OpenAI upgraded Codex with desktop computer use, image generation via gpt-image-1.5, parallel agents, and memory — a direct counter to Claude Code's dominance.
OpenAI announced a major Codex update adding desktop app control (macOS first), background multi-agent execution, image generation via gpt-image-1.5, new tool plug-ins, and opt-in memory for personalization. The rollout starts today for ChatGPT-connected Codex desktop users, with EU users and other OS support delayed. The update is a clear strategic response to Anthropic's Claude Code gaining significant developer mindshare. Memory features are launching as a preview for Enterprise and Edu users.
Codex now competes directly with Claude Code's agentic capabilities: parallel background agents, computer use on macOS, and memory that survives sessions. The computer use feature is specifically useful for frontend testing and interacting with tools that lack APIs — a gap Claude Code still has. Multi-agent parallelism means you can queue up multiple async coding tasks without babysitting a terminal.
Spin up the Codex desktop app on macOS this week and run a frontend testing task — specifically something that previously required Playwright or manual clicking — to benchmark it against your Claude Code workflow on the same task.
Suggested exercise · not independently tested. Access requirements and costs may vary.
Open the Codex desktop app and sign in with your ChatGPT account on macOS
Create a new task with this prompt: 'Open my local React app at localhost:3000, click through the signup flow, identify any broken UI states, and give me a list of issues with screenshots'
Let it run in the background while you continue working, then review the output report
Tags
Related