diff --git a/src/agent/dispatcher.rs b/src/agent/dispatcher.rs
index fcead248..78d23c31 100644
--- a/src/agent/dispatcher.rs
+++ b/src/agent/dispatcher.rs
@@ -33,12 +33,16 @@ impl Agent {
     /// Returns `AgenticLoopResult::Response` on completion, or
     /// `AgenticLoopResult::NeedApproval` if a tool requires user approval.
     ///
+    /// When `resume_after_tool` is true the loop already knows a tool was
+    /// executed earlier in this turn (e.g. an approved tool), so it won't
+    /// force the LLM to use tools if it responds with text.
     pub(super) async fn run_agentic_loop(
         &self,
         message: &IncomingMessage,
         session: Arc<Mutex<Session>>,
         thread_id: Uuid,
         initial_messages: Vec<ChatMessage>,
+        resume_after_tool: bool,
     ) -> Result<AgenticLoopResult, Error> {
         // Load workspace system prompt (identity files: AGENTS.md, SOUL.md, etc.)
         let system_prompt = if let Some(ws) = self.workspace() {
@@ -110,6 +114,8 @@ impl Agent {
 
         const MAX_TOOL_ITERATIONS: usize = 10;
         let mut iteration = 0;
+        let mut tools_executed = resume_after_tool;
+
         loop {
             iteration += 1;
             if iteration > MAX_TOOL_ITERATIONS {
@@ -193,12 +199,30 @@ impl Agent {
 
             match output.result {
                 RespondResult::Text(text) => {
+                    // If no tools have been executed yet, prompt the LLM to use tools
+                    // This handles the case where the model explains what it will do
+                    // instead of actually calling tools
+                    if !tools_executed && iteration < 3 {
+                        tracing::debug!(
+                            "No tools executed yet (iteration {}), prompting for tool use",
+                            iteration
+                        );
+                        context_messages.push(ChatMessage::assistant(&text));
+                        context_messages.push(ChatMessage::user(
+                            "Please proceed and use the available tools to complete this task.",
+                        ));
+                        continue;
+                    }
+
+                    // Tools have been executed or we've tried multiple times, return response
                     return Ok(AgenticLoopResult::Response(text));
                 }
                 RespondResult::ToolCalls {
                     tool_calls,
                     content,
                 } => {
+                    tools_executed = true;
+
                     // Add the assistant message with tool_calls to context.
                     // OpenAI protocol requires this before tool-result messages.
                     context_messages.push(ChatMessage::assistant_with_tool_calls(
diff --git a/src/agent/thread_ops.rs b/src/agent/thread_ops.rs
index c1d6442f..66f3723c 100644
--- a/src/agent/thread_ops.rs
+++ b/src/agent/thread_ops.rs
@@ -278,7 +278,7 @@ impl Agent {
 
         // Run the agentic tool execution loop
         let result = self
-            .run_agentic_loop(message, session.clone(), thread_id, turn_messages)
+            .run_agentic_loop(message, session.clone(), thread_id, turn_messages, false)
             .await;
 
         // Re-acquire lock and check if interrupted
@@ -1057,7 +1057,7 @@ impl Agent {
 
             // Continue the agentic loop (a tool was already executed this turn)
             let result = self
-                .run_agentic_loop(message, session.clone(), thread_id, context_messages)
+                .run_agentic_loop(message, session.clone(), thread_id, context_messages, true)
                 .await;
 
             // Handle the result