This guide offers comprehensive prompting tips for GPT-5, a new flagship model designed for advanced agentic tasks, coding, and general intelligence. It focuses on optimising model outputs by addressing concepts such as improving agentic workflow predictability through calibrating the model's 'eagerness' to act, and enhancing coding performance for tasks ranging from app development to refactoring. Today's episode also highlights the importance of instruction adherence and steering model behaviour through parameters like 'verbosity' and 'reasoning effort', drawing on real-world prompt tuning examples from the AI code editor, Cursor, to illustrate best practices. Ultimately, the guide aims to empower users to maximise GPT-5's capabilities by providing strategies for effective communication with the model and for leveraging new API features, including the Responses API for reusing reasoning context.

What topics should we do next?