not every task requires the most powerful (and energy-intensive) model - **Cache and reuse** solutions rather than regenerating the same code repeatedly - **Consider local models** for routine tasks where cloud-based inference is unnecessary