Is Qwen3.8-Flash-Next too trigger happy or is it just me?

I'm currently evaluating it for coding and our use-case at work (brain for voice agent). I feel it is really eager to get work done. Tends to just go ahead and make code changes, even though I intended it to just analyze, research or look up something. It runs tool calls like crazy. I don't know if it's double-triple-checking everything, but it feels way overboard. I discussed a bug in an open source repository with it, asked if there are issues for it already, and it went ahead and created an issue lol. As our voice agent, it asks a question and immediately calls the tool to save the answer in the same response. And it keeps doing it every step of the way. In comparison, DeepSeek v4 flash (either 0731 or v4.1) seems similarly coked up. MiMo-V2.6-Flash-MOPD on the other hand I found to be a much more pleasant coding agent in this regard. Has anyone noticed the same? Maybe gotten it under control via prompting or special instructions? Because to me it feels like I'd need to completely rewrite my voice agent harness to get the performance I want.