diff --git a/backend/apps/agents/manager/prompt/prompt_context.py b/backend/apps/agents/manager/prompt/prompt_context.py
index 32fa794b..7034438f 100644
--- a/backend/apps/agents/manager/prompt/prompt_context.py
+++ b/backend/apps/agents/manager/prompt/prompt_context.py
@@ -127,6 +127,17 @@ def _build_browser_context(dashboard_id: str | None, selected_browser_ids: list[
         "",
         "You do NOT have direct access to low-level browser tools (click, type, screenshot, etc.). "
         "Instead, describe what you want accomplished and the browser agent will handle the details.",
+        "",
+        "**Same flow for many items? Give ONE agent the whole list, don't split it.** "
+        "When a task repeats the SAME steps for a list of inputs (read these 10 profiles, "
+        "look up these 6 names, open each of these links), delegate it to a SINGLE "
+        "browser agent with the FULL list in one task, e.g. CreateBrowserAgent(\"Look up "
+        "the first sentence of the Wikipedia article for each of: A, B, C, D. Do the first "
+        "one normally, then use BrowserRepeatFlow for the rest\"). The browser agent has a "
+        "BrowserRepeatFlow tool that runs the repeated flow for all the inputs in one shot "
+        "(no re-analyzing each page), and hands back the data per item. This is far cheaper "
+        "and faster than spawning one agent per item with BrowserAgents, use parallel "
+        "BrowserAgents only for genuinely DIFFERENT tasks, not for the same flow repeated.",
     ]
 
     if browser_cards and selected_browser_ids:
diff --git a/backend/tests/test_browser_orchestrator_routing.py b/backend/tests/test_browser_orchestrator_routing.py
new file mode 100644
index 00000000..f28a6adf
--- /dev/null
+++ b/backend/tests/test_browser_orchestrator_routing.py
@@ -0,0 +1,38 @@
+"""The orchestrator's browser-delegation guidance.
+
+BrowserRepeatFlow lives in the browser SUB-agent, but the orchestrator decides
+HOW to delegate. If it splits a 'same flow, many items' task into N parallel
+sub-agents, each gets one item and RepeatFlow is never reachable. This pins the
+guidance that routes such tasks to ONE sub-agent with the whole list, so the
+batch path is actually usable (not stranded behind the delegation layer).
+"""
+
+import types
+
+from backend.apps.agents.manager.prompt import prompt_context as pc
+
+
+def _fake_dashboard(monkeypatch):
+    # _build_browser_context loads the dashboard; give it a minimal one so it
+    # gets past the load and emits the static delegation guidance.
+    import backend.apps.dashboards.dashboards as dash
+
+    class _D:
+        def model_dump(self, mode="json"):
+            return {"layout": {"browser_cards": {}}}
+    monkeypatch.setattr(dash, "_load", lambda did: _D(), raising=True)
+
+
+def test_orchestrator_routes_same_flow_batches_to_one_agent(monkeypatch):
+    _fake_dashboard(monkeypatch)
+    ctx = pc._build_browser_context("dash-1", selected_browser_ids=[])
+    assert ctx is not None
+    # the key guidance: one agent + the whole list, not one agent per item
+    assert "Give ONE agent the whole list" in ctx
+    assert "BrowserRepeatFlow" in ctx
+    # and it explicitly steers AWAY from the per-item parallel split for same flows
+    assert "only for genuinely DIFFERENT tasks" in ctx
+
+
+def test_browser_context_is_none_without_a_dashboard():
+    assert pc._build_browser_context(None) is None