AI entrepreneur and developer Mckay Wrigley recently shared a technique via social media aimed at significantly boosting the performance of Anthropic's Claude Code. In a tweet, Wrigley advised users to "create a hook for Claude Code that adds an 'ultrathink' prompt to requests ending in '-u'," claiming it to be a "super easy way to maximize performance on-demand" that "Takes <90 seconds."
Claude Code, developed by Anthropic, is an agentic coding tool designed to assist developers directly within their terminal environments. It enables users to accelerate software development by automating routine tasks, explaining complex code, and managing Git workflows through natural language commands. The tool leverages advanced AI models like Claude Opus 4 and Sonnet 4 to understand entire codebases and facilitate rapid iteration.
Wrigley's "ultrathink" suggestion falls under the umbrella of prompt engineering, a practice where users craft specific instructions or "hooks" to guide an AI model's behavior and output. This technique is akin to established prompt engineering strategies such as "chain of thought" or instructing models to "think step by step," which encourage the AI to engage in more deliberate and structured reasoning, often leading to improved accuracy and efficiency in complex tasks.
The appeal of Wrigley's method lies in its promised quick implementation and immediate impact. By integrating a custom "hook" that appends the "ultrathink" instruction to specific queries, developers can, according to the tweet, achieve on-demand performance enhancements without extensive setup. This highlights a growing trend among advanced AI users to customize their interactions with large language models for specialized workflows.
The advice underscores the evolving landscape of AI development, where understanding and manipulating prompt structures can unlock significant gains in productivity and capability. Developers are increasingly exploring such personalized configurations to tailor AI tools like Claude Code to their unique needs, pushing the boundaries of what these powerful assistants can achieve in real-world coding scenarios.