OpenAI amps up Codex Desktop with computer control, in-app browser, image generation and memory

A coder's tool growing teeth

OpenAI has rolled out a major update to its Codex Desktop app, pushing it beyond a pure coding assistant toward a more general productivity agent. The company showed features that let the app operate a user’s computer, run background automations, and include an in-app browser — all intended to let Codex do more without constant hand-holding. It has been reported that one slide in OpenAI’s briefing claimed 80% of the company’s staff use Codex, a telling bit of marketing that suggests the tool is meant to live on more than just developers’ desktops.

What’s new — click, generate, automate

Highlights: the AI can allegedly control apps on macOS (OpenAI’s note: Mac only for now), an in-app browser that understands page elements you click, and a new image-generation agent so automations can produce charts, diagrams or pictures as part of workflows. The click-to-select feature could be a real time-saver: instead of fumbling through a verbal description like “the headline in column two,” you click the element and tell Codex what to do. Neat. Nervy, too. Do you want your editor to be a point-and-click puppet? Some will love it; others will squirm.

Memory, nagging, and long-run work

Codex can now attach automations to conversational threads, pick up context from past interactions, and — according to OpenAI — “wake up automatically to continue on a long-term task, potentially across days or weeks.” It has been reported that the app also gains memory for preferences and corrections so it doesn’t need to be retrained at each launch. There’s even a proactive “nag” feature that proposes work to continue where you left off. Helpful, or just more background AI poking you to finish your to-do list? Depends on whether you enjoy being reminded that you’re behind.

Why this matters (and what still needs testing)

The shift in positioning — from agentic coding tool to broader productivity platform — is notable. Promising features need hands-on testing: OpenAI didn’t demo browser automations or real-world image-quality comparisons, so performance claims remain unproven for now. If the features are reliable, Codex Desktop could speed workflows in meaningful ways. If not, it risks becoming yet another shiny assistant that talks a good game. Either way, the era of AIs that don't just answer questions but actually act on your machine has arrived. Ready or not.

Sources: zdnet.com