Cognitive Tutors are effective AI-based learning environments. The following statement summarizes three core instructional features from past summaries of deliberate practice (shown in italics) along with three elaborations (shown in bold): Good learning-by-doing instruction requires repeated practice on well-tailored tasks in varied contexts with explanatory feedback and as-needed instruction. This chapter describes these six learning-by-doing principles and how they are achieved in Cognitive Tutors. We correct some key misconceptions about cognitive tutors, including that knowledge component (KC) decomposition does not exclude, but includes conceptual connections, teachers are not replaced but valued, and up-front-telling is not the instructional focus whereas learning-by-doing guidance is. We also point to a need for more experimentation on the benefits of as-needed versus up-front instruction, better integration of supports for enhancing student motivation, and better pathways for teachers to participate in co-design (during the inevitable need for Learning Engineering beyond Learning Science) and customization.