OpenAI announced a major update to Codex Desktop on Thursday that transforms it from an agentic coding tool into a full computer-use agent. The app can now run applications in the background, operate its own built-in browser, generate images, and automate multi-step workflows across your desktop. It is available on Mac, with other platforms to follow.

If that sounds like what Anthropic launched with Claude Cowork last month, that is because it is almost exactly what Anthropic launched with Claude Cowork last month.

What Codex Desktop Can Actually Do Now

The headline feature is computer use. Codex Desktop can operate applications on your Mac while you work on other things. It runs automations in the background, meaning the AI is not fighting you for screen control. OpenAI claims 80% of its own staff already use Codex, including non-programmers, which signals how broadly the company sees this tool's audience expanding.

The built-in browser is new and notable. You can click on any element in the browser and tell the AI to modify it, effectively turning web interaction into a point-and-command interface. If the feature works as demonstrated, it eliminates one of the most tedious parts of AI-assisted workflows: trying to describe what you are looking at.

Other additions: image generation within automation pipelines (charts, diagrams, graphics produced as part of a workflow), persistent memory across sessions (the app remembers corrections, preferences, and context from previous use), automations that can be attached to existing conversation threads, and a proactive suggestion system that proposes useful work when you launch the app.

The most ambitious claim: Codex can "wake up automatically to continue on a long-term task, potentially across days or weeks." That is not an assistant. That is an autonomous agent with persistence.

The Competitive Landscape Just Got Crowded

This is not happening in isolation. Anthropic launched Claude Cowork in March with remote computer control, background task execution, and desktop automation. Perplexity shipped a "Personal Computer" feature that can operate a Mac Mini. Google is expanding Gemini's desktop integration. The entire frontier AI industry has converged on the same thesis: the next battleground is not the chatbot. It is the desktop.

SiliconANGLE noted that several of Codex's new features "seem to mimic recent additions to Claude Code," including the remote computer control capability that Anthropic shipped first. The pattern is clear: Anthropic sets the pace on agentic features, and OpenAI follows with broader distribution. OpenAI has the user base (over 400 million weekly active users across ChatGPT). Anthropic has the product vision. The race is about which advantage matters more.

Why This Matters More Than Another Model Release

Model improvements are incremental. Desktop agents are a platform shift. When an AI can run your computer, manage your files, browse on your behalf, and remember what you told it three weeks ago, it stops being a tool and becomes an operating layer. The company that wins this race does not just sell an AI model. It sells the interface between humans and everything digital.

The risk for users is lock-in. If Codex Desktop learns your preferences, remembers your workflow patterns, and accumulates context over weeks of use, switching to a competitor becomes expensive in ways that have nothing to do with price. This is the same playbook that made Google Search and Apple's ecosystem sticky. Except now the switching cost is not your data. It is your AI's understanding of how you work.

OpenAI's briefing had genuine enthusiasm from its developers, but the messaging was, by their own admission, "a little murky." Codex Desktop is still aimed at programmers. But it clearly wants to be for everyone. The question is whether "computer use for everyone" ships as a clean product or a confused one.

Reported via ZDNET briefing, TechCrunch, and SiliconANGLE.