Codex evolving into desktop agent

OpenAI is trying to own the desktop work surface, not just the coding prompt. By giving Codex a cursor, an in app browser, image generation, and memory, OpenAI turns a developer tool into software that can click through apps, inspect webpages, draft visual assets, and keep working across repeated tasks. That matters because the winning product is no longer just the best model for writing code, it is the agent that can finish surrounding work inside the same interface.1 2 3 4 5

• The product move is explicit. In April 2026, OpenAI said Codex could operate the computer by seeing, clicking, and typing with its own cursor, plus use everyday apps, generate images, and browse inside the desktop app. That is a much wider job than patching files in a terminal.1 4
• This also changes the competitive frame. Anthropic introduced computer use as an API capability in October 2024, then pushed Claude Cowork as a desktop product that runs inside spreadsheet, word processing, and presentation tools. OpenAI is matching that shift from coding assistant to general knowledge work agent.6 7
• The built in browser shows how the workflow expands. OpenAI documents that ChatGPT desktop can open pages, handle sign in, move across tabs, download files, and let the model and user see the same page. For frontend work, research, QA, and ops tasks, browser control pulls web steps into the same loop as code edits.4 5

The next phase is a fight over habitual desktop time. As Codex adds memory, browser state, and tighter control of local apps, the product can move from helping with isolated coding tasks to becoming the place where technical and adjacent work gets planned, executed, and checked. That pushes the market toward full desktop agents bundled with model access and existing user distribution.1 2 3 5