Fun Friday: automating my publishing chore with n8n and Gemini
Again it is fun Friday, how lucky to start my 2026 with a fun Friday. I followed some software 前輩 advice to roll up my sleeves and build. I have built on a suite of new tools and workflows to eliminate a source of boredom in my routine.
Ever feel like you have a hard time maintaining a consistent online presence? Maintaining a vibrant online presence on multiple platforms can feel like an endless publishing chore. That was my Friday problem. I decided to start building an agent to do the heavy lifting of copy and pasting.
I spent 0 minutes deciding I will go with the default: Google Doc, n8n and Gemini. Most of the time I will spend time evaluating but I decided to combine my experience and collective wisdom with my intuition.
Then I outlined my multi-platform publishing nightmare to Gemini. It helped me think through the entire workflow and its potential roadblocks. My prompting techniques remained basic(Secret: “Be my most critical and nitpicking boss and comment on my plan.”), I still needed to clarify and craft the workflow in concept as concisely as possible.
After I was satisfied with the workflow “pseudocode”, I asked Gemini to generate the initial workflow in n8n JSON. Voilà! The whole JSON was instantly laid out. And I thought it was done. Turns out I am again too optimistic. The real learning happened in the “last miles”. I spent 67% of my time here and learn two things:
- Integrating services is still manual: For instance the GitHub API (which demands Base64-encoded files) while my initial process outputted binary data meant I spent half my time meticulously “gluing” those nodes together.
- The tools are evolving faster than the foundation model. The output from Gemini does not adhere to the latest n8n version, even when I instruct it to process the latest document. Luckily I am equipped with the patience and prior knowledge to try & read and understand.
I can see some of my time can be used to prompt a better model (Claude Code Opus 4.5 maybe?) with better questions. But this is the key: LLM as a new abstraction of software is still flaky by nature. I borrow one of the best tech founders coined: working with LLM is like working with a drunk intern, but this intern’s best performance IS getting better very fast.
Do you have any more interesting little projects that help you work a bit faster and learn new tools at the same time? Let us know!
#Automation #GenerativeAI #WorkflowOptimization