OpenAI released a major Codex update featuring background macOS automation, an in-app browser, persistent memory, and 90+ plugins. With 3M weekly users and nearly half of usage now non-coding, Codex is being repositioned as a general work agent.
OpenAI Repositions Codex as General Work Agent with macOS Automation and Memory
OpenAI has released a significant update to its Codex platform, announced just one hour after Anthropic's release of Claude Opus 4.7. The update marks a strategic pivot for Codex from a developer-focused coding assistant to a comprehensive "workbench for real work" capable of handling tasks before, around, and after writing code.
Key Takeaways
- OpenAI released a major Codex update featuring background macOS automation, an in-app browser, persistent memory, and 90+ plugins.
- With 3M weekly users and nearly half of usage now non-coding, Codex is being repositioned as a general work agent.
What's New: From Code Assistant to Desktop Agent
The update introduces several features that expand Codex's capabilities beyond the IDE:
- Background Computer Use on macOS: Codex can now click and type on a user's Mac autonomously while the user continues other work, enabling true desktop automation.
In-App Browser: A built-in browser allows Codex to navigate the web, gather information, and interact with web applications without leaving the interface.
Image Generation via
gpt-image-1.5: Direct integration of an image generation model expands Codex's creative and documentation capabilities.Persistent Memory: The agent can now maintain context and memory across sessions, a critical feature for long-running workflows.
Long-Running Automations: Support for extended, multi-step automated processes.
Expanded Plugin Ecosystem (90+ plugins): New integrations include Atlassian (Jira, Confluence), CircleCI, and the Microsoft Suite, connecting Codex to core enterprise productivity and development tools.
The Strategic Shift: Codex vs. Claude Code
The most significant revelation is not a feature, but a metric and a framing. OpenAI disclosed that Codex now has 3 million weekly active users, and claims nearly half of its usage is already for non-coding tasks. This data directly informs the product's repositioning.
As noted in the source, this represents a "different bet than Anthropic's." While Anthropic's Opus 4.7 focuses on pushing raw model capability and Claude Code remains specialized for software development, OpenAI's Codex is expanding its product surface area and distribution as a general work agent. The goal is to own the user's entire workflow, not just the coding portion.
Technical Details & Context
This release follows OpenAI's established pattern of leveraging its distribution advantage and integrating capabilities across its product stack. The gpt-image-1.5 integration suggests a tighter coupling of multimodal models within the agent framework. The background macOS automation feature is a direct step toward the vision of fully autonomous computer control agents, a competitive area that also includes startups like Cognition Labs (with its Devin agent) and Magic.dev.
The timing, one hour after Anthropic's announcement, underscores the intense, head-to-head competition between the two AI giants. This follows a pattern of rapid, reactive releases we've tracked throughout 2025 and into 2026.
gentic.news Analysis
This update is a clear escalation in the AI Agent Wars, moving the battlefield from benchmark performance to daily utility and workflow integration. OpenAI is playing to its strengths: a massive installed base (3M weekly users is a formidable starting point) and the ability to rapidly deploy integrated product experiences. The claim that nearly half of Codex usage is non-coding is a powerful justification for this pivot; the market is already using the tool beyond its intended scope, and OpenAI is now formally building for that use case.
This strategic divergence creates two distinct lanes in the race for AI dominance. Anthropic, as highlighted in our coverage of their Constitutional AI papers and the Claude 3.5 Sonnet launch, continues to bet that superior reasoning and safety will win the enterprise. OpenAI is betting that a useful, omnipresent agent that can manipulate your actual desktop environment will create unbeatable lock-in. It's a race of depth versus breadth.
The success of this move hinges on execution. Background macOS automation is a powerful but technically fraught feature, requiring robust security and reliability. If Codex can reliably automate complex, cross-application workflows without breaking, it will have carved out a unique and valuable position. However, if these agentic features prove brittle, it could push users back toward more reliable, single-purpose tools or pure-model interfaces like Claude.
Frequently Asked Questions
What is the new OpenAI Codex update?
The update transforms Codex from a coding assistant into a general work agent. Key new features include the ability to run in the background on macOS to click and type automatically, a built-in web browser, image generation, persistent memory, and over 90 plugins for tools like Jira and Microsoft Office.
How does Codex compare to Claude Code now?
They are now pursuing different strategies. Claude Code, part of Anthropic's ecosystem, remains focused on being a superior AI for software development tasks. OpenAI's Codex is expanding to handle a user's entire workflow, including non-coding tasks like data entry, web research, and document creation via desktop automation.
Is Codex safe to run with background computer control?
This is a critical open question. Granting an AI agent the ability to click and type on your computer autonomously carries significant security and privacy implications. Users should carefully review the permissions and security model implemented by OpenAI before enabling these high-access features.
What does "3M weekly users" mean for the AI market?
It indicates massive adoption of AI developer tools. A user base of 3 million weekly active developers and knowledge workers provides OpenAI with a huge advantage in distribution and real-world testing data, which it can use to refine its agentic models faster than competitors with smaller user networks.
Originally published on gentic.news
Top comments (0)