diff --git a/doc/source/whatsnew/v1.3.0.rst b/doc/source/whatsnew/v1.3.0.rst
index 381a05a18b278..413f43302e888 100644
--- a/doc/source/whatsnew/v1.3.0.rst
+++ b/doc/source/whatsnew/v1.3.0.rst
@@ -343,7 +343,7 @@ Plotting
 
 Groupby/resample/rolling
 ^^^^^^^^^^^^^^^^^^^^^^^^
-
+- Bug in :meth:`DataFrameGroupBy.agg` and :meth:`SeriesGroupBy.agg` with :class:`PeriodDtype` columns incorrectly casting results too aggressively (:issue:`38254`)
 - Bug in :meth:`SeriesGroupBy.value_counts` where unobserved categories in a grouped categorical series were not tallied (:issue:`38672`)
 - Bug in :meth:`SeriesGroupBy.value_counts` where error was raised on an empty series (:issue:`39172`)
 - Bug in :meth:`.GroupBy.indices` would contain non-existent indices when null values were present in the groupby keys (:issue:`9304`)
diff --git a/pandas/_libs/tslibs/period.pyx b/pandas/_libs/tslibs/period.pyx
index e276515ca1ae9..d518729b6ce67 100644
--- a/pandas/_libs/tslibs/period.pyx
+++ b/pandas/_libs/tslibs/period.pyx
@@ -1402,6 +1402,28 @@ cdef accessor _get_accessor_func(str field):
     return NULL
 
 
+@cython.wraparound(False)
+@cython.boundscheck(False)
+def from_ordinals(const int64_t[:] values, freq):
+    cdef:
+        Py_ssize_t i, n = len(values)
+        int64_t[:] result = np.empty(len(values), dtype="i8")
+        int64_t val
+
+    freq = to_offset(freq)
+    if not isinstance(freq, BaseOffset):
+        raise ValueError("freq not specified and cannot be inferred")
+
+    for i in range(n):
+        val = values[i]
+        if val == NPY_NAT:
+            result[i] = NPY_NAT
+        else:
+            result[i] = Period(val, freq=freq).ordinal
+
+    return result.base
+
+
 @cython.wraparound(False)
 @cython.boundscheck(False)
 def extract_ordinals(ndarray[object] values, freq):
@@ -1419,6 +1441,8 @@ def extract_ordinals(ndarray[object] values, freq):
 
         if is_null_datetimelike(p):
             ordinals[i] = NPY_NAT
+        elif util.is_integer_object(p):
+            raise TypeError(p)
         else:
             try:
                 ordinals[i] = p.ordinal
diff --git a/pandas/core/arrays/period.py b/pandas/core/arrays/period.py
index 2ec3282d30093..e0149f27ad6a6 100644
--- a/pandas/core/arrays/period.py
+++ b/pandas/core/arrays/period.py
@@ -38,6 +38,7 @@
     is_datetime64_dtype,
     is_dtype_equal,
     is_float_dtype,
+    is_integer_dtype,
     is_period_dtype,
     pandas_dtype,
 )
@@ -897,7 +898,7 @@ def period_array(
     if not isinstance(data, (np.ndarray, list, tuple, ABCSeries)):
         data = list(data)
 
-    data = np.asarray(data)
+    arrdata = np.asarray(data)
 
     dtype: Optional[PeriodDtype]
     if freq:
@@ -905,10 +906,15 @@ def period_array(
     else:
         dtype = None
 
-    if is_float_dtype(data) and len(data) > 0:
+    if is_float_dtype(arrdata) and len(arrdata) > 0:
         raise TypeError("PeriodIndex does not allow floating point in construction")
 
-    data = ensure_object(data)
+    if is_integer_dtype(arrdata.dtype):
+        arr = arrdata.astype(np.int64, copy=False)
+        ordinals = libperiod.from_ordinals(arr, freq)
+        return PeriodArray(ordinals, dtype=dtype)
+
+    data = ensure_object(arrdata)
 
     return PeriodArray._from_sequence(data, dtype=dtype)
 
diff --git a/pandas/tests/arrays/test_period.py b/pandas/tests/arrays/test_period.py
index 8fca2a6d83393..443eced3922ac 100644
--- a/pandas/tests/arrays/test_period.py
+++ b/pandas/tests/arrays/test_period.py
@@ -102,6 +102,17 @@ def test_period_array_freq_mismatch():
         PeriodArray(arr, freq=pd.tseries.offsets.MonthEnd())
 
 
+def test_from_sequence_disallows_i8():
+    arr = period_array(["2000", "2001"], freq="D")
+
+    msg = str(arr[0].ordinal)
+    with pytest.raises(TypeError, match=msg):
+        PeriodArray._from_sequence(arr.asi8, dtype=arr.dtype)
+
+    with pytest.raises(TypeError, match=msg):
+        PeriodArray._from_sequence(list(arr.asi8), dtype=arr.dtype)
+
+
 def test_asi8():
     result = period_array(["2000", "2001", None], freq="D").asi8
     expected = np.array([10957, 11323, iNaT])
diff --git a/pandas/tests/groupby/aggregate/test_other.py b/pandas/tests/groupby/aggregate/test_other.py
index 5d0f6d6262899..04f17865b088a 100644
--- a/pandas/tests/groupby/aggregate/test_other.py
+++ b/pandas/tests/groupby/aggregate/test_other.py
@@ -432,10 +432,14 @@ def test_agg_over_numpy_arrays():
     tm.assert_frame_equal(result, expected)
 
 
-def test_agg_tzaware_non_datetime_result():
+@pytest.mark.parametrize("as_period", [True, False])
+def test_agg_tzaware_non_datetime_result(as_period):
     # discussed in GH#29589, fixed in GH#29641, operating on tzaware values
     #  with function that is not dtype-preserving
     dti = pd.date_range("2012-01-01", periods=4, tz="UTC")
+    if as_period:
+        dti = dti.tz_localize(None).to_period("D")
+
     df = DataFrame({"a": [0, 0, 1, 1], "b": dti})
     gb = df.groupby("a")
 
@@ -454,6 +458,9 @@ def test_agg_tzaware_non_datetime_result():
     result = gb["b"].agg(lambda x: x.iloc[-1] - x.iloc[0])
     expected = Series([pd.Timedelta(days=1), pd.Timedelta(days=1)], name="b")
     expected.index.name = "a"
+    if as_period:
+        expected = Series([pd.offsets.Day(1), pd.offsets.Day(1)], name="b")
+        expected.index.name = "a"
     tm.assert_series_equal(result, expected)