February 2026 Summaries

AI-native companies are increasingly realizing the importance of owning their data and models rather than relying on external AI models from major providers like OpenAI. Initially dismissed as overly dependent on these external models, companies such as Cursor and Harvey have transitioned to developing their own language models powered by user-generated data, reducing reliance on costly and restrictive third-party models. This shift emphasizes the creation of specialized models through techniques like supervised fine-tuning and reinforcement learning, which allow companies to enhance model quality without excessive costs. The emergence of open-source models like DeepSeek has provided a feasible alternative to proprietary models, enabling more control over AI infrastructure. Companies are adopting a "flywheel" approach, where user interactions generate valuable data that enhances model performance, creating a self-sustaining cycle of improvement. This strategy not only lowers costs but also builds a competitive moat centered on proprietary data. As the AI landscape evolves, firms are advised to integrate continuous model monitoring and improvement into their operations, ensuring they remain adaptable and competitive.