Modern AI agents have evolved from simple reactive tools to sophisticated collaborators that adapt to user needs. This shift hinges on context engineering, which allows agents to remember user specifics—preferences, past interactions, and more—creating a personalized experience. The OpenAI Agents SDK includes a key component, the RunContextWrapper, enabling developers to manage structured state objects that evolve over time. This framework supports long-term memory, allowing agents to learn and adapt based on user behavior. A travel concierge agent serves as a practical example of this technology. It starts each session with a user profile and captures new preferences, like dietary restrictions. At the end of a session, it consolidates these preferences into long-term memory, resolving conflicts based on a clear hierarchy of inputs. The architecture ensures that the agent maintains a coherent understanding of the user across interactions, allowing it to apply knowledge consistently. When it comes to memory architecture, state-based memory is favored over retrieval-based memory for tasks requiring continuity, such as travel planning. State-based memory maintains a coherent user profile and allows for real-time updates, whereas retrieval-based systems can struggle with context and conflict resolution. Thoughtful memory design starts by considering what a human in the same role would remember, grounding it in task relevance. This approach ensures that agents can act more like persistent collaborators rather than disjointed search tools. In summary, AI agents that effectively personalize context can build trust with users while providing valuable insights for businesses. This not only enhances user experience but also equips support staff with a deeper understanding of customer needs, ultimately leading to better service and product development.