Infrastructure created by clicking is infrastructure nobody can reproduce, review or safely change. Terraform fixes that, and it introduces one genuinely new concept you must understand deeply: state. Almost every serious Terraform problem is a state problem wearing a costume.
So state comes early and gets a lot of attention. What is in the file, why it must be shared and locked for a team, remote backends, the plan as a diff between desired and recorded reality, and what actually happens when reality drifts because somebody changed a setting in a console. You will deliberately break state and repair it, using import, move and targeted operations, until the file stops being frightening.
The language itself follows: resources, data sources, variables with validation, outputs, locals, the loop constructs, conditional expressions, and the dependency graph that decides what runs in parallel. Then modules done properly, with a clear interface, sensible defaults, versioning and a composition pattern that avoids the tangled module trees teams regret.
The final third is team practice. Plan output reviewed in a pull request, applying from continuous integration with short-lived credentials, policy checks that block a dangerous change, and scheduled drift detection. You build a complete environment across the course, then destroy and rebuild it to prove it is reproducible.