I spent weeks using cloud code and

codecs to try to figure out which one

was better to be able to share with you

which one you might pick. Turns out that

was actually the wrong question to ask.

The real question is kind of which one

thinks like you. If you've been using

Claude Code and just wondering about

this whole codeex thing or maybe you're

a Codeex user and you hear everybody

rave about how cool and how easy it is

to use Cloud Code. If you're in one of

those camps, I'm going to give you a

glimpse into what the other side is

really dealing with and why it's kind of

cool on that side of the fence. But if

you're brand new, you're new to

engineering, or you're using one of

these, but you still feel kind of unsure

of yourself, I think this will help you

understand what these differences are.

And by the end of this, I think you'll

know which one belongs on your desktop

and maybe even understand a little bit

better about what kind of engineer you

actually are. Hey, I wanted to interject

here just for a second. This is a pretty

long video and if you're just trying to

get to a winner, I have it at the very

end. That's perfectly fine. Just jump

forward to it and I'll TLDDR it. It's

Claude Code, but not for everyone. If

you really want to know why I feel that

way, watch the rest of the video. But if

you're just looking for a which one do I

need, maybe Claude Code is the place to

start, I would advise the video, of

course, you know. So, let's get some of

these really basic measurements out of

the way. These used to be ways that we

would compare these kinds of systems.

And really these days, especially with

these two, everything has gotten so

close that I don't think that this is

necessarily the battle battleground any

longer. So, there's performance

basically how they do on Swebench or

some other scored and they're both not

quite capping that out, but they're in

the high7s or mid70s kind of

neighborhood. So, they're basically the

same. And really my honest take is as an

engineer both of them are basically

excellent engineers. So I don't think

that's necessarily a separation event

anymore. The next item might be speed

like how fast they get their job done.

This one I always kind of bristle at

because as an engineer I'm not really

interested in absolutely how fast it

gets done. Now if you're going to take

40 days to do something that someone

else can do in one day then yes I care.

But if you're talking something that one

model takes 2 minutes to do and the

other model does in 30 seconds, the

problem here is the quality out of the

other side. If the quality is absolutely

measurable and you would say the same

quality can be done in a quarter of the

time, great. I'm into that. But just

saying raw speed, I think this isn't a

real value to really talk about anyway.

Use a smaller model. Get worse quality

if you want better speed. There's

systems out there right now that are

saying look how fast we are. I would

question the quality coming out of the

other side of that. Okay, the third one

here I think is quite interesting, but

it's basically a variable tie and that's

just of the last couple weeks. This is

what I call access. Where can you use

these? We're going to be looking at them

as kind of CLIs. We've already seen what

they look like in a terminal environment

and how you use them. We'll be diving a

lot deeper into that in just a second,

but that's their basic main front. You

can also use them as plugins inside of

something like Cursor or Visual Studio

Code and other systems like that. They

both have nice extensions to be able to

be used there. You can use them in the

cloud and this is where Claw just

recently caught up by releasing a cloud

system. So, you can kind of issue

commands to these on asynchronously when

you're on the bus or you're on your

couch or something like that. Once

again, another tie. And that really

brings us to the big one. I will call

this upfront a tie, and this is cost.

Basically, both of these systems work

off of kind of a tokenbased system. So,

the $20 model, you can get a lot of

really good value out of either one of

these for 20 bucks. If you're already a

subscriber, dive in. You've already got

access. But both of them are consuming

at roughly the same rate. To me, this is

just a tie. Okay. But this brings us to

a couple of the items that I think are

not necessarily ties and they show the

distinct difference between these

projects. And this is where I want to

talk a little bit. One is ergonomics and

the other is personality. Weirdly

enough, I'm describing the difference

between tools by their personality. And

I think that's actually the

differentiation that matters the most.

But first, let's step into the

ergonomics of the tools themselves, how

nice they are to use, and what kind of

offerings they have. This is a space

that there is an absolute 100% clear-cut

winner. Oh, yeah. So, I have this great

idea. I'm going to tell you what I think

is the big differentiator between these

two systems in just a second. But, in

the meantime, let's give them a side

quest. We're going to give them the same

prompt and let them kind of cook on it

while we talk about what I think is

different about these systems and then

we'll come back at the end and kind of

compare the two builds and see if that

differentiation is identifiable in the

outcome. So, first let's actually take a

look at how we're going to build this. I

have four different terminal

environments up. So, we have clauds on

the left obviously and codeexes on the

right. I understand this is probably

maddening. So, here we go. We're telling

them to read through the files that they

find in their directory and go ahead and

build everything they can themselves.

Let's let that cook and that should do

it. So, we're going to first talk about

what I think the difference between

these systems actually is and then we'll

come back and see how they did on these

builds. Okay, let me talk about the

thing that I think actually

differentiates these in a major way. And

if this matters to you, you should hear

about it. When you take a look at the

idees themselves and of course I'm in a

terminal environment. I'm pulling up

cloud code here in an application that

already exists in both of these. You'll

see this is codeex and how codeex looks.

This is how cloud code looks. So this is

just the beginning step. You're looking

at a command line basically terminal

application. And so they don't look like

a typical UI of course, but Claude Code

gets so much closer to feeling like

something that isn't necessarily a

terminal application. It kind of helps

you around. You can see it has multiple

different disclosures with color and

codec's very much like the engineer. It

tends to be is very succinct and

discreet. Here's a list of the slash

commands. If you hit slash, you can look

through the other slash commands down

here and it can scroll through them and

show you what's available. Same thing

for clawed code. If you hit slash, you

can go through the different slash

commands. So, let's ask very briefly to

give a brief description of what this

system does. Okay, what is this system

about? Okay, and here's a great example.

So, what is this system about? This is

how codeex returns it to you. Very

succinct. We're talking I'm looking for

the periods maybe three sentences here.

Uh definitely gives you a couple of the

different uh folders that it's looking

for. Now this is not as intense as

codeex can do. If I asked for a complete

system breakdown or a flow diagram or a

sequence diagram or all this other

stuff, it can do all of it. But this is

one of those things that we'll talk

about in just a minute about the

personality of these different systems.

It's not just the CLI here that acts

this way. This is the codeex model

underneath that acts this way. This is

very succinct. For example, let's go

take a look at what claude code came

back with. So these are the key features

of the system, the user experience, the

technical stack, the architecture

highlights, a much more readable kind of

definition of what's going on inside of

this system. This is a clear distinction

between what I'm going to talk about in

the next section which is one is much

more of um kind of a personality and the

other is much more of just an

engineering system is what they feel

like. But what about the tools

themselves? All right, we talked about

the slash commands. Slash commands are

are much better in clawed code. The way

that you can set up your slash commands

where you might save them. You see I

have some here at the beginning that are

my own slash commands. Those are set up

in a folder that's not inside of this

project. It's in my global environment.

But I can also set some other slash

commands up inside of this folder and

check those into a project. That sounds

very obvious that I might be able to

have some kind of slash commands that

are just basically repeatable functions

that you're asking for inside of the

project so that anybody editing on it or

me in the future, I'll have the same

function to use and then others are in

my global environment. That is not the

way codeex works/comands and kind of all

of these settings are only driven

through your global environment. Uh

Claude also does a good job of trying to

allow you to point out different

locations that you might load these from

so that you can kind of use a variety of

different places to load all of your

settings from. It also packages things

up into tasks and sub aents and other

things like that that allow you to kind

of either share packages of of this kind

of utility to different team members.

They're really trying to figure out a

way to make this application usable by

teams. This is one of those spaces that

Claude Code just runs circles. Cloud

Code also does great things like you can

take a look directly at your context and

how much context you're currently using

and it describes it very cleanly. But

over on the codec side, if we do the

same thing, they are doing their best to

kind of describe some things to you. You

can definitely get to some definition of

it. But these are the differences of

these two systems. One's much more

engineering specific and trying to

succinctly give you the answer. And the

other is trying to work with you like a

normal application. So the the

environment that you're working in in

clawed code is basically a four out of

five. And to me codeex is basically a

two out of five. It is absolutely good.

But the way that it talks back to you is

always so succinct that many times you

have to ask a little bit more detail or

can you tell me a little bit more about

this or more about that because it will

come back after working for 20 minutes

and show you the change that it's made

and just describe three things to you

and say the rest of it's done. And it

usually is done. So, I'm not saying it's

wrong, but Claude Code will come back

and say, "Here's the 25 things that I

did in the four pages of scroll that you

might want to read about it." That can

also be overwhelming, but this gets down

to these systems are there to kind of

work with you and the way you need to

work. And so, that's what we really need

to talk about. Okay, one more thing on

kind of the UX, the usability. I just

wanted to show you remember we're

cooking away on the challenge the side

project that we have going and I wanted

to show you that both of them had

finished the uh analytical right that

first task that we had and in this this

is the output that we get from claude

code it tells us what it put together

the dashboard what's on each one of them

the activities page the technology stack

how to run it how to start it what the

layout of the files that it created and

the key design decisions if we come over

to codeex and we see okay here's the

next steps by the way the next steps

that codeex offers are really fantastic

very often so this I don't want to

overlook this is a really meaningful

detail that you get in codeex but you

don't get in cloud code um the testing

that it did and how to run those tests

and that it succeeded by the way that's

kind of meaningful codeex is also very

good at running tests and this is the

implementation and by the way that's it

so it worked for 10 minutes to do this

here is the output that it's telling us

out. I built this. I wired this up. I

crafted that. Here's the polished

surface. I ran some tests. Here's what

you might want to do next. But on the

other side, it's this whole write up

about everything that it has done and

what it looks like, what the files are.

So, there's a big distinction in the way

that these things communicate and you

can see it very very quickly when you

when you run a change that's similar

like this. Okay. So, we've talked about

kind of the usability and the UX or kind

of the what I call ergonomics of using

the system. There are other ergonomics

like the the extension that you might

use in Visual Studio Code or the cloud

experience. Those are very very similar

to one another. So, I've kind of left

them off. Uh kind of go take a look for

yourself, but those are really so very

close. They follow the same concept.

Like you'll get more readout from what

Claude says versus Codex in all of these

environments because underlying all of

this, what you just saw, the CLI, is

what's driving all of these. But now I

want to drive into what I think is the

biggest differentiator between these and

it's actually weirdly personality. Okay.

So what do I mean when I talk about

there's a personality difference between

these and there really is. You've

already seen it a little bit using the

ergonomic lens that we just used. But

comparing Claude Code and Codeex, it's

not about how good the outcome is

because you know these systems are

variable in a lot of ways. So, it's very

difficult to get an applesto apples

comparison of something. And really, the

engineering out of both of them is so

high. They both are, like I said, the

mid70s and the Sweetbench kind of

benchmark, if that means anything to

you. But they're both really very good

engineering uh kind of systems, but the

question really is how you work. And

that fit matters so much. And in the

end, when I was trying to compare these

two things and I realized, but I can't

possibly compare them. The thing the

penny that dropped for me and told me,

wait a second, I can't ask the question

of which one's better any longer. These

are so close. I was using both of them

myself on my own projects. I use both of

these systems. So, the weird thing is

I'm not picking one or the other. I

understand their distinct benefits and

their weaknesses and I will use them in

the places that I think that they're the

most useful for the task that I'm trying

to perform. All right. So, let's dive

into this whole idea of personality. All

right. The way I describe this at a very

high level, Claude is let's say more

artistic and Codex is more scientific.

Really, you might see that in the

readout that we got in the ergonomics.

Codex is talking to you much more like a

very succinct engineer that's trying to

describe how did I solve the problem

that you put in front of me. Claude is

trying to give you a full output of

explanation of what you might expect.

That little detail is really what we're

here to talk about. But I don't want to

leave this as too big of of a signal. So

really there's a huge amount of

intersection between the art and the

science. These these systems are not

that far apart. And in fact, in reality,

they overlap so much that you would say,

or I would say Claude Code kind of feels

like it has a little extra artistic

flare. And Codex has a little bit of

scientific or engineering focus. And

really, we'll see that in a way that

Claude will wander a little bit. Claude

will kind of make uh tangential changes

and kind of have collateral damage to

some degree. every now and then you ask

it to do something and it does other

things also that you weren't necessarily

asking for whereas Codeex very rarely

does those. So CEX will very deeply

solve a problem from beginning to end

and Claude will miss some of those

things and I'll talk about that in just

a second but Claude will kind of explore

on the edges and in that that one detail

is probably the biggest distinction

between these two systems codecs and I

don't want to say this too too broadly

or too strongly because it they are so

close to one another. This intersection

is huge. Codeex will stick to the task

that you've given it more specifically.

Claude code will wander and explore and

basically create. So Claude is a little

bit more creative. Codeex is a little

bit more specific. It will absolutely

run the tests. It will always check

lint. It will do all of these things for

me that Claude will forget. It will

forget to run linting. it will forget to

run the tests even though I might have

asked it in kind of the the request that

I put in. But it will wander around a

little bit and say, "Oh, but maybe you

also meant to update this. You're

changing the drop menu to look like

this. I'll also change the buttons that

will share the same styling so that

they'll be similar." Codeex will not do

that. It will change the drop menu. And

I will have to come back and say, "Well,

the drop menu now looks like this, but

the buttons don't look like this." And

that kind of comparison is the

difference I think. All right. Another

thing that I will say about kind of this

personality is Claude code will work on

what you're asking it to work on. It

will notice other shiny things. It's

kind of like a puppy is the way that I

describe Claude code is it absolutely

will go get the ball. It'll get the

ball. If there's another ball, it might

also pick up the other or it'll drop the

first ball and get it'll try to figure

out how to bring both the ball and then

it's going to pick one of the balls and

one of the ball it's you will get a ball

back. which ball you get back. Not

really sure. Often when I ask for a

change, it will go make that change. It

may make other changes. It does not

always do that. I'm not kind of

representing that Claude will constantly

create uh ancillary changes. It's not

that, but it will make that that leap

more often. Um but it will kind of fix

the problem that I've asked it to fix

without necessarily noticing upstream or

downstream impacts that it might have.

Whereas when I talk about something like

Codex doing the same thing, Codex will

take that same request, make the change

that I needed it to make, but also

notice upstream, oh, I no longer need

some of these constants because we've

removed them from that in this last

change. And also downstream, you might

have more impact. Let me go make sure

that I check those. It's kind of a more

holistic version of engineering to me

where Claude code will wander a little

bit and kind of explore and get lost

every now and then. In fact, that is a a

a real distinction in my head in how

they're used. I expect a little bit more

loss when I use clawed code. Meaning,

hey, can you go do this one thing and

maybe also make short X? It's going to

forget one of those two sometimes, maybe

5% of the time. Now, it's not all the

time at all. Codeex will almost never

forget those things these days. It feels

like CEX will absolutely do what you've

asked it to do, but it will not look

around and fix the other things that

you're thinking are obvious. Okay. And

so what does that really mean to me?

What I would say is claude code is what

I use when I'm exploring or I kind of

want to mention a kind of change and it

will kind of uh propagate that

throughout the system. Codeex is the

system that I use when I'm trying to

build an end toend kind of solution or

solve a specific problem. In fact, I

have examples where claude code will

spin and spin. I worked multiple hours

the other day trying to get Claude code

to solve a problem that is a problem as

an engineer I've had many times and I

understand how complex it is to find. So

I started walking it into the solution

saying well maybe check this other area

to see if it would find it doing that

and it didn't. And then I'd say oh well

okay maybe put logging in so you can see

what's going on and it didn't know how

to run the system to be able to see the

logging correctly. So I had to give it

the logs back every single time. And I

said,"Well, okay, how about if you

remove all of the the extraneous parts

and you just leave in the first and the

last change? Does it still occur?" And

it it noticed that these problems still

existed, but it couldn't understand the

impact between them. In fact, so much so

that Claude came back at one point I was

sharing screenshots with it saying, "No,

no, you can see here that it's not doing

what you are saying it's doing." It kept

coming back saying, "I've solved it." In

fact, that's like one of the driving

forces of clawed code. the puppy. I have

solved it. I'm glad you showed me that

problem. It was a problem. I'm finished.

I found the root cause. It's fixed. And

it's not fixed. And I would send it a

screenshot saying, "Nope, it's still

this. Here's the logging. You can tell

it's this." It actually came back at

some point and said, "No, I think you're

wrong. I think you're taking a

screenshot of it when it's just flashing

that." And it's actually turning into

something else. In fact, I'll tell you,

it was a dark mode versus light mode

theming issue, which is kind of at a

substrate problem. there's a much deeper

problem causing the thing, but it was

super clear to tell, do I end up in

light mode or dark mode? It's supposed

to go to dark mode. We're ending up in

light mode. And Claude Code was was

coming back saying, "No, no, no. I

understand your screenshot so shows a

white mode, but I think that's just a

flash of white while it's moving to

dark." So, I don't think that's really

the case. And I had to go back and say,

"No, no, trust me, this is actually the

the final state. Can you go write some

playwright tests and verify it yourself?

It never really could. I brought Codeex

into the game. It worked for about 20

minutes. It worked all the way through

writing multiple tests, failing at those

tests, evaluating the logs, never came

back to me, solved the problem entirely

until it was done and gave me basically

a small paragraph saying, "Oh, well, I

noticed this was the problem, so I fixed

that." And that's a great example to me

of when I use each one of them. So, it's

not Claude Code can't do things or is

not as good at engineering. It's just

got different skills. And that comes

back to the question of what kind of

engineer are you? So, first let's check

in on our challenge that we gave it to

see if this behavior and this

personality kind of shines through in

the actual results. Okay, so our side

quest is complete. Um, let's take a look

at the results here. So, all four of our

systems have completed and I've gone and

created a a server for each one of

those. What I'm going to do, this is

going to get confusing because there's

four different things to share. Some of

them are from Claude and some of them

are from Codeex. So, I'll show the

Claude one first and then flip over to

the Codeex similar example. So, this is

the Claude analytical dashboard. This is

it gets they're all running off the same

data, by the way. It gets uh some run

activity that I've I I have in in that

I've exported for my environment and it

gives me kind of a readout of here's the

different activities. I might be able to

go in and select only certain activities

or search for them. Um what trends do I

have? And so by the way again this is

claude. This looks kind of great right

if you ask me. This looks really really

great. Um I would love to have this

system. I guess I do have this system

which is just kind of crazy. Uh, all

right. And then the different

activities, different dashboard. As you

scroll around on the ash dashboard, you

have current activities, etc., etc.

Let's take a look at Codeex's version of

this same thing. So, this is the

analytical from Codeex. Um, and I will

say, by the way, right away, hey, this

looks really, really good. Very

interesting. Um, has a nice chart for me

to use here. I cannot seemingly move as

easily into different activities. They

just have it all mostly on one page. So,

I'm guessing paging through the

different activities down here is their

idea. Pace versus distance, heart rate

distributions. Still really very very

good. I I I think I prefer the clawed

one, but honestly, there's a certain

amount of um kind of big bubblegum

version of icons and things like that.

This gets back to some behaviors that

Claude has versus what Codeex has. Codex

being just a little bit more, I don't

know, sophisticated, more engineering

specific. So, okay, let's take a look at

the other one. All right. Again, here we

are on Claude. And this one was actually

to try to motivate people, not to dive

through your data. The intention wasn't

to kind of allow you to dive through all

the data and shuffle through it. It was

much more keep you motivated, show you

your streaks. Very simple kind of

request that I put into this. Same data.

So, I have a 2-day streak for running.

That's great. 4 days this week. I'm

doing great. Thank you. I appreciate it,

Claude. Um, my weekly distance averages.

Here's what it's looked like. Um,

alltime stats. Excellent. So, this

really, it would take me a second to

puzzle through what exactly it's trying

to tell me, but it does look pretty

attractive. I do kind of They've called

it run motive as well. That's nice. My

motivational running system. How much

rowing? No rowing. H not a surprise. All

right, let's take a look at what Codeex

did. And I will say Codeex really

knocked it out of the park here. Very

different. Uh completely different

perspective on the same problem. Here's

my weekly goal versus uh what I've done.

My consistency score kind of specific. I

have a 57% consistency.

Uh 2-day streak, 30 days this month. Uh

one day run, reset the tempo with the

next run. Uh, so it's a little more

difficult for me to understand what it's

trying to tell me, but they definitely

worked on design here. This is a very

different styling. They're not just

stealing a very simple style. And in

fact, I will say Codeex uh or Claude

Code used what looks like a very very

similar styling between the two. So I I

wanted to represent this this way so

that you could see if you go with claw

code when I say it's artistic it doesn't

mean it's the one that'll make something

look good or it's the only one that can

make something look good or anything

like that. Codeex is fantastic at this

kind of stuff as well. But as we saw on

the analytical one it's a bit more

analytical. Even on the run momentum one

I'm having to read between the lines a

little bit. Now, had I given it clear

instructions for any of these features,

which I intentionally did not, it

absolutely would have nailed each one of

them. It doesn't have a problem doing

it, it's just whether or not it knows to

wander and explore in that same way.

It's not quite the same puppy dog. Okay,

so here we are at the very end after a

lot of non-comparisons sort of and I

want to say one thing very very clearly.

Cloud codecs

extremely similar. you do not have to

worry about what's on the other side and

should I jump over to the other f uh the

other uh the other yard over the fence

kind of thing. That's not necessary.

Claude code as an IDE, as a system, the

way it's usable in your environment

actually has a little bit more

functionality in it. So, you might want

to kind of take a look at some of the

things Claude Code does. I share those

kinds of things on the channel pretty

regularly to be able to say here's some

of the things that Claude Code is doing

that CodeEx is not. But that doesn't

mean that you need them just because

they're doing them. So with that,

knowing that they're super similar, my

actual takeaway that I would say is

dependent upon your style. If you like

to kind of be very specific about the

way you ask for things or you don't want

wandering and you want it to only do one

thing, it's clearcut. Go with codeex.

Start there. Give it a shot. If it's not

quite doing what you want, try out cloud

code from time to time. If, and this is

a big if, and I think this one actually

covers a lot of people in this space,

and it's the reason that I even use

cloud code, and I actually have skills

in all of these areas. If you're an

engineer and you're not great at UI, you

need to take a look at cloud code. If

you've always been one of those

engineers, and I know a lot of you who

say, I don't know, do UI, UI, I don't do

any of that stuff. I just do the

engineering, somebody else's problem.

Cloud code can actually help you in this

space. It's creative. It will help give

you ideas. It's very valuable there. If

you're a new engineer, meaning not just

in the new engineering space,

congratulations. We're all working in

the new engineering space. The question

is, if you don't have an engineering

background, which means you don't

necessarily know all of the moving parts

of engineering and how to ask for them.

Claude code is also for you. And so this

distinction of claude code will cover

ground you did not ask for. It will

interpolate what you've asked for. It

can be irritating especially to an

engineer who's trying to get one thing

done and other things happen to also get

done. That can be kind of irritating.

But if you don't know your surface area

all that well, be it engineering or UI,

Cloud Code definitely has your back in

this space. And so I have to call a

winner here, which is if you've never

started on this or you're really trying

to figure out which one should I be in,

I would say Claude Code is the clear

winner because it covers ground that you

might not be an expert or have deep

sophistication in. It is trying to do

all of it. Codeex is trying to do what

you ask. Now, if you know how to ask for

all of these different moving parts, UI

or engineering or deep depth, all of

those things, Codeex can do perfectly

well. So, all right, that's my outcome

is I'm gonna have to call it for clawed

code. Weirdly, it's not a tie, but

really, you're not missing anything if

you're in one or the other. So, don't

worry about it. All right, this one was

pretty long. Thanks for coming along for

the ride on this one and I'll see you in

the next