Human-level AI is a statement about performance: the system does about as well as a person. AGI is a statement about scope: the system can operate across domains and transfer skill between them. A system could be general but below human level, or narrow and above it.
Where the terms converge
In practice, most people mean both at once: a system with human breadth and human competence. That combined target is what the popular usage of AGI refers to, and it is why the words get swapped.
Where they separate
Chess engines are above human level and utterly narrow. A general agent that reasons across domains at the level of a bright ten-year-old would be AGI by scope and below human level by performance. Both distinctions are real, and collapsing them causes bad predictions.
The measurement problem
Human level is also harder to define than it sounds, because human performance varies enormously by domain and by individual. Any threshold you pick is a choice, not a discovery.