David Gewirtz/ZDNET
Follow ZDNET: add us as a preferred source on Google.
**ZDNET’s Key Takeaways**
OpenAI has introduced a new generation of agent-driven tools while simultaneously removing some of the most practical, day-to-day functions users relied upon. Although ChatGPT Work—an enterprise-oriented version of the AI assistant—has finally launched, the experience of using ChatGPT via its desktop app has deteriorated markedly. For the majority of users, the web-based ChatGPT remains the most reliable and complete way to access the service.
Initially, my intention in writing this article was straightforward: I wanted to produce a detailed comparison between Anthropic’s *Claude Cowork* and the newly debuted *ChatGPT Work*. However, the path toward that simple goal took an unexpected detour. The process of onboarding and configuring ChatGPT Work turned out to be far more complicated than anticipated, transforming what should have been a routine evaluation into a full-fledged experiment punctuated by frustration and discovery.
My daily computing setup revolves around a Mac, and ChatGPT has long been an essential companion on this platform. Mac users generally have two principal methods for engaging with ChatGPT: directly within a web browser or through the standalone macOS application. Each approach offers distinct strengths. The browser version, for instance, provides advanced capabilities such as creating and customizing GPTs, while the native desktop app focuses on executing them. Both have historically complemented one another perfectly, creating a seamless workflow.
*(Disclosure: Ziff Davis, ZDNET’s parent company, filed a lawsuit in April 2025 against OpenAI, alleging copyright infringement in the training and operation of its AI systems.)*
Until very recently, my preference leaned heavily toward the desktop version, primarily because of its extraordinary convenience when juggling dozens of active browser tabs. One of its most unheralded yet incredibly useful capabilities was the built-in screenshot tool. With it, I could capture any portion of my screen and instantly drop that image directly into a chat window. This process bypassed the need to open third-party screenshot utilities, dramatically speeding up troubleshooting sessions and documentation work. When working through technical issues—say, a misbehaving system process—I could scarcely imagine operating without it. Rapidly capturing and transmitting screenshots to ChatGPT became an almost rhythmic part of problem-solving. Unfortunately, that elegant feature has now vanished.
But the losses do not stop there. Another intuitive function that once distinguished the desktop app from its browser counterpart was the *Work With* option. This feature allowed ChatGPT to interpret and operate within the context of open documents or apps such as Notes, Notion, or TextEdit. With a simple click of the *Work With* button, the AI could analyze whatever appeared in the current window, instantly providing suggestions, edits, or summaries based on real-time content visibility. It transformed ChatGPT into an intelligent, always-available assistant aware of your workspace. Now that tool, too, has disappeared—taking with it a core piece of what made the app so effective for multitasking professionals.
In practice, it would be fair to say that the former ChatGPT desktop environment has been obliterated rather than updated. What has replaced it is, in essence, a rebranded and visually downgraded version of Codex—the developer-oriented programming tool previously integrated with Visual Studio Code. The new window design evokes the clunky aesthetics of low-quality third-party Mac utilities. The transition feels less like a modernization and more like a regression in both appearance and usability.
OpenAI has also confirmed that this modified desktop experience will soon debut on Windows, which means those who have so far escaped the change will soon face similar frustrations. This overhaul bears all the hallmarks of a product decision focused more on monetization strategy than user happiness. The reasoning behind it, however, becomes clearer upon closer inspection.
For OpenAI, chat-only users occupy what might be considered the lower-cost tiers of their ecosystem—those on the free, $8/month Go, or $20/month Plus plans. Higher-end services, such as Codex or agent-based systems like ChatGPT Work, operate at a much greater computational cost and therefore are gated behind subscription tiers starting around $100/month. Advanced AI agents require substantial processing power—and therefore more billing tokens—to perform complex actions autonomously. It follows, therefore, that the more customers migrate toward using Work’s agentic features, the greater the revenue potential for OpenAI.
Yet, this strategic expansion comes at the price of user satisfaction. The newly merged app no longer feels like the conversational assistant that millions of users have grown accustomed to. When opening the so-called ChatGPT app after the upgrade, a splash screen bearing the Codex name greets you. The app even prompts you to adopt the Codex icon instead of the ChatGPT one. Inspecting its internal information panel reveals a confusing contradiction: some parts identify it as Codex, others as ChatGPT. This confusion mirrors the broader identity crisis now facing OpenAI’s desktop software.
For instance, within the interface, the task list on the left side of the window displays coding projects recently accessed through Codex—a stark reminder that this is no longer ChatGPT as users know it. My own project history included entries from unrelated development work, illustrating how deeply Codex’s DNA permeates the new environment. If you switch to *Work Mode*, you are essentially invoking the same tool but directed toward business-oriented workflows, not natural conversation. In either case, the familiar and user-friendly ChatGPT interface—the layout that supported chat histories, GPT libraries, and productivity shortcuts—is nowhere to be found.
Even the so-called “chat” element of ChatGPT has been buried behind a small, easily overlooked button. Hovering over this miniature icon reveals only the last handful of queries. Clicking it opens a small pop-up chat window that looks rudimentary compared to the polished environment users once enjoyed. Gone is the structured sidebar with pinned conversations, project groupings, and seamless access to the GPT Library. The sense of organization and continuity that once defined the ChatGPT desktop workflow has evaporated.
This transformation leads to a clear and discouraging conclusion: the ChatGPT desktop experience as we knew it has been effectively retired. The combined Codex-and-Work interface may deliver added functionality for developers or enterprise teams, but for existing users who relied on ChatGPT’s streamlined communicative workspace, the shift represents a major step backward.
To its credit, Codex remains an impressive tool in its own right. It excels at what it was designed to do—code completion, automation, and task-oriented agent execution. Yet merging it wholesale with ChatGPT without preserving the latter’s signature features feels unnecessarily destructive. A more thoughtful integration—perhaps through toggles or separate tabs—could have preserved the old while ushering in the new.
Still, not everything has been lost. The core ChatGPT experience remains entirely intact within a browser. There, users will still find the familiar sidebar housing conversations, GPTs, the Library, and other creative resources. Even the Work tab exists within that environment, capable of executing multi-step tasks in the cloud. Although this version can’t directly interface with a user’s desktop system, it offers continuity, stability, and access to ChatGPT’s evolving web-centric features.
Thus, for those who have grown accustomed to ChatGPT in a browser, daily workflows will continue unaffected. For others—particularly those who embraced the desktop app for its efficiency—the change represents a tangible setback. I have reached out to OpenAI to ask whether the company intends to restore the previous desktop features or provide a dedicated mode for traditional ChatGPT use. Once a response arrives, I will share it.
In the meantime, I plan to conduct an in-depth comparison between the new cloud-based and local implementations of ChatGPT Work, alongside Anthropic’s Claude Cowork. The results should reveal whether this pivot toward agentic computing truly justifies the trade-offs users are being asked to accept.
Meanwhile, if you’ve experienced these changes yourself, I’m eager to hear how they’ve affected your workflow. Share your impressions in the comments—and for updates, follow my ongoing project notes across social platforms and newsletters. You can find me on Twitter/X (@DavidGewirtz), Facebook (Facebook.com/DavidGewirtz), Instagram (Instagram.com/DavidGewirtz), Bluesky (@DavidGewirtz.com), and YouTube (YouTube.com/DavidGewirtzTV). Stay tuned as this evolving story continues to unfold.
Sourse: https://www.zdnet.com/article/openai-gutted-chatgpt-desktop-app-for-codex-work/
