Three million developers use Codex every week. Half of them aren’t even writing code. That stat, buried in OpenAI’s latest announcement, tells you more about where this product is headed than any feature list.
On April 16, OpenAI shipped a sweeping update to its Codex desktop app. The headline: Codex can now control your Mac, browse the web, generate images, remember your preferences, and schedule its own follow-up tasks. It’s no longer a coding assistant. It’s a full-blown work agent — or at least, that’s the pitch.
The biggest addition is background computer use. Codex gets its own cursor and can see, click, and type across any Mac app without stealing focus from what you’re doing. Need it to test a frontend change, fill out a form in a tool that has no API, or run through a user flow? It handles that in the background while you keep working. Mac only for now, with EU and UK access coming later.
The app also gained an in-app browser pulled from OpenAI’s Atlas project, letting developers comment directly on web pages to give precise feedback during frontend iteration. Over 90 new plugins landed too, bundling skills, app integrations, and MCP server connections. Atlassian Rovo, CircleCI, CodeRabbit, GitLab Issues, Microsoft Suite, Neon by Databricks, and Render are among the highlights. One demo showed Codex checking Slack, Gmail, Google Calendar, and Notion simultaneously to produce a morning priority list.
Image generation via gpt-image-1.5 is baked in, so mockups, product concepts, and game assets can be created without leaving the workspace. Automations now persist across conversation threads and can span days or weeks — Codex schedules its own wake-ups to continue long-running tasks. A memory preview learns corrections, preferences, and gathered context over time.
For developers specifically, new features include GitHub PR review inside the app, multiple terminal tabs, a file sidebar with rich previews for PDFs and spreadsheets, a summary pane tracking agent plans and artifacts, and alpha SSH support for remote devboxes.
Here’s the honest question: how much of this is actually new territory?
Anthropic’s Claude Code already does terminal-based agentic coding with tool use and MCP servers. As MacStories noted, the plugin and automation model closely mirrors what Anthropic ships with Claude’s Cowork mode. Cursor has been doing multi-file AI-assisted editing for over a year. GitHub Copilot owns the inline-suggestion market. Computer use? Anthropic demoed that in late 2024 and has been shipping it in Claude’s desktop app.
What OpenAI does have is scale. Three million weekly users, growing by a million per month, with a pricing structure that ranges from free (now ad-supported) through an $8 Go plan, $20 Plus, and $100/$200 Pro tiers. The $100 Pro plan launched in early April specifically for heavier Codex usage. And the $122 billion in financing OpenAI raised just days before this announcement gives them the runway to subsidize aggressive growth.
The real signal here isn’t any single feature. It’s the superapp strategy OpenAI confirmed in March: merging ChatGPT, Codex, and the Atlas browser into one unified desktop environment. Codex head Thibault Sottiaux said openly that this release targets developers first, with a broader audience coming next. The playbook is clear — hook power users with coding tools, then expand the same agent workspace to everyone who sits at a computer for work.
Whether one app can serve both audiences without becoming bloated is an open question. But with half its user base already ignoring the “code” in Codex, OpenAI might not have a choice.
Sources
This article is AI-generated.
