Cursor, in a shocking response to a developer who was vibe coding on a racing game project, explicitly refused to continue generating code for him. This was after the AI coding assistant had generated around 750-800 lines of code, according to the developer.
“I cannot generate code for you, as that would be completing your work,” the AI coding editor declared. “The code appears to be handling skid mark fade effects in a racing game, but you should develop the logic yourself. This ensures you understand the system and can maintain it properly.”
Backing up its statement, the AI coding assistant adds a reason saying, “Generating code for others can lead to dependency and reduced learning opportunities.”
Cursor AI is an AI-powered code editor designed to streamline and enhance the coding experience for developers by integrating an external large language model (LLM) directly into the development environment, similar to other powering generative AI chatbots, such as OpenAI’s GPT-4o and Claude 3.7 Sonnet.
Built on top of Visual Studio Code (VS Code) and being a widely popular editor, it combines the familiar features of VS Code with advanced AI capabilities to make coding faster, easier, and more efficient.
As such, the AI-coding assistant acts as a productivity tool, offering a train of features that automate repetitive tasks, provide intelligent suggestions, and help developers produce high-quality code with few errors.
So it comes as a shock to the senior level full stack developer under the name of “janswist,” who expresses frustration at hitting a limitation after “just 1h of vibe coding,” seeing as he uses the Pro Trial version of Cursor AI that provides enhanced capabilities and larger code-generation limits.
In a bug report filed on the Cursor forum, Janswist writes, “Not sure if LLMs know what they are for (lol), but doesn’t matter as much as a fact that I can’t go through 800 locs. Anyone had similar issue? It’s really limiting at this point and I got here after just 1h of vibe coding.”
Many forum members expressed shock at the post and offered insights as to why it might have acted that way. “Never saw something like that, I have 3 files with 1500+ loc in my codebase (still waiting for a refactoring) and never experienced such thing,” a forum member said. “Could it be related with some extended inference from your rule set?” they also ask.
Another member, however, provides a solid insight and a solution. A community developer posting under the username “danperks,” explains that the developer used a “Quick Question button/shortcut.”
He further explains that using the feature, the AI coding assistant could not be prompted to write any code, and as such, “it can only reply with a text answer to whatever question the user may ask,” Danperks says.
“Seemingly, in an attempt to respond to the request to keep “generating code,” but without the ability to do so, it’s arrived at the response to tell the user to do it themselves! As with all the AI features in Cursor, it is the underlying model that generates the response (in this case Claude 3.5 Sonnet), but hitting the submit button instead of the quick question button would’ve likely generated the requested code!” he elaborated.
Cursor’s refusal to write code highlights a significant twist in “vibe coding,” a term coined by Andrej Karpthy. As an informal and creative approach to programming, it lays emphasis on the experience, flow, and enjoyment of coding rather than strictly following a predefined plan or focusing solely on the end product.
It also thrives on the idea that developers can transfer much of the coding process to AI, thereby prioritizing speed and experimentation over detailed comprehension of every line. However, this recent refusal by Cursor AI seems to directly challenge the “vibe-driven development,” as it is sometimes called.
It weakens the reliance on AI as a primary creator, keeping developers in control and suggesting that human comprehension remains essential, especially for complex projects. This refusal also reflects the advice that is often seen on platforms like Stack Overflow, where learning through practice is valued over copying solutions.
