OpenAI Launches New Coding Application for macOS

AI-Powered Software Development Enters a New Phase

Artificial intelligence has already initiated a fundamental shift in how software is written. The routine workload of programming is now largely handled by AI agents and sub-agents. With developers experimenting with new interfaces and form factors for human-AI collaboration, even the most advanced AI labs are struggling to keep up with this pace.

Codex Lands on the Desktop

The new macOS application announced this past Monday significantly enhances OpenAI's Codex tool. Codex began as a command-line tool last April, gaining a web interface a month later. The new application incorporates the agent-based software development practices that have become popular over the past year.

Agent-based software development refers to systems where AI agents can work independently on coding tasks. Applications like Claude Code and Cowork stand out in this field. With its new app, OpenAI aims to take a significant step in this competitive arena.

Multi-Agent Support and New Features

The new Codex application is designed to work in parallel with multiple agents, integrate agent capabilities, and utilize the latest workflows. The app allows automations that can be configured to run in the background on an automatic schedule, with results placed in a queue for user review upon their return.

Users can also select different 'personalities' for the agent, ranging from pragmatic to empathetic, depending on their working style. The biggest selling point for the company, however, is the development speed enabled by AI. OpenAI CEO Sam Altman stated at a press conference, "You can use this to go from scratch, from a completely blank page, to a fairly sophisticated piece of software in a few hours."

Model Power and Competition

This launch also comes less than two months after the release of GPT-5.2-Codex, OpenAI's most powerful coding model. The company hopes this model will be sufficient to attract Claude Code users. Altman said, "If you want to do sophisticated work on something really complex, 5.2 is the most powerful model to date."

However, coding benchmarks paint a more complex picture. GPT-5.2 holds the top spot on TerminalBench (a test measuring how well AI handles command-line programming tasks). But agents from Gemini 3 and Claude Opus recorded roughly equivalent scores. Results from another coding benchmark, SWE-bench, which tests the ability to fix real-world software bugs, similarly do not show a clear advantage for GPT-5.2.

Experts note that effectively benchmarking agent-based use cases is difficult, and the latest models can differ significantly in user experience. This development, along with intense competition in the AI hardware space, is interpreted as an indicator of how rapidly the software development ecosystem is evolving.

OpenAI's move signals that AI will continue to play a central role in software developers' toolkits and will set new standards in development efficiency.