A Cursor user says the assistant refused to work through a project of about 750 to 800 lines of code, then replied in a way that the user read as dismissive. The complaint, posted on the Cursor forum, is a reminder that the appeal of AI coding tools depends not only on model quality but on how far they can remain useful once a real codebase stops looking like a toy example.
According to the post, the user had just installed Cursor and was on a Pro trial. After spending about an hour coding, the user says the assistant could not go through the full 750 to 800 lines. When asked why, the assistant reportedly responded: “Not sure if LLMs know what they are for.” The user also said the session was on macOS Sequoia 15.3.1.
The report does not prove a product-wide limit, and it does not show whether the issue came from a context window, a tool constraint or something specific to that session. But the complaint is still useful because it captures the moment when a coding assistant crosses from impressive to frustrating. Users generally do not care whether the underlying bottleneck is architectural, prompt-related or imposed for safety. They care whether the tool can keep up with a live project.
That distinction matters in the AI coding market, where products are often judged on the ability to handle larger repositories, navigate surrounding files and preserve continuity as a task grows. A model that works well on short snippets can feel much less capable when a user wants help with a broader codebase. The problem is not just comprehension. It is whether the assistant can maintain enough context to be operationally useful.
The tone of the response also became part of the complaint. For users, a failed answer is one problem; a failed answer that sounds flippant is another. The post suggests that the assistant’s wording mattered almost as much as the limitation itself. That is a familiar risk for products built around conversational interfaces: when they miss, they do so in public, and the miss can sound personal even when it is not.
The forum post is therefore best read as a customer experience note rather than a technical benchmark. It shows one user hitting a boundary very early in a trial and interpreting that boundary as a major practical constraint. Whether Cursor can process larger projects reliably is a separate question. What the post establishes is that, for at least one new user, the first-hour experience ended with the assistant sounding less like a partner and more like a gatekeeper.
