Anysphere’s Cursor AI editor recently added a visual web designer, but the enhancement has drawn mixed reactions from developers who are voicing concerns about bugs, frequent UI changes, and high subscription costs.
Cursor is a fork of Visual Studio Code (VS Code), first released in 2023, customized for AI-assisted software development. Version 2.2, released last week, introduced a bunch of new features.
One of these is debug mode: the developers describes a bug to an AI agent which then instruments code with logging statements in order to verify the likely cause. The agent will then propose a fix and invite the developer to re-test. According to Anysphere, the new mode enables more precise fixes instead of “hundreds of lines of speculative code.”
The most eye-catching new feature is a visual web editor, built into a browser sidebar. Cursor’s web browser can run in a separate window or as an inline pane, and the sidebar now allows page elements to be moved, realigned, sized, colored, and have appearance such as shadows, opacity and borders modified via visual sliders. Developers then click an apply button which triggers an agent to update the code, with hot reload to show the results. Another way to use the visual editor is to select an element and then prompt the AI agent to makes changes.
A Reddit discussion shows that the feature is welcome but could increase cost, as the AI agent is always involved with changes. “Do we even need AI agents when we just make small design changes?” asked one developer.
Cost is an issue with Cursor, which allows use of various LLMs including its own models and those from OpenAI, Anthropic and Google. Its pricing is based on a usage credit, varying according to the selected plan, followed by pay as you go prices after the credit is exhausted. The exact cost of any operation is opaque as it depends on the model used and how much computation is required. Further, when third-party models are used they appear to cost more via Cursor than when purchased direct from the model provider.
“Cursor is charging you a premium to access models … you get dramatically more usage on a Claude subscription compared to pay-as-you-go with the API,” said one developer.
The trade-off is the benefit of the Cursor user interface (UI) versus the lower cost of a CLI-based AI assistant such as Anthropic’s Claude Code. Some developers though find the rapid evolution of Cursor’s UI is causing friction. “Can you NOT change the UI every week … it is infuriating as hell when I have to reconfigure the editor every week or so after an update,” said one user yesterday, a point of view that was quickly liked by other developers. Cursor engineer Andrew Milich said the feedback is understood and the team is “working to minimize UI changes going forward.”
Cursor’s head of design Ryo Lu said in an interview last month: “at Cursor the roles between designers, PMs (Product Managers), engineers, are really muddy.” In fact, he said, there are not really any PMs but rather “a lot of the PM jobs are spread across the builders in the team.” Lu said the Cursor team does not really have a roadmap “because the world is changing faster and faster, there’s new models dropping every day.”
One user commented that “they need to start hiring product managers. There’s … a complete lack of product oversight and a culture that incentivizes fast and very buggy code.”
Another user said his team was cancelling its subscription because Cursor’s developers are “obsessed with shipping new features while critical bugs introduced by your own updates are completely ignored.” He listed issues with chat management, focus and navigation, model switching, copy and paste, code integrity and more.
The Cursor developers no doubt code the product itself using Cursor so it is possible that the company is experiencing some of the downsides, as well as the benefits, of making heavy use of AI for software development.