The article explores the concept of "model weight first" companies, which leverage large language models (LLMs) more effectively than traditional corporate environments. These companies build their software using the preferences embedded in the LLMs, rather than imposing rigid corporate rules on the AI's output. By working with the inherent strengths of the models, they avoid the pitfalls of context engineering that many corporations face when trying to enforce naming conventions or coding standards. The author likens this approach to woodworking, suggesting that just as one should work with the grain, companies should align with the model's preferences to achieve better results. The author posits that model weight first companies can significantly outperform traditional firms in terms of productivity and success rates. They argue that when tasks like building a Docker container are simplified by using the natural capabilities of an LLM, the experience is seamless. In contrast, corporate environments burden the AI with additional rules and specifications, leading to suboptimal outcomes. For example, configuring a Docker container in a corporate setting often involves navigating numerous restrictions, which complicates the process and results in frustration. The article suggests that a company's approach to AI can be a determinant of its success. Firms attempting to mold AI outputs to fit outdated practices may struggle to realize the full potential of the technology. In contrast, those willing to adapt and embrace a model weight first philosophy could find themselves at a competitive advantage, particularly as AI continues to evolve. The author encourages a reevaluation of existing corporate dogmas, implying that flexibility and openness to change could lead to more effective AI integration.