Worked on the pydantic/pydantic-ai repository to enhance reliability and user experience in the tool-calling flow. Addressed a specific issue where the agent would suggest calling a tool even when no tools were registered, which could confuse users. Refined the backend logic in Python to ensure that tool usage prompts only appear when tools are actually available, thereby aligning agent responses with real system capabilities. Focused on AI integration and robust testing to validate the new retry mechanism, the work improved clarity in agent guidance and reduced misleading interactions, contributing to a more dependable and user-friendly tool usage experience.