Unlike stackful coroutines (like Go goroutines or Lua coroutines), C++20 coroutines are completely stackless.

When a function contains co_await or co_yield:

1. The compiler transforms the local variables into a heap-allocated coroutine frame object.
2. Generates an internal state machine.
3. The promise_type customizes return values via get_return_object(), initial_suspend(), and return_value().
4. Calling handle.resume() continues execution at the exact yield suspension point without allocating an OS thread stack.

Stackless coroutines have virtually zero memory footprint (typically 32 to 96 bytes per frame), enabling millions of concurrent async tasks.

The hardest part for beginners is writing custom awaitables. Once you implement a reusable Task<T>, it becomes as ergonomic as C# async/await.