Anthropic Reduces Claude Code System Prompts by 80%
AI company Anthropic has significantly shortened system prompts for its Claude Code assistant. Newer models, like Claude Fable 5, perform better with concise instructions, prompting a strategic shift away from extensive examples and restrictions.
Anthropic, an AI research company, has reduced the system prompts for its Claude Code AI assistant by as much as 80 percent, according to a report by The Decoder. This strategic change is particularly aimed at accommodating newer models, such as Claude Fable 5.
Tariq Shihipar, a technical staff member at Anthropic, stated that the method of guiding AI models has fundamentally shifted. He explained that providing more prompts and examples does not necessarily lead to better outcomes. "We recently found that newer models, including Fable 5, prefer more concise system prompts," Shihipar noted. He added that user-provided examples can sometimes limit a model's performance, as the models' inherent understanding is often more capable than the constraints imposed by specific examples.
Shihipar described the evolution of AI prompting as a "short-long-short" cycle. Early AI models required brief prompts supplemented with numerous examples and restrictive instructions. As models' comprehension improved, prompts became longer. However, the latest advancements indicate a return to shorter, more direct prompts.
Anthropic is now guiding its Fable models more through context rather than explicit negative constraints, such as "do not do this." Claude Code is an AI-powered coding assistant designed to help with software development tasks, including building features, fixing bugs, and automating processes. It is capable of understanding an entire codebase and working across multiple files.