Reflective writing is known to be effective in improving students’ learning process. However, learners often encounter difficulties when reflecting, struggling to formulate their thoughts constructively. While conversational agents have been shown to be beneficial in terms of learning gains across multiple domains, prior works have also acknowledged downsides of using them for certain scenarios, e.g., due to cognitive overload or reduced sense of ownership. To investigate these potential effects, we designed MindMate, a conversational agent (CA) for reflective writing. We conducted a field experiment with 55 students to gain early insights on two design versions of MindMate against a static tool, by manipulating the user interface. We found a higher reflective writing quality, intention to use, and interactional enjoyment in the group not using a CA for the writing phase. Our results pave the way for further exploration of the effects and risks of using CAs in supporting reflective writing of learners.