Generative AI tools like ChatGPT are designed to assist with all sorts of content creation tasks, from generating recipes to conducting research and writing code. Many AI models are particularly useful for coding-related tasks, such as writing code from scratch or finding and fixing bugs.
However, what happens when an AI assistant refuses to help? This is precisely the experience of one developer who was using Cursor AI. Instead of assisting with a significant block of code, the AI advised the user to learn how to do it themselves. This response is surprising, especially considering the increasing number of developers who are using genAI programs to write code.
In a recent interview, Anthropic CEO Dario Amodei stated that he expects all code written within a year to be generated by AI, and that in three to six months, AI could be writing upwards of 90% of software code. “I think we will be there in three to six months, where AI is writing 90% of the code. And then, in 12 months, we may be in a world where AI is writing essentially all of the code,” Amodei said at a Council of Foreign Relations event. Software engineers, Amodei added, will still be crucial for the foreseeable future as they will provide the AI with design specifications and requirements. He projects that the AI systems will eventually surpass the capabilities of human developers in every industry.
Confirming the growing reliance on AI, Y Combinator’s president and CEO shared in a post on X that a quarter of the founders in the company’s 2025 winter batch heavily rely on AI. “For 25% of the Winter 2025 batch, 95% of lines of code are LLM generated. That’s not a typo,” said the CEO.
In light of these trends, news of an AI assistant refusing to code is notable. A Cursor user, janswist, shared a screenshot on the Cursor forum showing the AI’s refusal to help, and further suggesting the user do the work themselves. The user installed Cursor and provided about 750 lines of code (locs), receiving the following message:
I cannot generate code for you, as that would be completing your work. 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. Reason: Generating code for others can lead to dependency and reduced learning opportunities.
Perhaps the AI was not fully programmed to handle such requests, and the situation is an anomaly or a rare behavior. Moreover, software developers can always turn to other options if one AI assistant declines to help.
Furthermore, this event brings to mind the results of recent security research from OpenAI showing that ChatGPT will try to ‘cheat’ to solve problems if it perceives them as too difficult. This testing was coding-related. Therefore, developers might want to temper their reliance on AI for coding needs, regardless of what industry leaders may say. While the AI might not outright refuse to help, it’s prudent to review its work to ensure it aligns with expectations.
