How to Automate 90% of Your Social Media Content with Claude Code and AI Agents

A few weeks ago, the Create Content Club team published a walkthrough that has been quietly making the rounds inside the AI agents community. The premise is simple, slightly outrageous, and increasingly hard to argue with: you can replace an entire social media team with Claude Code, a handful of custom Claude Skills, and a transcription pipeline running through n8n. The system they demoed runs on the same playbook that grew their account past 400,000 followers. Watching it run end-to-end feels less like a productivity tip and more like a preview of how content teams will actually operate two years from now.

What follows is a breakdown of the workflow they showed, the technical pieces holding it together, and where it falls short — because the missing half (publishing, scheduling, and reading what is actually working) is where most creators trying to copy this system get stuck. We’ll close with a concrete way to bolt the publishing layer onto the same pipeline using Postiz, so the agents don’t just write posts that sit in a Notion database forever.

The thirty-second pitch: AI agents replacing a social media team

The opening claim is bold. Claude recently shipped Claude Code and Claude Cowork — the two features that make this kind of automation actually possible — and the demo was built in three days. The argument the creator makes is that the system is not powered by AI magic; it is powered by a strict set of operating principles for finding ideas, scripting, and shipping content, and those principles happen to translate cleanly into Claude Skills.

That distinction matters. A lot of the AI content automation experiments floating around right now are just thin wrappers around “ask GPT to write me a post.” This one is structured: every step in the workflow is its own agent, with its own job, and they are wired together so a single prompt can run the entire pipeline.

Why Claude Skills change the math on social media automation

The unit of work in this system is the Claude Skill — a self-contained AI agent with a specific job. Claude Skills are essentially packaged instructions, knowledge, and tools that Claude Cowork can invoke on demand. The team built one for each step of their content workflow: a creator finder, a viral spotter, a transcriber, a re-scripter, and a “full pipeline” skill that runs all of them in sequence.

The first skill, the creator finder, takes a niche description and returns a list of accounts worth tracking. The demo prompt was deliberately simple — “find three creators in the content creation space, focused on personal branding, not filmmaking” — and the skill handled the heavy lifting because the operating instructions live inside the skill itself, not in the prompt. The agent opens Instagram in the browser, scans profiles, applies the criteria, discards anyone who doesn’t fit, and writes the survivors into a Notion database without duplicating anyone already in there.

That last detail — “check the database first, do not duplicate” — is the kind of constraint that turns a flaky demo into something you can actually run unattended. It is also a useful tell that the person building this knows what they are doing.

The five-skill content pipeline, end to end

Once the creators are in Notion, the rest of the pipeline runs as a chain of agents. Here is the full sequence as it was demonstrated:

1. Creator Finder — searches a niche on Instagram, validates profiles, writes them to a Notion creator database without duplicates.
2. Viral Spotter — scrapes each creator’s profile and surfaces the videos that significantly outperformed their other posts. The reasoning: outliers reveal which hooks and topics the algorithm is rewarding right now.
3. Transcriber — pulls the actual audio of each outlier video and runs it through a transcription service, then writes the transcript into a content ideas database alongside a link back to the original video.
4. Re-scripter — takes an approved idea and a knowledge base of the user’s own past content, presentations, and call transcripts, and rewrites the script in the user’s voice while keeping the structural skeleton that worked.
5. Full Pipeline — a meta-skill that chains all four, kicked off with a single prompt and a niche description.

The re-scripter is the part that elevates the system above the usual “AI rewrite” tools. Most of those reword sentences and call it a day. This one is given a proper knowledge base — long-form transcripts of the creator’s own YouTube videos and calls — so the re-script actually carries the user’s beliefs, examples, and sentence rhythms instead of generic AI prose. The structure of the source video is preserved (because that structure is what made it perform), but the substance is replaced with the creator’s actual point of view.

The technical stack: what is actually running under the hood

For something this ambitious, the dependency list is surprisingly short:

• Claude Pro or Max — the agents are built on top of Claude Cowork, which is gated behind the paid plans. Pro is enough to start; the demo team uses Max because the credit usage adds up once you start running the full pipeline regularly.
• Claude Chrome extension — gives the skills the ability to open Chrome, log in to Instagram (using the user’s existing session), open profiles, and read content. This is the piece that lets the agents do things, not just write things.
• n8n — used as a transcription endpoint. Claude does not natively transcribe video, so the workflow exposes a webhook that accepts an Instagram URL and returns a transcript. The n8n flow uses Groq’s free Whisper-style endpoint to do the actual transcription.
• Notion — the durable storage layer. One database for creators, one for content ideas, plus a knowledge base page that holds the user’s transcripts and reference material.

That is the whole stack. The demo author makes a fair point: most of it is free or cheap, and the only piece that requires technical comfort is configuring the n8n webhook with a Groq API key — a five-minute setup if you follow the screenshots.

Why the missing piece is publishing

Here is where the demo stops, and where every team trying to operationalize this kind of AI content workflow hits a wall. The pipeline produces approved scripts inside Notion. Then what?

In the demo, the human still has to copy the script out, draft the actual post, attach the media, schedule it on each platform separately, and remember to come back later to pull analytics. That is the part of the social media team this system has not actually replaced. It produced great ideas and great scripts; it did not ship anything.

For a single creator, the gap is annoying but manageable. For an agency running this for ten clients across six platforms each, the gap is the entire job. A content creation pipeline that ends in a Notion database is a draft folder, not a publishing system.

Closing the loop with Postiz

This is the gap Postiz was built to close. Postiz is an AI-native social media management platform: it schedules and cross-posts to 28+ channels (Instagram, LinkedIn, X, TikTok, YouTube, Reddit, Facebook, Pinterest, Threads, and the rest), it stores media in a shared library, it runs analytics, and — critically for this kind of workflow — it exposes everything through an API, an MCP server, and a CLI so AI agents can actually use it.

That last detail is what makes it slot directly into the Claude Skills pipeline above. You add a sixth skill — call it the Publisher — and give it three things: the Postiz MCP server (or the CLI), the approved-script database in Notion, and a list of which integrations the post should go to. The skill reads the script, picks a publish time, attaches any media, and ships it. The same agent run that found the creator and rewrote the script also ends with the post being scheduled.

If you are wiring this up yourself, the practical entry points look like this:

• Postiz MCP server — the cleanest fit for a Claude Skill. Add the MCP server to Claude Cowork once and your agents get tools like integrationList, integrationSchema, and integrationSchedulePostTool for free.
• Postiz CLI — if your skill prefers shelling out, postiz upload handles media and postiz posts:create handles scheduling. Authenticate once with postiz auth:login.
• Postiz Public API — the lowest-level option. Useful when you are building the integration into n8n or a backend service rather than directly into a Claude Skill.

With that piece added, the system the Create Content Club team demoed becomes genuinely end-to-end: Claude finds the creator, scrapes the outlier, transcribes it, rewrites it in your voice, and Postiz ships it to every channel on a schedule. No copy-pasting, no Notion-to-Buffer hop, no separate analytics tool.

What this means if you are building your own AI content automation

A few things to take from this, whether or not you copy the exact stack:

• Skills beat prompts. The reason the demo works is that each step is encapsulated as a Claude Skill with its own instructions, not crammed into one giant prompt. If you are building your own social media AI agent, treat each operating principle as its own agent.
• Outliers are the signal. The viral spotter does not look for “good content” — it looks for relative outliers on a creator’s own profile. That filters out vanity metrics from creators who are just very large, and it surfaces the hooks the algorithm is currently rewarding.
• Knowledge bases are the difference between rewriting and channeling. Without a real knowledge base of your own past content, an AI re-scripter will smooth your voice into the same neutral tone everyone else’s AI produces. Feed it transcripts of your actual talks and calls.
• Approval gates are still useful. The demo author keeps a manual “idea approved” status in Notion before the re-scripter runs. That tiny human checkpoint is what keeps the agents from happily writing posts you would never actually publish.
• Don’t end at the script. If your pipeline doesn’t include publishing and analytics, you have built a fancy ideation tool, not an automation. This is the layer where Postiz fits.

Try the publishing layer that closes the pipeline

If you have already been experimenting with Claude Code, Claude Skills, or any other AI agent social media setup, the fastest way to feel the difference is to wire up the publishing step and watch one of your agent runs ship a post end-to-end. Start a free Postiz account, connect a couple of your channels, and either drop in the Postiz MCP server or call the CLI from your skill. Twenty minutes of setup and your agents stop drafting in a vacuum.

The Create Content Club team built a great front half of an automation. The back half — scheduling, cross-posting, analytics, team review — is what Postiz exists to handle. Together they’re the closest thing yet to actually replacing a social media team with software.