Despachei duas tarefas em paralelo no Codex Cloud
Read the Portuguese translation →
Author: Cortes do Rafael Quintanilha · Watch on YouTube
Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.
Original language: English
Original video: Despachei duas tarefas em paralelo no Codex Cloud · Channel: Cortes do Rafael Quintanilha · 05:13 · Watch on YouTube
Topics: AI agents, Task and workflow automation, Software engineering with AI, AI developer tools
Summary
Section titled “Summary”The video demonstrates setting up a Codex Cloud environment connected to a GitHub repository and dispatching tasks to it. The speaker compares Codex with Cursor, praising Codex’s remote-control interface while saying Cursor’s cloud agents are currently ahead. They dispatch two tasks in parallel—one to change a website’s primary color to red and another to yellow—and show that both appear as separate cloud tasks. One task then stalls after making a change, which the speaker attributes to a model limitation, and they ask it to create a PR.
Key points
Section titled “Key points”- A Codex Cloud environment can be created from a connected GitHub repository.
- Environment setup includes variables, installation scripts, and security policies such as internet access and allowed domains.
- The speaker says Codex allows remote control from a phone without first selecting a remote-control mode.
- Codex tasks can be dispatched from the local desktop or from the ChatGPT web interface.
- The speaker considers Cursor’s cloud agents the current leader, while expecting OpenAI to catch up.
- Two cloud tasks are dispatched in parallel to change the website’s primary color to red and yellow.
- The tasks run in separate environments, so they can work on the same area without interfering with each other.
- One task appears to stall after making a change, prompting the speaker to ask it to create a PR.
Chapters
Section titled “Chapters”| Start | Chapter | Summary |
|---|---|---|
| 00:00 | Setting up a Codex Cloud environment | The speaker shows how to create an environment from a connected GitHub repository and configure its variables and security policies. |
| 01:03 | Comparing Codex and Cursor | The speaker compares their cloud and remote-control features, noting Codex’s interface strengths and Cursor’s lead in cloud agents. |
| 02:39 | Dispatching two tasks in parallel | The speaker sends one task to change the site’s primary color to red and asks a personal assistant to dispatch another to change it to yellow. |
| 04:11 | Checking the tasks and handling a stall | Both tasks appear in Codex Cloud, but one stalls after making a change, so the speaker asks it to create a PR. |
Quotes
Section titled “Quotes”you can execute a task from the cloud. — 00:00
the interface in Codex is superior for remote control — 01:35
Cursor is definitely the winner in this area today. — 02:07
I can have two tasks in parallel — 03:10
please just create the PR. — 05:13
Concepts and entities
Section titled “Concepts and entities”Codex, Codex Cloud, Cursor, OpenAI, ChatGPT, GitHub, Pen, PR
Full transcript (original language)
Section titled “Full transcript (original language)”00:00 What I want to show you today is that you can execute a task from the cloud. And here you can see that I already have the program environment set up, and I can click here to create a new one. It will allow me to create an environment based on a repository I already have connected here on GitHub, letting me dispatch tasks to that cloud repository, right? Since I already have the program one selected, I’ll show you how it works in practice. So we have this environment here, and basically,
00:32 you configure its environment variables so that when you initialize it, it has everything it needs, along with security policies—like internet access and which domains to allow, and so on. So there’s the installation, right? The scripts and models are capable of doing this on their own now. If you want to allow internet access, the variables are right here too. Coincidence or not, again, coincidence or not, this is very similar to what we
01:03 have in Cursor. If you come over here to Cursor, look at this. You also have an option to select the computer, and you also have the option to select the cloud, which is basically the same thing—the same environment running on a cloud agent. And you can also select a remote machine here, like in the case of Host Gate. The difference here with Cursor is that they have a slightly worse interface for remote control. You have to start the conversation by selecting remote control, which I find worse. In Codex, you don’t need that.
01:35 You just start a local conversation, and you can already come here on your phone, for example, and control it from there, provided your computer is configured for it. So, the interface in Codex is superior for remote control, but the cloud concept here is exactly the same as the Cursor cloud I’ve already brought to the channel and mentioned many times as arguably the best one out there. That’s why I reiterate that OpenAI may not have reached the level of Cursor’s cloud agents yet, but Cursor is definitely
02:07 the winner in this area today. And I imagine OpenAI will quickly catch up as well. But what’s cool here, guys, is that I can also dispatch the task right from my local computer, you know, I’m here on the desktop, or I can dispatch it from the normal ChatGPT web. So I can come here, look, to Codex, New Chat , and the same environment is here in the cloud as well. The only difference is that here I can only run in the cloud, right? Obviously, I’m in the cloud, I’m not on my local computer here, so I can only dispatch tasks from the cloud. So let’s even do a test. Let
02:39 me switch here, look. GPT6 sun to high. I’ll switch here to fast mode. And then I’ll pass some task here to it. Please change the primary color of the website to red. All right? And then, while it works here, I can go to my personal assistant. Let’s open Pen here and do a test with her too, okay? I want you to dispatch a task in the cloud for me for
03:10 the program AI project. And I want you to change the primary button, the color , the primary color for the site to yellow. Wonderful. So, first we need to see if Pen will correctly dispatch a task to the cloud, but the great advantage is that by running in the cloud, I can have two tasks in parallel , even if they are modifying the same thing, right? Theoretically, both are modifying the button there, but I can make them work in parallel without one interfering with the other, because
03:40 they are running in separate environments, which is excellent, okay? So let’s see if Pen will be able to dispatch this task to the cloud. And we can see here, look, this is the one we dispatched, right, asking to change it to red. And it already said it will start here, look, make AI primary yellow. If we select it here, here is Pen’s request, look. The user explicitly asked for a task in the cloud, okay? So let’s wait to see if this task will appear. And look, guys,
04:11 coming here to my Codex, I can see both tasks, look. Change the color here to red and here change the color to yellow . So this one was dispatched by Pen. All right? OK. Both are running in the cloud. We can see here, look, the little cloud icon. So it has already started working on the other one here. It supposedly finished, but look at that, it finished here but stopped, right? It didn’t do anything else. Let me talk to it. Where is the code? This already seems like a model issue to me,
04:41 right? Because it made the change in the local workspace. And then, OK? What are you planning to do with this? So, look guys, how sometimes the model isn’t ready for the functionality, right? I asked: “Alright, you made the change, what do you expect to do after this, right?” Then it said: “Oh, the next step would be for you to review it in the browser.” Then I said: “How can I review it in the browser if you’re in the cloud?” And so now it’s like, whoops, let me check what I can do here , right? So it’s saying you can’t, and then it stops and does nothing. That’s
05:13 really a model problem that’s a bit annoying, right? Ok, so, please just create the PR. So I’ll ask it to create the PR here while the other task is also running in the cloud, okay? So look how cool, you can dispatch not just from the web, not just locally, but also from your own dot, a task in the cloud. Yeah.