A few days ago, I purchased an annual membership for Kimi Cli. I tested its performance in common scenarios, and here are the impressions: 1️⃣ Help me install software. On Mac, for software like NetEase Cloud Music, it can fully download the DMG and execute the installation. Completion rate: 🌟🌟🌟 2️⃣ Help me organize Gmail emails. In this case, because I have over 30,000 emails, Kimi often needs me to confirm whether to continue organizing. I told it to proceed without asking me, but this indicates that long-term tasks are somewhat limited by token consumption or RL guidance issues. For such fixed long-term tasks, performance is somewhat lacking. Completion rate: 🌟 3️⃣ Help me find suitable jobs. I provided it with my resume and asked Kimi to search for suitable jobs across the internet. The recommendations are quite good, but the downside is that there’s no specific job information—only the company name at most. The job types vary greatly, with impressive ones like World Labs, and less known Eastern European companies. This might be limited by the tool use capability of the model. It’s very good as a directional guide, but not yet truly guiding. Completion rate: 🌟🌟 4️⃣ Help me plan travel itineraries. In scenarios where the goal isn’t clear, I asked it to suggest options for the next five days based on whether I want to watch football in the UK, rest at home, or travel in Japan. The experience here isn’t quite as good as Doubao; the suggestions are generic and lack follow-up questions. Compared to Claude Code, which repeatedly provides three options to converge on a goal in multi-objective fuzzy task definitions, Kimi Cli could improve in this aspect. Completion rate: 🌟🌟 Overall, I think Kimi Cli is a very capable worker, willing to work tirelessly. However, it always asks for my confirmation on fixed tasks, while it can freely perform in non-fixed tasks, which confuses me. I believe @Kimi_Moonshot urgently needs an Agentic RL to optimize the experience. Besides that, other daily functions like summarizing PDFs, scripting locally, and managing schedules surprised me and made me feel that the annual membership was worth it.