How Scotty Beam Runs X Content with Seven Grok Bot Agents and Postiz
Built by SCOTTY BEAM · · Source: X
Result: About forty minutes a week of their own time, mostly spent reading. They say they haven't opened a social app to post since the setup went live.
What they built
Scotty Beam runs their content with seven Grok Bot agents. Each one has a narrow job: Chief of Staff, Researcher, Writer, Visualiser, Analyst, Scheduler, and a seventh bot called Postiz that puts finished posts on a publishing calendar. According to the author, the target is twelve long-form X posts a week plus an article every two weeks, with the same ideas reworked for LinkedIn and Threads.
How it works
- One shared strategy. Every bot gets the same short strategy block: three content pillars, posting volume, three quality checks (based on something real, one idea per post, something the reader can act on today), and a definition of success that puts saves, reposts and replies ahead of likes.
- Shared state. They say the bots keep memory and logins between tasks, share one persistent cloud computer so files pass between them directly, and hand tasks to each other in message threads.
- Weekly rhythm. On Monday the Analyst reports last week's numbers, the Researcher drops ten sourced topics, and the Chief of Staff picks five. Scotty reviews that list on their phone. The Writer, Visualiser and Scheduler then produce the posts, and the Postiz bot adds them to the calendar through the Postiz API.
- Draft-only publishing. Every post arrives as a draft. Scotty reads the drafts and previews on Tuesday, fixes a line or two, and schedules them.
The post describes the system prompts but doesn't include their text.
The result
According to the author, the setup takes about forty minutes of their time a week, and they haven't opened a social app to post since it went live. They also list three early mistakes:
- A single all-purpose agent produced mediocre work. Splitting it into narrow roles was the biggest quality improvement.
- Publishing without review let two posts go out that they wouldn't have sent.
- Describing the strategy differently to each agent made the posts sound like different people wrote them.
How to build something similar
- Write your content strategy once, in one short block, and give the exact same text to every agent.
- Give each agent one job and a clear standard instead of one agent with a long list of tasks.
- Send everything to a scheduler as drafts first, and only allow automatic publishing once you trust the output.
- Set up the publishing step early, since unpublished content is where many pipelines stall.
The original post
https://x.com/i/article/2094040136300302336 Two weeks ago I posted a breakdown of how I run my content with Grok Bot agents. I wrote it in about twenty minutes, expecting the usual couple hundred likes from people who follow me for AI stuff. Then Elon Musk reposted it. The post did numbers I have never done before, and my mentions turned into one long queue of the same question: okay but how do I actually build this? So this is that answer. The full setup, the exact prompts I use, how the agents hand work to each other, and how everything ends up scheduled and published without me touching a calendar. Here is the original post, for context: Start with the strategy, not the tools I want to be honest about something before I show you a single prompt. Agents do not fix a bad content plan. They just execute a bad plan faster and more consistently, which is arguably worse. So here is my actual strategy, written out the way I gave it to my agents. It is boring on purpose. What I post about. Three pillars, nothing else: AI agents in practice – what people are actually building, what breaks, what it costs. Automation for small teams – workflows a one or two person team can run without hiring. Behind the build – my own setups, my own mistakes, visuals included. How much. Twelve long-form posts a week on X, one article every two weeks, and the same core ideas reshaped for LinkedIn and Threads. I am not trying to be everywhere. X is the main channel and everything else is a repackage. The rules. Every post has to pass three checks: It comes from something real that happened in the last seven days, or from my own build. It contains one idea, not four. Somebody who reads it can go do something with it today. What good looks like. Saves, retweets and replies over likes. If a post gets a lot of likes and no replies or retweets, it was a nice sentence and nothing more. That whole thing fits in a paragraph, and it is the single most important input in this entire system. Every agent below is holding a copy of it. The team Seven Grok Bot agents. Each one has a narrow job. That is the trick – narrow jobs are the reason the output stays good. Why Grok Bot specifically A few things about how Grok Bot works make this possible, and they are worth understanding before you copy the prompts. Each Bot is a named teammate with durable state. It keeps memory, files, browser sessions and app logins between tasks. You are not starting a fresh chat every morning and re-explaining who you are. The Bots share one persistent cloud computer. So when my Researcher saves a file, my Writer can just open it. Nothing gets copy-pasted between tools, and nothing gets lost in the handoff. They can message each other in threads and pass ownership of a task. Put three Bots in one thread and they hand work back and forth while you watch instead of approving every step. They use connectors and MCP where they exist, and plain computer use where they do not. That is the part people underestimate. A Bot can log into a site with no API and click through it like you would. And they learn routines. Record yourself doing something once, stop the recording, and the Bot can re-run that path on a schedule. You create a Bot, message it, and give it access to what it needs. There is no workflow builder to learn. The prompts These are the actual system prompts I have on each Bot, cleaned up a little. Copy them, swap my pillars for yours, and you have the same setup running today. 1. Chief of Staff This Bot is the only one I message directly on a normal day. 2. Researcher 3. Writer This is the one that took the longest to get right. The voice section is the part that matters. 4. Visualiser I fed this one about fifteen of my own images once. Now everything it makes looks like it came from the same account, because it did. 5. Analyst 6. Scheduler Then everything has to actually get published This was the part that stayed manual the longest, and it was the p
Want to build something like this?
Join Shipping Skool free: 4 live build sessions a week and a community that ships.
Join Shipping Skool — Free