At a time when artificial intelligence is in the news for writing a growing pool of codes, OpenAI is rolling out a new AI agent for ChatGPT that’s designed to streamline software development. Microsoft CEO Satya Nadella recently said that as much as 30 per cent of his company’s code is being written by AI.
The Cloud-based software engineering agent, called Codex, is meant for professional coders. It will let them automate more of their work while keeping the process less opaque than many existing tools.
Codex is in the early stages and has limited functionality. It will first be released as a “research preview” to paid ChatGPT Pro, Enterprise and Team users.
OpenAI joins a growing number of tech companies — Microsoft-owned Github, Alphabet’s Google and Anthropic — that offer AI tools for programmers. Startups like Cursor maker Anysphere and Windsurf are also attracting investments because of their AI-driven coding assistants that can scan a software developer’s actions and suggest the next few lines.
OpenAI explains the route Codex will take. Accessible through the sidebar in ChatGPT, you can assign it new coding tasks by keying in a prompt and then clicking “Code”. If a question needs to be asked about a codebase, click “Ask”. The tasks are processed independently in a “separate, isolated environment preloaded with your codebase”.
The company said Codex can “read and edit files, as well as run commands including test harnesses, linters, and type checkers”. Task completion typically takes between one and 30 minutes, depending on complexity.
OpenAI’s technical staff uses the coding agent daily, from repetitive tasks to helping build new features. Companies like Cisco Systems and Kodiak Robotics, are also using the tool, OpenAI said. Temporal uses Codex to “accelerate feature development, debug issues, write and execute tests, and refactor large codebases”.
Meanwhile, OpenAI is in talks to acquire Windsurf, another AI-powered programming tool, for around $3 billion, according to The New York Times. With the deal, OpenAI wants to go beyond its popular chatbot, ChatGPT.
OpenAI has launched two other agentic AI tools in the last year: Operator, which controls a web browser and can automate online chores, and Deep Research, which does detailed web searches to compile reports.
AI coding agents, like generative AI systems, can make mistakes. A recent study from Microsoft found that leading AI coding models, such as Claude 3.7 Sonnet and o3-mini, struggled to reliably debug software.
