XCom ("cross-communication") is how tasks hand each other values. Whatever a task returns is stored as an XCom under the key return_value; TaskFlow pulls it for you when you pass one task's result to another. In classic operators you do it by hand, with the task instance ti from the context:
def report(ti=None):
total = ti.xcom_pull(task_ids="total_revenue")
ti.xcom_push(key="row_count", value=3) stores an extra value under a key of your own, and xcom_pull(task_ids=..., key="row_count") reads it. A pull that finds nothing returns None, so a typo in a task id fails quietly.
XComs live in Airflow's metadata database, so keep them small: counts, dates, file paths, table names. For a large dataset, write it to storage and pass the path. Values must also be serializable: lists, dicts, numbers, strings and dates are fine; an arbitrary Python object, such as an open connection, is refused and fails the task.
One task returns its rows and pushes a count under its own key; another reads only the count.