Developed adaptive thinking capabilities for Claude Sonnet 4.6 across the anomalyco/opencode and badlogic/pi-mono repositories, focusing on enhancing AI reasoning and optimizing compute usage. Leveraged TypeScript and full stack development skills to implement dynamic effort mapping and clamping, allowing fine-grained control over model reasoning levels. Automated tests were introduced to ensure stability and robustness of these features across multiple models. The work emphasized cross-repository collaboration, with clear traceability and concise, co-authored commits. These contributions advanced the reliability and scalability of AI orchestration, integrating software engineering best practices and thorough testing to support evolving adaptive reasoning requirements.