After two years of experimenting with AI coding, the author has shifted back to manual coding. Initially, the excitement of AI's capabilities leads many to give it increasingly complex tasks, only to encounter limitations. While AI can handle simple requests well, it struggles with more nuanced demands. The author describes a common pattern among developers: they start believing that if they provide detailed specifications, the AI will deliver exactly what they want. However, this approach often fails because AI models can't adapt or revise their plans as real-world projects evolve. As the author dives deeper into the code produced by AI, they notice a troubling trend: the code may appear impressive on the surface but is riddled with inconsistencies and flaws. This “slop” becomes apparent when reviewing the codebase in full. The AI's output may satisfy requirements in isolation, but it lacks a cohesive understanding of the overall architecture or design. The author likens the AI's narrative capabilities to writing a novel where individual chapters might read well, but the entire story lacks coherence. Ultimately, the author concludes that manual coding is more effective. They find themselves faster, more accurate, and more creative than relying on AI, especially when considering the quality of the output. The experience underscores a growing skepticism about the reliability of AI in serious coding tasks, leading the author to prioritize transparency and integrity in their work over the allure of technology.