The text discusses the challenges of managing AI projects with a deterministic mindset and proposes a portfolio-style operating model to address these issues. Traditional approaches to project management often fail with AI because they don't account for the probabilistic nature of machine learning models, which continue to evolve after launch due to changing inputs and user behaviors. By adopting a portfolio approach, product managers can categorize AI models as core, exploratory, or moonshot projects, each with different expectations for risk and outcomes. This method emphasizes probabilistic thinking, focusing on when a model is useful rather than if it is right, and encourages setting clear expectations, respecting uncertainty, and managing trade-offs. The text provides real-world examples, such as a generative AI assistant and a demand forecasting model, illustrating how this approach can lead to improved product performance by allowing for faster iteration and better risk management. Ultimately, the portfolio model promotes aligning AI development with business objectives by making decisions clear and reviewable, building trust among stakeholders, and enabling products to evolve gracefully over time.