OpenAI’s latest tool, Codex, launched May 16, 2025, is changing how computer programmers’ work. Instead of writing every line of code themselves, developers can now type what they want in plain English, and Codex turns those instructions into working computer code.
“There are lots of good AI systems out there,” says Srinivas Narayanan, vice president of engineering at OpenAI. “Competition is clearly there.” This acknowledgment comes as companies like Microsoft, Amazon, and Google offer similar tools.
Here’s how it works: When programmers need help, they type their request into ChatGPT. Codex then works in its own secure space – imagine a virtual workshop where it can safely write and test code without affecting other parts of a computer system. It can handle multiple tasks at once, from writing new code to fixing bugs, typically taking between 1 to 30 minutes to complete each task.
The early results are promising. At tech company Temporal, developers use Codex to speed up their work by letting it handle basic tasks while they focus on more complex problems. Superhuman, another early user, found an unexpected benefit – their product managers can now make simple code changes without always needing a programmer’s help.
But it’s not all smooth sailing. Wayne Chi and Valerie Chen, doctoral students at Carnegie Mellon University who test these systems, point out that competing tools from companies like Anthropic are often preferred by developers. They note that while different tools excel at various tasks, it’s too early to declare a winner.
Similar Posts
The numbers tell an interesting story. When tested, Codex achieved 75% accuracy on the SWE-Bench Verified benchmark when allowing for multiple attempts. This shows both its capabilities and limitations – it’s helpful, but not perfect.
Right now, Codex is available to users who pay for ChatGPT Pro, Team, and Enterprise services. OpenAI plans to add it to other subscription levels soon. While it’s free to use initially, the company will introduce usage limits and pricing in the coming weeks.
Security is a key focus. Codex works in an isolated environment without internet access, similar to working in a sealed room. This setup helps prevent misuse while keeping developers’ code safe.
OpenAI is also upgrading its simpler version, Codex CLI, which programmers can use directly on their computers. This update makes the tool respond faster and includes $5 in free API credits for Plus users and $50 for Pro users who sign in with their ChatGPT account over the next 30 days.
The tool has clear limitations. It can’t work with website designs yet, and sometimes takes longer than traditional coding. “We’re still in the early phase, so we’re going to have to learn a lot,” Narayanan admits.
For software companies, this represents a significant shift. Cisco, one of America’s largest tech companies, is already exploring how Codex can help their engineering teams work faster. At OpenAI itself, engineers use it for routine tasks like organizing code and writing documentation, allowing them to focus on more challenging work.
This development raises important questions about the future of programming. While Codex makes coding more efficient, it’s designed to assist rather than replace human programmers. The tool’s success or failure will likely depend on how well it can consistently deliver accurate code while fitting into existing work practices.
The shift toward AI-assisted coding is happening rapidly across major tech companies. As these systems become more common, the key will be balancing their helpful features with the need for human oversight and creativity in software development.
