The competitive landscape in large language models has intensified.
Since the arrival of OpenAI's ChatGPT in late 2022, which brought conversational AI into widespread use and ignited what many refer to as the LLM wars. OpenAI initially gained prominence with specialized tools for developers, including the original Codex model released in 2021 that powered early versions of GitHub Copilot.
That early Codex focused primarily on translating natural language into code snippets, and later, as an autonomous, cloud-based software engineering agent. It was later phased out in favor of more general-purpose models, but the name has now been revived for a far more ambitious agentic system.
This is because OpenAI has announced a significant update to its Codex agent, transforming it from a coding assistant into a more comprehensive computer collaborator.
The new version introduces background computer use on macOS, allowing Codex to control its own cursor to see, click, and type within applications, even those without APIs.
Codex now helps with more of your work, from coding to staying on top of everything around it. https://t.co/TCaOJhdj1q
— OpenAI Developers (@OpenAIDevs) April 16, 2026
Multiple agents can operate in parallel on the same machine without interfering with the user's own work, making it possible to run background tasks like testing apps or iterating on frontend designs while the developer continues other activities.
The update adds an in-app browser that lets users interact directly with web pages by commenting on rendered content, providing precise instructions for changes.
This is particularly useful for frontend development and local applications running on localhost, with plans to expand browser control further. Image generation has been integrated through OpenAI’s models, enabling Codex to create and iterate on visuals such as mockups, product concepts, or game assets within the same workflow.
So excited to share that we're bringing Computer Use to Codex. Computer Use lets Codex see, click, and type into your Mac apps, with its own cursor.
It's a magical feeling to have agents using your apps in the background, and still get to use your computer at the same time. pic.twitter.com/wdgxiHAKyX— Ari Weinstein (@AriX) April 16, 2026
Additionally, more than 90 new plugins expand its reach, connecting to tools like JIRA, GitLab, Slack, CircleCI, and various other development and productivity platforms.
These enhancements allow Codex to support a wider range of activities across the full software development lifecycle.
It can now address GitHub review comments, manage multiple terminal sessions, connect to remote development environments via SSH, and provide rich previews for different file types including PDFs, spreadsheets, and documents.
A new summary pane helps track agent plans, sources, and generated artifacts. Memory features enable the system to retain user preferences, past corrections, and project-specific context, while expanded automations support long-running tasks that can pause and resume over days or weeks.
Background Computer Use
Computer Use in Codex has some deep OS-level wizardry. Codex can see/click/type in apps in the background, without taking over your computer, and you can work in parallel.@AriX and team absolutely crushed here. Windows soon. https://t.co/kajg3QoSp3— Alexander Embiricos (@embirico) April 16, 2026
This latest release follows closely on the heels of Anthropic's Claude Opus 4.7, announced around the same time.
Claude Opus 4.7 brought improvements in agentic coding, long-horizon reasoning, high-resolution vision capabilities, and more thorough problem-solving, intensifying the ongoing competition between the two companies in building advanced AI agents.
OpenAI's renewed Codex builds on its initial launch as a cloud-based software engineering agent in May 2025, evolving it into a desktop-integrated workspace that blurs the lines between code generation, application control, research, and project management.
The features are rolling out first to users of the Codex desktop app signed in with ChatGPT accounts. Some personalization elements, including memory and context-aware suggestions, are expected to reach enterprise, education, and additional regional users in the coming weeks.
Computer-use capabilities are initially limited to macOS, with broader platform support planned.
Codex can also proactively suggest next steps based on open issues, connected tools, and accumulated knowledge from previous interactions.
You can now generate and iterate on images with gpt-image-1.5 in Codex to create frontend designs, mockups, game assets, and more without leaving your workflow.
Usage is included with your ChatGPT account, no API key needed. pic.twitter.com/ay17I3Nxoa— OpenAI (@OpenAI) April 16, 2026
Since its reintroduction, Codex has seen steady growth, reaching millions of weekly active developers who increasingly rely on it not just for writing code but for understanding systems, debugging, reviewing work, and coordinating longer-term projects.
This development highlights a broader shift in AI tools for software engineering.
What began as code-completion engines has progressed toward autonomous agents capable of handling complex, multi-step workflows that span local machines, cloud environments, and external applications. As both OpenAI and Anthropic continue to push these boundaries in quick succession, the distinction between specialized coding assistants and general-purpose AI collaborators is becoming less defined, pointing toward more integrated systems that aim to support the entire process of building and maintaining software.
Despite these impressive advances, Codex and similar agentic systems still face important limitations.
We’ve also added support for 90+ plugins in Codex, giving it more ways to gather context and take action across the tools you already use for docs, project management, code review, creative work, deployments, and more. pic.twitter.com/IkmpDJwrLq
— OpenAI (@OpenAI) April 16, 2026
Reliability can be inconsistent, with the AI occasionally producing subtle bugs, misinterpreting requirements, or struggling with large codebases and legacy systems. Security and privacy concerns are significant, as granting deep control over the user’s machine, browser, and files introduces risks around data exposure and unintended actions. High subscription costs, usage limits, macOS-only availability for full computer control, and occasional performance slowdowns further restrict accessibility.
There is also a growing concern that overreliance on such tools may gradually erode developers’ core skills in debugging, system design, and critical thinking.
Nevertheless, the broader positive implication is profound: these evolving AI agents are steadily transforming software engineering from a solitary, labor-intensive craft into a highly collaborative and accelerated discipline.
By handling repetitive tasks, maintaining context across long projects, and enabling faster iteration, tools like the new Codex have the potential to dramatically increase developer productivity, shorten development cycles, and allow humans to focus on higher-level creativity, architecture, and innovation, and ultimately raising the overall quality and ambition of what software teams can achieve.
Just before this, OpenAI released GPT-5.4-Cyber after Anthropic launched a secretive Project Glasswing after knowing its Mythos AI model is powerful in cybersecurity.