We introduced the term plan steering to describe the process of predicting and avoiding plan failures, much as one steers a car around potholes. We win describe a plan steering system for managing transportation networks, and a suite of experiments which show that mixed-initiative plan steering works better than humans or programs working alone. More recently we bnilt a mixed-initiative plarmer for data analysis. In a large dataset, the space of relationships between variables is vast, and many look attractively informative. It proved challenging to keep the human/ computer analyst team working toward the same goals and informing each other of their progress.