My Motivation: One Man Agency Syndrome
Let me set the stage. Like lots of marketing teams these days, I’m a team of one. Strategy, content, ads, analytics… all me. And as I learned over the last 12+ years in Big Corp, I am not alone. How many teams have a blog they don’t populate? How many have social channels they can’t use properly? Even in Big Corp you always get topics that do not get the budget and it is frustrating. I’ve been there. But burnout is real, and working 80 hour weeks is not sustainable. So the go to strategy for most is “Leave it, do the bare minimum, just keep it alive somehow.”
So why not outsource the repetitive stuff to a 24/7 AI workforce? I rolled up my sleeves and set out to crack the wildest automation challenge I’ve tackled yet: a self running AI content operation that could churn out blogs, LinkedIn content, application letters and more… all while I sleep or hit the gym (ok, or watch football on TV…I am not THAT nerdy). The dream.
What I Actually Built
You know those BS LinkedIn posts about “AI generation in 30 seconds”? Well, that’s not happening. Built on GPT-4o with a custom-trained model using uploaded data, meticulously crafted prompts, JSON-based workflow logic, a chunk-based Supabase RAG system, writing styles, brand guidelines… this was anything but plug and play. Can you get a blog article or a LinkedIn post in 5 minutes? Sure, with or without automation. What you can’t get that fast is content just the way you want it, personalized as if you had invested a lot of time into it yourself.
And N8N? Well, it is marketed as “no code,” but welcome to the land where you better become fluent in JSON or be lost in a maze of trial and error. Future articles might cover recipes for building your own stack or provide more details on how I did certain parts to build my own AI agents (yes, they exist and, well mostly at least, work).
But for now, it is about what got this half crazy system to life:
N8N: How it Works
At the end of the day, N8N is kind of a hub where you can build your automations. There are other tools out there like Make.com or Zapier that have more native integrations with apps and systems ready to use, so they are a bit more user friendly if you will. N8N however doesn’t charge you for every step of the process like these other tools. They charge you for an entire execution of your workflow. And they let you host it locally and run it free of charge. That’s why I opted to try it in the end.
So how did I approach this? I said to myself: I will treat these agents like very stupid, inexperienced new employees that don’t know anything about the company, the brand, the tone of voice, the topics to cover etc. You get the point. If you ask a 12 year old to write you an article on marketing campaign tactics it will hardly fit your brand’s tone of voice and meet your target group’s needs, right? So we need a team then.
Idea Manager: Delivers specific article ideas based on a topic provided
Prompt Engineer: Lays out the blog’s required sections, key insights, things to keep in mind and things to avoid
Brand Manager: Responsible for keeping every piece of content on brand
Copy Writer: Making sure the tone of voice is on brand
Content Manager: Providing personalized information from my RAG system so my content matches my beliefs and work experience
Research Manager: Handles expert level research at warp speed, citing links. Yes, I used Perplexity here
Team Leader: Keeps it all together, checks the quality, instructs the team members and helps them along
My Secondary Automation Pipelines for Pre and Post Publishing
The content machine doesn’t automatically publish articles anywhere and it doesn’t only use AI tools to steer the entire process. Webhooks are a magical thing. I used a mix of LLMs, Notion, Google Sheets, Google Docs and Framer webhooks to get to a solid pipeline that can handle not only producing content but also images for every article and LinkedIn post, fully automated of course.
Overall? It is a functional system that still requires me to correct, change, adapt, but it works and saves me time for sure. It took 21 different pipeline iterations and four complete system reconstructions over endless hours of frustration to get here. The time “wasted” testing and failing wasn’t actually a waste. If you want to become great at AI, accept the challenge and build an infrastructure that thinks and learns… and then let it transform your workflows for tomorrow.
The N8N Reality: Patience and Precision
Let’s shatter some dreams before they even cobweb. If you envision a few clicks to automation heaven, sorry to burst your bubble.
N8N Tutorials Get You 50% (estimate), if You’re Lucky – and it won’t be in 5 or 10 minutes. Honestly, if you are working full time, have a family, want to leave the house occasionally, this is a long term project. I burned countless hours on YouTube only to find custom setups that don't transfer (outside of the basic ones, this are actually helpful). My advice? Use tutorials for initial orientation; then you’re on your own. Use ChatGPT etc, but don’t rely on them. I can’t tell you how much bad advice that was not even close to functional I got from ChatGPT for the JavaScript coding parts.
You’ll Break. You’ll Curse. You’ll Rebuild. Often. And when you think it works, it crashes. Or it works once and then does not work for some random reason the next time. Call yourself “Chief bug fixing officer”… because that’s what you’ll mainly do.
JSON Isn’t Optional or “Confusing Tech.” It’s Your Best Friend. It is the difference between a basic setup and one that actually helps you. If code is a language, JSON is the universal translator. I couldn’t adapt my N8N workflows without learning basic JSON syntax. Trust me, it’s worth the weekend headache.
No, N8N Doesn’t Have All the Answers. It automates workflows, it does not create content for you. You are essentially connecting loads of different tools… and the more complex it becomes, the easier it breaks.
Stop Listening to “I Replaced My Entire Team” Posts. Those “I saved USD 600,000 with this” LinkedIn posts by 20 year olds are eaxctly what they sound like, a mediocre at best hook. And even more importantly, they use that tool all the time. Any template they send you is worthless if you do not know how to use N8N. Because guess what: you will need to change everything in those templates.
Why the Madness is Worth It
All doom and gloom? Far from it. Despite the chaos, the system now works for what I need it. And when it works, it saves 60% (estimate, but I feel that's close) of my content creation time. I am on a learning journey, I do not want to spend countless hours writing articles that do not help my learning curve. But I still want to share some of my experiences once in a while. And that I can now do a lot easier than before.
My Main Advice: Treat Your AI Like a Junior Team
Don’t expect it to freewheel brilliance. Structure your workflows, provide strategic inputs, and review the outputs before publishing. Think of it as your 24/7 junior content team that needs training above all.
Your Turn Now
Automation is a workflow booster, not a replacement for real teamwork. Think about your own workflows and what could be automated. Be smart. Get help if you need it. An agency, freelance help, or some other way. If you don’t have the time or don’t want the hassle, that is absolutely ok. Teams and agencies are not the devil, as some 20 year olds want to make you believe.
By the way: Ever noticed that the same group of people all share each other’s posts and comment on them? Good old social media strategy that the algorithm long crushed… not on LinkedIn however.
See you in Part 2, where I will dig a bit deeper on some of my AI team members.
This article was written by me and my AI team in collaboration.
Marc
