Another step toward automating programming?
On Friday, May 10, 2025, OpenAI announced a preview of Codex, a new conversational agent designed for software engineering, integrated into ChatGPT1. This model, not to be confused with the first version of Codex launched in 2021, is capable of assisting developers by generating code, fixing bugs, writing unit tests, and even creating complete pull requests. Following in the footsteps of Operator (web browsing) and Deep Research (information synthesis), Codex aims to transform programming into a dialogue guided by artificial intelligence. Some companies, such as Stripe and Notion, have already begun testing its capabilities in real-world environments2.
What Codex Can Already Do
The Codex agent acts as a software co-pilot capable of working iteratively with the user. It can handle complex environments, understand existing software architectures, suggest corrections, and adhere to coding standards. Thanks to its native integration with ChatGPT, users can interact with the agent using natural language, minimizing technical friction. It is also capable of contextualizing its responses based on a project’s dependencies, which enhances its relevance in production environments. Several experts are already praising its ability to generate structured code from simple prompts.
A strategic tool for businesses
Codex is poised to become a major productivity driver for tech companies. By shortening development cycles, it could save up to 30% of development time for certain teams, according to initial internal feedback from OpenAI. This automation could free senior developers from repetitive tasks, allowing them to focus on architecture and software innovation. As part of a DevOps strategy, Codex could also streamline coordination between the development and continuous integration phases. Use cases in pilot companies, such as Stripe and Notion, are currently being evaluated.
Rapid development—but at what ethical cost?
While Codex promises to accelerate software development, it also raises ethical questions. Who is liable in the event of an error in generated code? The user, the company, or the agent itself? Furthermore, code transparency poses a problem: how can one audit logic generated probabilistically? Codex could also reproduce biases or incorporate code protected by licenses, due to a lack of rigorous filtering of training data. In a regulatory environment still under development, these gray areas require heightened vigilance from both developers and decision-makers3. These ethical issues tie into broader debates on algorithmic accountability in production environments.
Codex and the development professions: transformation or redefinition?
The emergence of tools like Codex calls for a rethinking of developer training and practice. Key skills are gradually shifting: it is no longer just about knowing how to write code, but also about evaluating, correcting, and managing a generative agent. This transformation could reposition developers as AI supervisors, responsible for code quality, compliance, and maintainability. In this context, mastery of prompts and model architectures is becoming a new technical skill in its own right.
What are the implications for education and software engineering?
Software engineering could evolve toward a more collaborative and dialogic approach, where developers become AI “orchestrators,” capable of designing systems in cooperation with models. Codex could thus herald the era of augmented development, where execution speed is combined with conversational intelligence. This hybrid model will require new working methodologies and quality standards adapted to the involvement of autonomous agents. It also involves equipping developers to detect and correct any potential deviations in code generators.
References
References
1. TechCrunch. (2025). OpenAI’s Codex agent wants to be your pair programmer.
https://www.ccomptes.fr/fr/publications
2. VentureBeat. (2025). Stripe and Notion experiment with OpenAI Codex to accelerate software development.
https://www.numerique.gouv.fr
3. Ministry of the Economy. (2024). France 2030: AI Strategy for Public Services.
https://www.economie.gouv.fr/france2030