So, we've gotten pretty comfortable at

delegating tasks to our AI coding agents

like Cloud Code and our terminal or an

IDE like Cursor. But what about when

you're away from your desk or if you

want to kick off a big complex build

before you leave for the day and you

just want to come back in the morning

and find it done? That's where

cloud-based background agents come in.

And right now, we've got three major

players who are offering pretty good

options for running coding agents in the

cloud. We've got Claude Code on the web.

We've got Cursor's background agents.

And we have OpenAI's codecs on the web.

Now, on the surface, all three of these

tools are pretty similar. They all spin

up a cloud environment. They connect to

your GitHub repository and then they

create their own branch and eventually

they'll submit a pull request. So, the

basic mechanics are the same. But where

they differ, and this is what actually

matters, is the workflow. So, how do you

actually initiate a new task with a

cloud agent? How do you monitor its

progress? How do you check it out on

your mobile device? Or how do you wrap

things up when it's done? The interface

and the workflow varies quite a bit

between cloud code on the web, cursor

background agents, and codecs on the

web. So, in this video, I'm not just

comparing whether they can complete a

feature in the background. That's table

stakes. I want to show you what it

actually feels like to use each one of

these tools day-to-day because that

workflow experience, that's what

determines whether you'll actually reach

for one of these cloud background agent

tools or just stick with your local

setup. Now, if you're serious about

staying ahead of the curve when it comes

to building with AI, and I mean actually

building, not just following the hype,

then I want you to read my builder

briefing. That's a 5-minute read that I

send every Friday. No fluff, just

practical notes on what's working right

now in AI first development. You can get

yours by going to buildermethods.com and

subscribers are always the first to know

about my next live workshop where we

talk about what's working right now when

building with AI. All right, so to give

these three different uh cloud agent

systems uh a really good go at it, I

want to make an update to my builder

methods uh application and my website uh

specifically on the workshops area. So,

I run these live workshops on building

with AI. I do these about once a month.

And currently, it's just a really simple

index view which just has a list of the

workshops. Sometimes they're upcoming in

the future, sometimes they're in the

past. Uh, and then it links to the uh

the page where, you know, you can either

purchase a ticket or if you're a Builder

Methods Pro member, then uh you get

complimentary access to all past and

future workshops. Uh, but I want to

improve this layout. I want to give them

nice thumbnails here. I want to be able

to filter by which workshops are

upcoming in the future and which ones

are in the past and just have recordings

available. Um, and I've also designed

some new thumbnails that I want to, you

know, display on each of these

workshops, you know, kind of like that

one. Um, and the nice thing is that I've

already designed a section of the site

that has this layout that I'm going for.

So this is our video library uh which

has you know a bunch of videos and um

it's a grid layout or you can toggle to

a list layout like that and the system

will remember your layout and then you

can also filter by topics and here in

this case uh the person viewing is not a

logged in builder methods pro member so

we show this call out that points to

that um but if I do log in and then I go

to library you know we have the same

grid layout here with the list view. Uh,

but instead of the call out, now we have

some filters. So, I can see like which

videos I have recently viewed and I can

see which videos I'm adding to my list

and so on. So, I want to actually reuse

this same layout, the grid and the list

view toggle. And I also want to

establish a topics filter for workshops.

And then I also want to have a filter so

that users can easily see which

workshops are coming in the future and

which ones are uh in the past. So to do

this uh we're going to run this in cloud

code on the web. We're also going to run

a background agent in cursor and I'm

going to use codeex uh on the web. And

I'm going to sort of compare all three

of these and I'll give you my

impressions uh all the way through this

process. Um so before I start to kick it

off, I want to show you the prompt that

I'm going to be using. And as you can

see, it's a pretty long one. Um and it

includes most of what I just described.

And this is what I typically do. I will

draft a prompt in detail. I'm giving it

very specific instruction on which files

and patterns and interfaces and

components that I I've already

established in this interface for

browsing the library. So hopefully the

agents will reuse a lot of that code and

and that interface. Okay, so let's kick

this off. I'm going to start with cloud

code on the web. And what I like about

it off the bat is that this this seems

to be like the closest to the cloud code

CLI experience. It's it it really feels

like uh just a web version of the same

experience. It's very consistent. Okay,

so now first thing you will need to

connect your GitHub repository for your

project and I've already done that for

my builder methods project. Also in

cloud code you can select which model.

Uh, for smaller things, I would probably

go with Haiku, but uh, for this is a

somewhat larger piece, so I'm going to,

um, use the Sonnet 4.5 model. That's

sort of my go-to these days uh, on most

things. I use that and cursors composer

model, which we'll I'll show you in just

a minute. So, I'm going to take my long

prompt and I'm just going to drop it

right in here. And we're going to kick

this off.

Okay, so Claude is getting to work and I

can already see it's updating live on my

iPhone through the uh, Claude mobile

app. Um, so that's nice and and it all

looks really consistent between the two.

It actually seems to be happening in

sync. So I could I could see like if I

ever out or traveling or something and I

want to kick off a new task, like it's

it's all right here. So that's really

cool. Um, so that's working. Now, let's

go ahead and get a cursor agent up and

running. So this one I'm really

interested in using because I use cursor

every day as my IDE. So uh, here I am in

the web interface. Now, Cursor, as far

as I know, I don't believe they have an

actual mobile app the way that Claude

does and ChatgBT does uh with with

Codeex. Um, but cursor.com, I find

actually is totally usable on the mobile

web. It's totally responsive. It's

actually a really good interface. Um, so

I've been pretty happy with it in both

cases. So, I'm going to go ahead and So,

I'm here in the web interface. I'm going

to, you know, I've already started up a

new agent here. Once again, I you know,

you also need to connect your GitHub

repository, which I've already done here

for builder methods. It's the same repo

that I'm working in. This indicates that

we're going to be branching off of the

main branch. Uh I think there was a

similar

You know what? I'm not sure if cloud

code even had that selection. There

might be a setting for it somewhere, but

yeah. So, this is specifying that I am

branching off of main. And then we can

select the model. So, I am going to use

Cursor's Composer 1 model, which is

their new model. I did an entire video

here on the channel last week covering

uh Cursor 2.0 and the new Composer

model, which I'm super impressed with.

It's incredibly fast, and it's also

really accurate, and I find it's almost

as good as set 4.5 in most cases. The

other cool thing about Cursor 2.0, you

know, and you can do this through their

their web background agents is you can

actually run multiple uh different

agents and even run multiple instances

of them, you know, multiply them. We're

not going to do that here. Uh I've used

enough token usage uh in my cursor

account this month, but um I'm going to

stick with curs uh composer one and I'm

going to give it that prompt. And I bet

that even though I'm starting this way

later than the cloud code agent, I bet

this is going to finish even faster

because composer is just lightning fast.

Okay, so composer model using cursor is

off and running. And I can see that here

in the web interface and I'm going to

check it out on mobile by going to

cursor.com.

And there it is. It's already there. Uh

I'll show this on screen. You know, it's

got a it's got a pretty nice mobile

interface that I can easily use using,

you know, like mobile Safari or mobile

Chrome or something like that. Now, what

I'm also interested to see is how does

this work? If I click open in cursor

because I am a cursor user. Cursor is my

IDE. So if I am sitting here in the

office, I'm going to click open in

cursor. And here it is. It's actually

showing me the cloud agent here in the

uh in cursor agent. And I can see down

here it indicates uh this little icon

means that I'm looking at the cloud

agent, not a local agent. And if I go

into my agents list in in agents view

again, I'm seeing that agent here in the

middle and I'm also seeing that agent in

this uh in progress section here in in

cursor. So I really do like that that

integration. I can jump, you know,

straight from the web interface right

into cursor which is right here on my

computer and I can see it all happening

right there. So that's really nice.

Okay, let's go over to codecs. You know,

this is linked right off of my chat GPT

account. Um, once again, you do need to

connect your GitHub repository, but I

found that Codeex sort of tripped me up

a little bit because in addition to

connecting your repo, you also need to

connect what they call an environment.

And all three of these are technically

creating a cloud environment where it's

running your application in the cloud on

their servers, not on your local

machine. Right? Codeex is a little bit

more more explicit about that. And this

is this is what tripped me up because

you can connect I think multiple

repositories into a single environment

and I had accidentally connected my

builder methods repo to a different

environment that I had been working on a

few months ago and that sort of tripped

it up. So I had to both create a repo

like connect my repo and create the

environment in codec. So I did that.

Something to be aware of if if it might

trip you up. So in codeex I can also run

they call them versions. This is like

running multiple agents. So, that's kind

of cool. You can try out different takes

on the same build, similar to what you

can do with cursor. Right now, it is

giving me the option on which branch I

want to branch off of cuz all three of

these are going to create a new branch.

The question is where are we branching

off of? I'm going to branch off of the

main branch. That's fine. One thing that

I'm not seeing, maybe I'm missing it

somewhere, but I can't actually select

the model. I'm assuming that this is

going to be GPT5 and it might be or it

probably is their GPT5 codeex flavored

model. Um, and that's, you know, what is

typically used with, uh, codeex CLI. And

I have a separate video doing a deep

dive on that as well. So, I don't see

the model selector. That's a little bit

strange to me. But anyway, I'm going to

go ahead and uh, copy my prompt into

here. And let's kick this off as well.

So I see my task here at the top of my

list and I can click into it and I can

see uh it is spinning up uh the

environment. This is sort of like you

know a look at at the terminal there.

Let's see if I click view log now. We're

going to sort of like zero in. I think

it's just spinning it up right now. So

it's just getting started. We're going

to check in on the progress as it goes

along. But I'm also going to open up my

chat GPT mobile app and I'm going to go

to codeex. And I do see it right here.

So here here it is at the top and I can

I can see the progress. So all three of

these are working although composer is

already done as and I'm not surprised by

that and we're going to review all of

its work here in just a minute but let's

wait for all three of them to be

finished. It looks like sonnet 4.5 with

cloud code on the web. It's about

halfway through its uh its to-dos. So uh

so yeah I'm going to I'm going to pause

here and I'm going to come right back

when all three agents are done and then

we'll take it from there. Okay. So, it

looks like all three agents are now

complete. Let's take a look at cloud

code first. And so, we can see that all

of the to-dos are finished, and it gave

me the the typical summary for a cloud

code task. And it's all set here in

cursor with uh with composer. So, here's

the finished work there. And then over

in codeex, uh here is the summary here.

I can see that it worked for 11 minutes.

Um so, cloud code probably took about 12

13 minutes. Uh I guess codeex took about

11 minutes and composer I don't have a

time on this but it was done in like 2

minutes something like that. Now let me

just compare what the experience is

after the agent has finished their works

and kind of reviewing the work there.

There are a few slight differences

between the three tools here. So with

cloud code I see here that this is the

branch and I just copied that to the

clipboard and that's really nice because

I can just you know pull that down and

and check it out. And I'm going to do

that here in just a second. Um, it does

have a link here to create a PR, but

Claude Code doesn't automatically create

a PR, which is a little bit weird. I I

kind of wish it would just create it. I

don't know why it doesn't. But one of

the really cool things of Cloud Code on

the web is this button. I can click open

in CLI, and I'm going to click that and

then it's going to copy a command to my

clipboard. So then I can go into I'm

going to go into editor mode in cursor.

And instead of just running claude, I'm

going to paste that in. So here I am

going to run claude, but it's going to

add this flag for teleport and then this

session ID. So I'm going to run that.

And so that is going to start up cloud

code and it's going to pull in the

entire session. All of the agents work.

All the history here is automatically

pulled in, which is really pretty cool.

And then the other thing that I noticed

is that in addition to that, it also

automatically opened or pulled this

branch and checked it out. So the thing

with background agents or really any

agents for that matter is of course we

can review their code. Of course we can

do a pull request and and do a code

review there. But what's always most

important to me is to actually run the

server and check it out in the browser

and check out the user experience of the

product that I'm building. Right? So

here I have it checked out. We're going

to check the work that Cloud Code did uh

in the browser. I'm going to run the

server and I bet there's going to be

some uh migrations that are pending.

Yeah. So, we're going to need to run

those. Okay. So, uh I'm here on my local

machine now and I'm going to go down to

resources. I'm going to go to workshops

and all right. So, here is the uh the

workshops index view. Now, I have not

uploaded my thumbnail images yet. So it

looks like these are like temporary

placeholders when there's not a

thumbnail for the workshop. So that's

pretty good. This is exactly according

to my prompt that I gave it. So this is

really looking good so far. Um I can

toggle between the list view and the and

the grid view and it should remember my

preference there when I refresh the page

and it does. So that's pretty cool. And

then I have this toggle switch. Now, I

actually already have a toggle switch

component, and this looks a little bit

off from that. Functionally, it works,

but it didn't actually use the right

toggle switch. All right, so topics.

Now, I don't I haven't actually created

any workshop topics yet. So, let's go

ahead and check out the admin interface.

This is the workshops area. Okay, so

I've created some uh workshop topics or

workshop tags. And then I can go into

each workshop now. And nice, I can I

have my topic and tag interface here.

So, this interface is a little bit

buggy, but again, it's just for me to

use. Um, and it functionally works. So,

that's that's good enough for me. Okay.

So, now I have some topics in the

system, and I can filter them like that.

Okay, we've got a little error here that

I'll need to get worked out and I can

fix that later. I I just want to move on

with the video here and compare uh the

other two agents work here. So overall,

I'm actually pretty happy with the work

that Claude Code and Sonnet 4.5 did

here. I do have some cleanup, some

errors to fix. Oh, one more thing I

wanted to check actually, and that is

the workshops view when I'm logged out.

And good, it shows this call to action

for builder methods pro. Uh because you

know, if you're a pro member, you get

complimentary access to all the

workshops, whereas if you're not, then

you can, you know, like purchase a

ticket to any individual workshop. So

that's why I wanted to have this uh call

out here. Okay. So let me uh sort of

backtrack and we're going to shift over

to the work that cursor did the cursor

background agent with the composer one

model. So in order to do this since

there was a a database migration uh and

I need to get the database back to its

previous state. Rails DB roll back. This

is just a Ruby on Rails thing here of

course. And now I'm going to go back to

main and we should be back to our

original state. And yeah, we are. So

this is how we started off before the

the agents did any of their work. This

is sort of like the boring uh index view

for workshops. So now let's go to

cursor. I'm looking at the cursor web

interface here. Now cursor unlike cloud

code and unlike codeex it does

automatically immediately create a PR

and I can just click this button and

that's going to open up the pull request

in GitHub and I can view it here. This

comment it looks like it's from me but

it was actually written by the agent

before I merge it in. Um even if I

review the code I always want to check

it in the browser. And by the way, also

in cursor itself, I can open this agent

and it shows me all the history here.

Looks like it's a little bit different

if I go between editor mode and agents

agents mode. So when I'm in agents mode,

I can actually see

um the I don't know what this review

button does. Let's see. Okay, so that

that basically just opens up all of the

um all the files that were changed. Uh,

but this one is also a link to the PR

and I can check out the branch. You

know, you can use cursors interface for

checking out the branch, but actually

I'm going to use my own alias for doing

this. So, I'm just going to grab this

branch name. And I have an alias that

fetches a branch off of GitHub and then

checks it out and switches to that

branch automatically. I guess that

cursor, you know, their interface here

would would probably do the same, but

it's just part of my muscle memory and

my and it's like a workflow that I that

I trust that always works uh the way

that I want it to work. And I think

that's just really important in general

when we're when we're tweaking and we're

trying out new tools in our workflow.

So, I'm going to go ahead and do my

alias for this and I'm going to paste

that branch name in here. And so, okay.

So, now I've just checked it out. We're

good to go. I'm going to run this

server. And again, we're probably going

to run into those uh migrations. But

just to make sure, let's see how

composer one did from cursor. Here we

are. All right. It looks looks pretty

good. So, here are our four um workshops

and it has my toggle UI all all set up.

Uh you know, the nice thing is that I've

already designed this same UI before, so

it just reused a lot of the same logic.

So, it it did that really well. Here is

the toggle switch. And that one actually

works a lot better than the one that

Claude Code created. So that's that's

interesting. I'll go back to workshops

and I will apply some tags.

Interestingly, uh this interface works

much better than the one that Claude

Code did. So that's another it's another

data point that makes me think I I might

keep the composer work instead of the

claude code work on this one. Um so

that's looking good. Uh but overall I'm

actually very happy with the work that

the composer model from cursor did. And

it did it so much faster than than the

other two did. Okay. But for now, let's

let's backtrack again and let's check

out the work from Codeex and GPT5 from

OpenAI. So we're going to do the same

thing again. We're going to Okay, so

let's go over to Codeex. Again, Codeex

does not automatically create a pull

request in GitHub. Uh, in order to do

that, you'll need to click this button,

create PR, and then we can click view

PR. All right, so now I'm just going to

do the same thing. I'm going to grab the

branch name that Codeex had created in

my system, and I'm going to fetch that

and switch to it. I'm going to run my

server. I'm guessing once again, we're

going to have some pending migrations.

We're going to run those, and let's

check out the workshops index view. And,

uh, yeah, this one has a little issue

here. So let's go ahead and have let's

at least get this working, right? Let's

see if composer can uh can fix this

issue. So now we have the composer model

fixing the work of GPT5 with with

codecs. Okay, all fixed up. Okay, so uh

this one looks pretty good as well.

We've got our uh list mode and the grid

mode. This one has the thumbnail. This

one is coming up in the future and the

rest of these are in the past. I would

want to actually look into this time

zone issue here. And then we've got the

upcoming toggle switch, which seems to

be working just fine. So, that's good.

Let's go ahead and check this out. And

this one works just fine. Cool. All

right. Let's uh look at this in another

browser. And it got the uh the call to

action to join Builder Methods Pro.

That's all looking good. And uh here is

the link that points to the the page to

get a ticket. Good. All right. So I

would say like overall I was most

impressed with the cursor uh both the

cursor experience with with running uh

background agents but also uh the work

that the composer model did. They were

all very similar. They all mostly

finished the the feature pretty well.

I'm a bit surprised that Claude Code and

Sonnet 4.5 uh didn't quite hit the mark

on this one even though that's typically

the the best performer of the three. Um

but you know it really is different from

from one feature to the next. It's kind

of an interesting time. So, I hope you

found this comparison helpful as you

look at different tools for running

cloud coding agents. Uh, as you saw in

the video, I kind of like cloud code

since I use that quite a bit. Cloud code

on the web is pretty impressive. But in

my experience right now, I'm sort of

leaning toward the cursor background

agent workflow, especially since I use

cursor every day as my daily driver, as

my IDE. Now, if you haven't seen it yet,

cursor 2.0 0 recently dropped along with

the composer model which I demoed in

this video. I did a deep dive into

cursor 2.0 and the composer model

specifically the two features that have

actually completely changed how I work

day-to-day. So right after you hit

subscribe on this channel I'll see you

over there next. Let's keep building.