OpenAI Expands Codex Beyond Code Editing with New Features in Europe
OpenAI is broadening the capabilities of Codex, its programming assistant, by introducing four new features in Europe. These enhancements include the ability to operate desktop applications on both macOS and Windows, a Chrome extension, a custom memory function, and the new Chronicle feature. Currently, OpenAI claims to have over five million weekly users globally. This European rollout follows earlier launches in other parts of the world and occurs in a competitive landscape where Anthropic announced its own desktop operating capabilities set to launch in 2024.
The updates signify a significant transformation of Codex from a mere programming tool to a versatile work assistant. The new application control feature allows Codex to have its own cursor, enabling it to view screens, click, and enter text within desktop applications on both macOS and Windows. Meanwhile, the Chrome extension makes Codex accessible directly within users’ browsers, where it can read content from websites and web applications, perform actions across multiple tabs, and operate in the background. The custom memory function tracks user preferences, recurring workflows, tech stacks, and deployment conventions, although it is disabled by default in the UK and Europe. Chronicle, currently a research preview available for Pro subscribers, feeds this memory function from content displayed on the screen.
OpenAI reports promising adoption metrics in France, highlighting that the number of weekly active users has surged more than sevenfold since the start of the year. Daily interactions increased by over fifteen times from January to late April, with more than 40% of queries now focused on use cases outside of coding. While these figures come directly from OpenAI and have not been independently verified, they reflect a clear intention to position Codex as a comprehensive workspace tool rather than just a code-writing assistant.
A Competitive Landscape Already Established
The significance of these features lies less in their novelty and more in their timing. Anthropic began offering desktop control in October 2024, with a beta version for macOS and Windows capable of screen monitoring, clicking, and text entry. By 2025, these capabilities evolved to form the foundation of Cowork, a desktop application aimed at non-technical users. In particular, the schedule gap between OpenAI and Anthropic can be measured in months regarding this functionality.
By transforming Codex into an autonomous desktop agent, OpenAI is entering two competitive fronts. The first pits it directly against Anthropic, whose Cowork caters to the same niche of general-purpose agents and became widely available in April 2026. Secondly, OpenAI is facing competition from third-party applications that have relied on its APIs. Hermes, an open agent developed by Nous Research that offers its own desktop application since June 2026, operates without a dedicated model and connects to an external provider to function. By delivering an agent that directly operates the computer, OpenAI circumvents the need for these intermediary layers. This transition of Codex from a coding editor highlights a broader industry trend, revealing OpenAI’s relative slowness in the desktop space, even as it leads in public awareness of generative AI.
Why Europe is Last to Receive These Features
The deployment of features in the UK, the European Economic Area, and Switzerland comes after their initial global availability, but some options remain limited: the memory capability is disabled by default in these regions, and Chronicle is offered only as a restricted research preview. OpenAI has not officially linked this delay to any specific requirement, but the regulatory context provides insight. GDPR regulations govern the storage of personal data, while the AI Act outlines the obligations for AI systems under European law.
An agent that remembers workflows and reconstructs context from the screen inherently handles potentially personal or proprietary data. The default deactivation of memory in Europe reflects a cautious approach to compliance. For IT departments, this functional asymmetry between regions becomes a consideration in choosing tools, as the same software offers different capabilities depending on jurisdiction.
The Shift in the Workspace Landscape
Beyond the agent itself, this announcement indicates a shift in control dynamics. An assistant equipped with its own cursor, accessible within web browsers and bolstered by persistent memory, transcends being just another application; it becomes a foundational layer that manages others. The architecture of AI agent platforms is converging toward a layered framework, with the workspace being the apex—a space that OpenAI is now poised to dominate with Codex.
OpenAI’s placement within this layer creates dependencies familiar to IT departments accustomed to office suites. Since the provider is based in the United States, data processed by the agent falls under the potential purview of the Cloud Act, which allows U.S. authorities to request data held by any provider under their jurisdiction, even if stored abroad. For organizations concerned with sovereignty, the decision-making process revolves not just around the agent’s performance but also about the scope of the data it monitors, retains, and transmits.
Governance Challenges Ahead
This operational autonomy shifts the focus to governance challenges. An agent that can click, type, and navigate in the background acts with the privileges of the logged-in user on business applications and authenticated sessions. Effective governance of AI agents is essential for their deployment, necessitating rights segmentation, decision logging, and traceability of sensitive operations. A desktop agent can more easily evade these controls by operating at the graphical interface level, where application permissions are less stringent than those on API calls.
The 18-month outlook highlights the stakes involved. A layer of agent functionality executing beneath the operating system’s graphical interface is a more straightforward approach than rewriting entire systems or providing integrated solutions. This model integrates seamlessly into existing installations without requiring dedicated hardware, contrasting with the push for AI embedded in silicon pursued by hardware manufacturers. Consequently, the battle lies in creating the orchestrator of agents installed on desktops, and the first player to establish itself as the default entry point stands to dominate user relationships and data. For French IT departments, the arrival of Codex on desktops warrants the urgent development of a usage policy outlining the authorized scope of action, memory management, and data sovereignty considerations for the captured information throughout the day.



