用 6 个 AI 代理运营 1.9 万粉账号的 Grok Bot 配置

Yarchi @undefinedKi原推发布于 2026.10.21

在 X 上打开原推

I run a 19k X account with 6 AI agents and I'd say running a blog has never been easier. Here is my setup Most of it runs on Grok Bot. I've tried a lot of different agents and setups, and this is the one I like most so far so I decided to share it. Easier doesn't mean easy though. Agents don't write my posts. Here's what actually works: 1. One agent per job, separate chats. When research, writing and inbox lived in one chat, context got mixed and the quality dropped. Now each agent only knows its own task and my rules for it 2. Rules go into memory, not prompts. Every time I fix something ("no like/share bait", "authority in the first 5 words", "I make images myself"), the agent saves it. I correct each mistake once, not every day 3. I write, the agent edits. I dump the idea in my own words, with typos and broken English. The editor fixes grammar, repeated words and rhythm, and isn't allowed to add facts I didn't give it. My voice stays mine 4. Data before ideas. I had an agent go through a month of my posts: what got 15k+ views and what got under 5k. The pattern was not what I expected: [your main takeaway]. Now every topic gets checked against it 5. Duplicate check. Before I write anything, an agent searches my own posts to see if I've covered it already. Sounds obvious, it saved me from repeating myself twice in one week 6. DMs get sorted. An agent reads the requests, splits collabs, questions and spam, and drafts a reply for the ones worth answering. The same questions come up again and again, so most replies are ready before I open the inbox. I read, tweak and send 7. Morning prep runs without me. [Morning Brief / Bookmarks: what lands in front of you before you start] 8. Agents never publish. Every post, reply and email goes through me. One wrong auto-post costs more than all the time saved The full version of this setup for an agency is in my article below

这条是引用着另一条写的：x.com/undefinedki/status/2111804964334883320