diff --git a/rdagent/components/coder/CoSTEER/evolving_strategy.py b/rdagent/components/coder/CoSTEER/evolving_strategy.py
index beb5aaf5..720cf1a4 100644
--- a/rdagent/components/coder/CoSTEER/evolving_strategy.py
+++ b/rdagent/components/coder/CoSTEER/evolving_strategy.py
@@ -21,9 +21,10 @@ from rdagent.core.utils import multiprocessing_wrapper
 class MultiProcessEvolvingStrategy(EvolvingStrategy):
     KEY_CHANGE_SUMMARY = "__change_summary__"  # Optional key for the summary of the change of evolving subjects
 
-    def __init__(self, scen: Scenario, settings: CoSTEERSettings):
+    def __init__(self, scen: Scenario, settings: CoSTEERSettings, improve_mode: bool = False):
         super().__init__(scen)
         self.settings = settings
+        self.improve_mode = improve_mode  # improve mode means we only implement the task which has failed before. The main diff is the first loop will not implement all tasks.
 
     @abstractmethod
     def implement_one_task(
@@ -93,6 +94,7 @@ class MultiProcessEvolvingStrategy(EvolvingStrategy):
             elif (
                 target_task_desc not in queried_knowledge.success_task_to_knowledge_dict
                 and target_task_desc not in queried_knowledge.failed_task_info_set
+                and not (self.improve_mode and last_feedback[index] is None)
             ):
                 to_be_finished_task_index.append(index)
 
diff --git a/rdagent/scenarios/data_science/dev/runner/__init__.py b/rdagent/scenarios/data_science/dev/runner/__init__.py
index 55db014a..d37dd436 100644
--- a/rdagent/scenarios/data_science/dev/runner/__init__.py
+++ b/rdagent/scenarios/data_science/dev/runner/__init__.py
@@ -48,11 +48,6 @@ class DSRunnerMultiProcessEvolvingStrategy(MultiProcessEvolvingStrategy):
         workspace: FBWorkspace | None = None,
         prev_task_feedback: CoSTEERSingleFeedback | None = None,
     ) -> dict[str, str]:
-
-        if prev_task_feedback is None:
-            # if no prev_task_feedback, it is the first loop; we do not make any changes and goto evaluators directly.
-            return {}
-
         # Get evolving history
         task_info = target_task.get_task_information()
         queried_former_failed_knowledge = (
@@ -157,7 +152,7 @@ class DSCoSTEERRunner(CoSTEER):
             single_evaluator=eval_l, scen=scen
         )  # Please specify whether you agree running your eva in parallel or not
         settings = DSRunnerCoSTEERSettings()
-        es = DSRunnerMultiProcessEvolvingStrategy(scen=scen, settings=settings)
+        es = DSRunnerMultiProcessEvolvingStrategy(scen=scen, settings=settings, improve_mode=True)
 
         # In runner, we don't need very big loops, so we set max_loop to runner_max_loop
         super().__init__(