AI-driven development has reshaped the way code is written. Much of the routine work once handled manually is now delegated to networks of autonomous agents. As developers continue to explore new ways of working alongside these systems, tools must evolve quickly, a challenge even the most advanced labs are struggling to meet.
Agentic development, where autonomous assistants handle end-to-end tasks, is gaining traction through platforms such as the Claude Code and Cowork apps. OpenAI has been building steadily in this space, rolling out its Codex tool last year first as a command-line interface, then as a web version. But the pace of competition has intensified, making this moment critical.
Codex steps up with a macOS App
OpenAI’s latest move is a dedicated macOS app for OpenAI Codex, bringing many of the agentic techniques that have surged in popularity. The app can orchestrate multiple agents simultaneously, combining their skills across different workflows. It arrives only weeks after the release of GPT-5.2-Codex, positioned as OpenAI’s most capable coding model to date and a potential magnet for users of rival tools.
Benchmark results, however, tell a more nuanced story. GPT-5.2 currently leads TerminalBench, which evaluates command-line performance, but competitors such as Claude Opus and Gemini 3 have posted scores close enough to fall within experimental margins. Tests like SWE-bench, which assess real-world bug fixing, reveal no clear dominance among top models.
What remains difficult to measure is user experience, an area where agent-based tools often differ widely, regardless of benchmark parity.
Features built for real-world workflows
The new Codex app introduces capabilities designed to surpass competitors. Automations can now run in the background on a schedule, queuing results for later review. Users can choose from different agent personalities, creating a working style that suits their pace and preference.
The desktop interface is more approachable than earlier Codex versions, making it easier for newcomers, not just engineers, to build and manage agents. This aligns closely with the direction set by Claude Cowork, launched in January, which similarly helped non-technical users deploy coding assistants across their existing software.
The app also includes a library of skills used widely within OpenAI, including:
- Implement designs: Convert Figma files into production-ready UI code.
- Manage projects: Triage bugs and track releases with Linear.
- Deploy to cloud hosts: Ship apps to Cloudflare, Netlify, Render, and Vercel.
- Generate images: Create or modify visuals using GPT Image.
- Build with OpenAI APIs: Pull updated documentation directly into projects.
- Create documents: Produce PDFs, spreadsheets, and DOCX files with clean formatting.
Codex has already proven its value. Independent developer Peter Steinberger, creator of the viral OpenClaw tool and its spinoff Moltbook, says his productivity doubled after switching to Codex. A four-person OpenAI team also built the Sora for Android app in just 28 days using Codex.
Availability and pricing
The Codex app is now available on macOS. Anyone with a ChatGPT Plus, Pro, Business, Enterprise, or Edu subscription can access Codex across the CLI, web, IDE extension, and desktop app. Usage is included in existing plans, with optional credits available for heavier workloads.
For a limited time, Codex is also open to ChatGPT Free and Go users. During this period, OpenAI is doubling the rate limits for all paid users.
Competition intensifies across the ecosystem
With over a million developers using Codex in the past month across startups and enterprises, OpenAI is tightening its grip on the developer tools market. But competition is heating up. Anthropic has made rapid gains in enterprise adoption, with companies like Uber, Spotify, Salesforce, Accenture, and Snowflake adopting Claude Code.
OpenAI is betting on centralised control as the next big differentiator. The app lets users coordinate multiple agents across projects, automate repetitive workflows, and monitor everything in one place. For now, the company recommends using GPT-5.2-Codex for coding tasks and GPT-5.2 for analysis and writing, with adjustable intensity and personality settings.
The battle to become developers’ default platform is far from settled, but with the Codex app, OpenAI has signalled its intent to compete on experience, depth, and speed.
What’s next for Codex?
Codex adoption has surged since the launch of GPT-5.2-Codex. Next, OpenAI plans to bring the app to Windows, further expand model capabilities, and speed up inference.
Multi-agent workflows will continue to improve, offering smoother task switching and better parallel execution. Automations will soon support cloud-based triggers, letting agents run continuously even when a device is offline.
OpenAI’s core belief is that everything comes back to code, and the more capable an agent is at producing and reasoning about it, the more broadly useful it becomes. Codex is designed to close the gap between powerful models and the practical, everyday work people need to accomplish.
OpenAI’s latest move signals one thing clearly: the new era of software development will be faster and more autonomous, more intuitive, and open to far more people than ever before.
