OpenAI Boosts Codex, Targeting Anthropic with New Desktop‑Control Power

OpenAI’s Codex upgrade introduces a background‑execution model that lets the AI run autonomous agents on a Mac without monopolising the user’s active session. According to the company’s blog, the agents can open any installed application, move the cursor, and type commands while the user continues to work in other windows. This “parallel‑in‑background” capability is framed as a “coding buddy” that can perform auxiliary tasks—such as iterating on frontend UI tweaks, testing applications, or interacting with software that lacks a public API—while the developer focuses on higher‑level work. The design hinges on a lightweight daemon that monitors system idle time and injects input events, allowing the AI to operate discreetly and resume control when the user is not actively interacting with the same screen region.

The new feature mirrors functionality recently added to Anthropic’s Claude Code, which TechCrunch noted can “remotely control your Mac and desktop on a user’s behalf while they were away from their keyboard.” OpenAI’s announcement therefore represents a direct response to Anthropic’s earlier agentic push. By enabling Codex to issue mouse clicks and keystrokes, OpenAI hopes to close the gap in “agentic assistance” that has so far given Claude Code a competitive edge in enterprise deployments. The company’s blog emphasizes that the agents are sandboxed to avoid interference with the user’s primary workflow, but the underlying mechanism—system‑level input injection—remains identical to the approach Anthropic demonstrated last month.

In addition to desktop control, Codex now ships with an in‑app browser that can be commanded via natural‑language prompts. The browser interface lets developers ask the AI to navigate to a specific web page, extract data, or interact with web‑based development tools, then feed the results back into the coding environment. OpenAI positions this as a boon for frontend and game development, where developers often need to toggle between local servers, documentation sites, and asset pipelines. The roadmap, as outlined by the blog, includes plans to “fully command the browser beyond web applications on localhost,” suggesting future support for remote services and possibly headless automation.

A preview feature called “memory” adds session persistence to Codex’s workflow. According to TechCrunch, the memory module records prior interactions and synthesises contextual cues about a user’s coding style, enabling the AI to generate more relevant suggestions in subsequent sessions. This persistent state is intended to reduce repetitive prompting and accelerate iterative development cycles. Coupled with the new image‑generation capability—described as a tool for creating product concepts, slide visuals, mockups, and placeholder graphics—Codex is being positioned as a multi‑modal assistant that can handle both code and visual artefacts within a single workflow.

OpenAI’s broader strategy appears to be a push toward a more integrated, agency‑rich development environment. By combining background desktop agents, browser automation, persistent memory, and image synthesis, Codex aims to become a “multifaceted tool” that can be woven into corporate pipelines ranging from UI prototyping to automated testing. The move underscores the intensifying rivalry with Anthropic, where each firm is rapidly expanding the operational envelope of its coding assistants. As TechCrunch points out, the upgrades “demonstrate OpenAI’s desire to not only make Codex a competitive coding assistant but also a more multifaceted tool that can be integrated into a variety of corporate workflows,” signaling that the next battleground will be not just model performance but the breadth of system‑level capabilities each platform can safely expose.