OpenAI has launched a new agentic coding tool with the potential to make even the most fresh-faced code monkey a software superstar. Codex, which the company announced on Thursday, represents a new sidebar feature inside OpenAI’s flagship product, ChatGPT. While obviously an exciting event in the world of software development, the tool also seems like it has the potential to replace skilled coders altogether.
The company says Codex can accomplish coding tasks, and that, much like the rest of ChatGPT, all the user needs to do is input a prompt and then click a button labeled “Code.” Among other things, Codex can read and edit computer files and also run commands. It can also probe a user’s codebase and answer questions about it. To accomplish this, all the user has to do is input a query and click a button labeled “Ask.” OpenAI says that “task completion” by the program “typically takes between 1 and 30 minutes,” depending on the complexity of the requested task.
OpenAI also notes that Codex’s workflow is 100 percent auditable, and that users can review “its actions through citations of terminal logs and test outputs,” which allows the user to “trace each step taken during task completion.” The coding takes place in an isolated environment, the code from which can then be easily integrated into a local environment or uploaded to GitHub.
Tech companies are increasingly automating their code-writing, something Codex will undoubtedly assist with. Unfortunately, it seems not out of the realm of possibility that such automated tools could spell big trouble for the coder economy. If you can hire a junior coder for half the cost of a senior one, plop Codex in their hands, and get a basically finished software product, why go to the trouble of hiring a skilled engineer?
It would stand to reason that Codex could be used maliciously (if unskilled coders can enjoy this tool, it seems like script kiddies would have a field day), but OpenAI claims that it has built safeguards into the tool to stop that from happening. “Safeguarding against malicious applications of AI-driven software engineering, such as malware development, is increasingly critical,” the company writes, explaining that Codex was engineered to somehow “identify and precisely refuse requests aimed at development of malicious software, while clearly distinguishing and supporting legitimate tasks.” It’s not exactly clear what those safeguards are. Gizmodo reached out to OpenAI for more information.
Codex is currently enjoying a research preview and will initially be available for subscription-tier ChatGPT users. OpenAI says that ChatGPT Pro, Enterprise, and Team users will all have free access to the tool, starting today. Eventually, access to the tool may be expanded. While Codex is currently available to subscribers at no additional cost, the company says that “rate-limited access and flexible pricing options” for additional usage will be rolled out at a later date.