Vibe Coding Killed Cursor (ischemist.com)

In a recent analysis, the author critiques the shift towards "vibe coding" within AI development environments, particularly its impact on Cursor, a tool designed for code generation using large language models (LLMs). Since the advent of ChatGPT and its successors, numerous LLMs have emerged, complicating the landscape for users not deeply engaged in AI discussions. The author argues that vibe coding—where users excessively interact with LLMs for minor changes—has led to inefficiencies and increased costs, particularly in how Cursor handles token usage. This approach encourages users to rely on LLMs for even trivial modifications, resulting in ballooning token consumption that ultimately undermines productivity. The author emphasizes a more efficient coding workflow by contrasting various LLMs, specifically pointing to the effectiveness of Gemini 2.5 Pro in long-context scenarios. Unlike Cursor’s approach, Gemini allows for more strategic communication by enabling comprehensive coding conversations without the high costs associated with vibe coding. The article highlights innovative features in AI Studio, such as message editing and context management, which enhance productivity. This conversation about optimal use of LLMs underscores the significance of effective engineering and user behavior in driving the future development of AI tools, suggesting a pivot back to more traditional programming methods may be necessary for serious software engineering tasks.