to
create the PR yet, but we can create
iterations of this and keep talking to
Claude until we are satisfied with the
results. And of course, again, you could
in theory create the PR right now and
then not merge it and then create
iterations with reply to Claude. So I'm
going to pull the most recent code again
because we just merged two different
PRs. There we go. You'll know it's
successfully merged when you see this
little thing right here. Hit commits.
You can see it's getting pulled here. I
say pull is just grabbing the code. So
perfect. Coming back over to our
counter. We have a dark UI. Plus two
minus two minus two minus two plus two
plus two. I'm a robot. 0 0 1 1 0 0
binary.
Nice. But how does it compare to the
other coding platforms? If you already
find yourself entrenched in the clawed
ecosystem, then yeah, you can leverage
this. Here are some things that this
doesn't have that Codex has though. What
Codex has, and I'm actually going to
release a video here pretty soon, so
make sure to subscribe here of how to
upgrade Codeex is it has something
called versions and it has something
called code review. Here's a TLDDR of
what that means. Essentially, versions
allows you to do, hey, I want a dark
theme. It gives you three different
versions of that for your front end.
Three different little code differences,
right? What does code review mean? It is
extremely dope. Literally, you do at@cex
in the PR and then Chad GBT will check
your PR and point out bugs or situations
that could break your application. A
whole another video is coming out on
that. So, just subscribe here. It should
be coming out literally tomorrow, I'm
pretty sure, after this one. So, where
does that leave us with which tool to
use to be honest with you, it's
diversified. Eg. We use the tools that
work best for whatever part of the
pipeline of your software development
you are currently in. So for example, if
you absolutely love claude, I love cla
code here. Or in theory, our base level
could be okay local code, I work in the
cursor IDE. When I'm mobile, maybe don't
want to pull out my cursor IDE, I use
cloud code. And when I push a PR, I use
codeex to review that code. And then
obviously it all just shows up here and
you have your fresh new application. Let
me know in the comments if you want me
to dive into a video that really shows
you how to use every single one of these
tools in the most effective way because
in reality it's not a one-sizefitit all
situation. But without further ado, as
you already know with these style of
videos, make sure you leave a like. It
is completely free. I'll see you in the
next. Why do I feel like in one or two
weeks, Claude Code is also going to come
off a new ability that does code review.
So then everything you watch in this
video, you're going to be like, Corbin,
hey, do you know, by the way, that Codex
is not the only one that can do code
review, but so can Claude. Well, my bad.
I posted this video two weeks before
type of video. Nice.
Loading video analysis...