Self-guided tutorials are popular resources for learning new tasks, but they lack important aspects of in-person guidance like feedback or personalized explanations.

Adaptive guidance systems aim to overcome this challenge by reacting to users' performance and expertise and adapting instructions accordingly.

We aim to understand the users' preferred balance of automation and control, what representation of instructions they prefer, and how human experts give instructions to match users' needs.

We contribute an experiment where users perform different virtual tasks, guided by instructions that are controlled by experts using a wizard-of-oz paradigm.

We employ different levels of automation to control instructions and alter their level of detail and step granularity to match the user's needs.

Results indicate that while users preferred automated systems for convenience and instant feedback, they appreciated a degree of manual control since they felt less rushed.

Experts relied on factors such as expected expertise, hesitation, errors, and their understanding of the current task state as main triggers to adapt instructions.