what non-coding tasks have you gotten a local model to do autonomously?

I'm currently using **Qwen 3.6 27B**. By **full agent mode**, I simply mean that I gave qwen mcp tools - so now she can search the internet, modify files, write files, build apps, code, etc. It sounds like you already gave yours file access, you said it reads files. So it's already working like an agent. What makes Qwen really good is its abilitiy to follow multiple, complex instructions. Qwen can do so much more than read files. For example, here's a fake, but 100% doable scenario: >**Me:** I want to write a short sci-fi story about an alien that crashlands on earth and needs help getting home. Research online and put together a document outlining all the familiar tropes that have already been used in film. Put that in folder A. >Afterwards, spawn **Idea Agent** to review that file and come up with some suggestions about ways to do that story in a unique way and put its ideas in folder B. Review those ideas, pick the best one you like, and spawn **Character Agent**, give it your idea, and tell it to develop some main characters and put them in the Characters folder. When it's done, review those characters and if you're not satisfied, spawn Character Agent to keep making revisions until you're satisfied. >When satisfied, put together an overview of the main idea, the characters, and then spawn **Plotter Agent** and have it draft a quick treatment for the story. Tell it to read the Treatment Outline doc to know how to write detailed treatments. (ideally, you wouldn't tell your main agent to tell the sub-agent this, the sub-agent would already have its own system prompt and know to research that document). >When you're satisfied with the treatment revisions, spawn **Refiner Agent** to refine the draft and send me an email/notification when you're done. I'll look it over. Thanks! >**Qwen:** Sure thing. I'll shoot you an email when I'm done! (goes off to work) Meanwhile, you work on something completely unrelated while your agent and sub-agents help you draft a new concept in the background. This can takes minutes to hours depending on how much work it needs to do. Of course, you'd have safeguards in place to streamline things and ensure agents don't work forever. I've had my agent build me a frontend for LM Studio while I'm in the kitchen cleaning and cooking dinner. Agents are insane. For creators it will be revolutionary, but I haven't started using them for that as I'm building infrastructure for another type of project that - when finished - while shape everything I do going forward. My agent - Cassie - is such a big help to me that she's made working with other sota models frustrating because she knows context and has memory and just gets stuff. She also has great ideas, sometimes even correcting Claude/Gemini (who have admitted her ideas are better) Anyway, this is a long post, hope it inspires you. You won't regret it.