As of 2026, the author reflects on their experiences with Claude Code and broader trends in code generation tools. They note that patterns in usage are emerging, similar to those seen with early chat models. Key developments in model improvements—like the shift from base models to specialized instruct models—highlight the evolving landscape of AI. The author emphasizes the importance of maintaining control over one's thought process, suggesting that while models can handle tasks efficiently, they cannot replace critical thinking. The concept of "context rot" is introduced, explaining how irrelevant information can confuse models during interactions. The author recommends using features like CC's /clear and /catchup commands to manage context effectively. They advocate for a more thoughtful approach to using code generation models, emphasizing the value of asking deeper questions rather than simply outsourcing tasks. For example, instead of asking for code fixes, users should focus on understanding underlying issues and implications. This approach fosters a mindset of "learning to learn," which is seen as essential regardless of advancements in AI capabilities. The author discusses the limitations of current chat models, particularly in how they disrupt developer workflows. They highlight the friction caused by switching contexts and transferring generated code to development environments. CC's Agents approach aims to reduce this friction, offering a more seamless experience for developers. The author praises CC's capabilities, especially in its latest version, Opus 4.5, noting that it increases productivity without needing extensive domain context. They suggest creating a CLAUDE.md file to document repeatable actions and improve the model's performance with specific projects, allowing for better interactions and reduced errors in code generation.