AI agents use language generation for internal reasoning and interacting with users. Agents often employ tools beyond language generation, such as calculators or search, to further augment these capabilities. We focus on how such tools can give the agent too much external context, diverting it from the user's original intent. According to Conversation Analysis, human-human dialogue often uses "insert-expansion" - inserted utterances for clarification - to resolve ambiguities. Building on this, we introduce a "user-as-a-tool" approach, enabling the AI agent to solicit clarification from the user while still reasoning, thereby realigning it with the user's intent. Initial evidence shows that our approach has benefits for conversational recommendation systems. We present a novel interaction method and empirical findings that enhance the user's role in guiding agent reasoning. This research is especially relevant as AI agents become increasingly common, and holds significance for optimizing the human-chatbot interaction loop.