OpenAI’s Codex for (Almost) Everything expands agent powers — and grabs the mouse

April 16, 2026
Hand operates a computer mouse beside a white keyboard on a dark desk, showcasing modern technology use.
Photo by Maria Stewart on Pexels

It has been reported that OpenAI released a major update to Codex that it says will make the assistant a broader partner for developers, allegedly reaching more than 3 million weekly users. The headline features are striking: Codex can now operate your Mac alongside you, interact with apps by seeing, clicking and typing with its own cursor, generate images with gpt-image-1.5, and remember preferences across sessions. Exciting? Absolutely. A little unnerving? You bet. Ready to hand over part of your workflow to a digital sous-chef?

What’s new — browser, images, and plugins

The app adds an in‑app browser (with page annotations), native image generation and iteration, and over 90 new plugins tying Codex into tools like Jira, CircleCI, GitLab Issues, Microsoft Suite, Neon, Render, and more. Multiple agents can run in parallel on a single machine without clobbering your active work, and screenshots plus code can be combined to produce frontend mockups and game assets inside the same flow. It’s the natural next step in the “agentification” trend: give AIs more context, let them act, and watch bottlenecks melt away.

Dev workflows and memory

Codex now supports GitHub review comments, multiple terminal tabs, remote devboxes via SSH (alpha), rich file previews in a sidebar, and a summary pane to track agent plans and artifacts — essentially stitching the entire dev lifecycle into one workspace. It has been reported that automations can schedule future work and that a memory preview will allow Codex to retain context, corrections and preferences to speed future tasks. The payoff is clear: faster iterations, fewer context switches, and a system that can suggest exactly where to pick up a stalled project.

Risks, reactions, and the big question

All of this sounds like productivity on steroids, but it raises real questions: who controls the cursor, what gets stored in memory, and how auditable are the agent’s actions? OpenAI’s blog lays out capabilities and plugins, but it has been reported that details on permissions, security controls, and enterprise audit logs will be important for adoption — especially in regulated teams. Will developers treat Codex like a teammate or like a faster, more opinionated tool? Either way, the update makes one thing clear: the future of coding looks less like typing in solitude and more like orchestrating a small army of helpful — and sometimes cheeky — assistants.

Sources: openai.com, Hacker News