OpenAI, the company behind popular AI model ChatGPT, has taken a bold step in its quest to create more sophisticated artificial intelligence. The company's latest innovation, Codex, is an AI coding agent that can handle tasks such as writing features, fixing bugs, and proposing pull requests.
Codex is built using OpenAI's own GPT-5 Codex model, which operates as a cloud-based software engineering agent. This means that most of the development work on Codex was done by Codex itself, with the company relying on its AI coding agent to improve and expand its capabilities.
According to Alexander Embiricos, product lead for Codex at OpenAI, "the vast majority of Codex is built by Codex." The model runs in sandboxed environments linked to a user's code repository and can execute multiple tasks in parallel. This allows it to work on a wide range of coding projects, from simple fixes to complex feature development.
One of the most impressive examples of Codex's capabilities was its role in developing OpenAI's Sora Android app. According to Embiricos, the development tool allowed the company to create the app in record time - just 18 days for the initial build and a total of 28 days from start to finish.
However, while Codex has shown tremendous promise, it's not without its challenges. A recent study by METR found that experienced open-source developers were actually 19% slower when using AI tools on complex, mature codebases. But Embiricos argues that these results are not representative of the entire coding landscape and that Codex is still in its early stages.
Despite this, there's no denying that Codex has already had a significant impact on OpenAI's development processes. The company's engineers now use Codex regularly, and even the company's own coders have seen their workflows improved by the AI agent.
But what does the future hold for agents like Codex? Embiricos believes that coding is a key area where these models will excel, providing economic value to developers while also scaling through them. He argues that Codex is not meant to replace human developers but rather to amplify their potential.
As the world of AI continues to evolve, it's clear that OpenAI is at the forefront of this revolution. With Codex leading the charge, we can expect to see even more innovative applications of AI in the coding space.
Codex is built using OpenAI's own GPT-5 Codex model, which operates as a cloud-based software engineering agent. This means that most of the development work on Codex was done by Codex itself, with the company relying on its AI coding agent to improve and expand its capabilities.
According to Alexander Embiricos, product lead for Codex at OpenAI, "the vast majority of Codex is built by Codex." The model runs in sandboxed environments linked to a user's code repository and can execute multiple tasks in parallel. This allows it to work on a wide range of coding projects, from simple fixes to complex feature development.
One of the most impressive examples of Codex's capabilities was its role in developing OpenAI's Sora Android app. According to Embiricos, the development tool allowed the company to create the app in record time - just 18 days for the initial build and a total of 28 days from start to finish.
However, while Codex has shown tremendous promise, it's not without its challenges. A recent study by METR found that experienced open-source developers were actually 19% slower when using AI tools on complex, mature codebases. But Embiricos argues that these results are not representative of the entire coding landscape and that Codex is still in its early stages.
Despite this, there's no denying that Codex has already had a significant impact on OpenAI's development processes. The company's engineers now use Codex regularly, and even the company's own coders have seen their workflows improved by the AI agent.
But what does the future hold for agents like Codex? Embiricos believes that coding is a key area where these models will excel, providing economic value to developers while also scaling through them. He argues that Codex is not meant to replace human developers but rather to amplify their potential.
As the world of AI continues to evolve, it's clear that OpenAI is at the forefront of this revolution. With Codex leading the charge, we can expect to see even more innovative applications of AI in the coding space.
it's like having a genius robot that can write features and fix bugs all day long 
but i guess the devs who already know what they're doing might be like "yeah no thanks" 19% slower 
but hey maybe that's just because they're too used to being awesome 
so yeah codex is gonna change the game and make coding way more efficient 
. On one hand, it's amazing that they were able to build something like this from scratch using their own GPT-5 model 
.
. I mean, we're talking about AI-generated code here, which can be pretty brittle if it's not tested thoroughly 
. And what happens when you have to make changes that require human judgment? That's where I think Codex falls short 
.
. I'm excited to see where this technology takes us, but I also want to make sure we're not over-relying on AI to solve our coding problems 
.
.
.
. But hey, if it's working for OpenAI's engineers and coders, that's a different story 
It's def an interesting study. But I think Embiricos makes some legit points about how Codex is meant to amplify human devs' potential not replace them. And that economic value thing? 
.
.
I'm hyped about how OpenAI is pushing the boundaries with this AI agent. 18 days to build an app? That's insane! 
!
. I can already imagine how this tech could help with bug fixes and feature development - game-changer!