Workflow 4 is the finish line. n8n’s native WordPress connector takes the AI-approved draft and creates a post in seconds, title, body, excerpt, and tags already filled in. Here’s how the simplest build of the series works, and why I still hit publish myself.
Building in Public: How Workflow 4 Takes Approved Drafts to WordPress Read More »
Workflow 3 is where the actual writing happens. Verified topics go in; a complete blog draft — title, body, excerpt, and tags — comes out. Here’s how it works, and what I learned about the difference between the way most people use AI and the way an agent uses it.
Workflow 1 finds the ideas. Workflow 2 makes sure they’re actually true. Here’s how I built a fact-checking layer using Tavily’s AI search API — and what I learned about designing the email workflow around real human behavior.
Building in Public: How Workflow 2 Added a Fact-Check Layer to My AI Content Pipeline Read More »
Workflow 1 of my AI content pipeline is live — and building it taught me more than I expected. From picking the right AI partner to making a small inbox decision that changed how the whole system works, here are five honest lessons from the build.
Building in Public: How Workflow 1 Is Getting Ideas For A Blog Posting AI Agent Read More »
We constantly hear that AI is going to automate everything, but what happens when the AI gets stuck in a loop of its own mistakes? When my partner’s restaurant lost their automated beer menu tool, I decided to build a replacement. What should have been a straightforward task turned into a frustrating battle with the “AI Hallucination Loop” across Claude and Gemini. Discover why automation still requires a human at the wheel, and learn three practical product management lessons to help you successfully build your own AI workflows without losing your mind.
The AI Hallucination Loop: Why Automation Still Needs a Human at the Wheel Read More »
Building with AI makes it easy to prioritize technical novelty over the actual problem. In this case study, I share the strategic lessons learned from rebuilding an AI content agent—shifting from a brittle, monolithic structure to four disciplined, modular workflows that prioritize educational quality and long-term flexibility over sheer content volume.
Rebuilding the AI Agent: A Lesson in Strategy Read More »
So, I started over based on what I had learned firsthand, through my research, online reading, and talking to others. My first change was to break up my agent into several workflows instead of one long one. This makes troubleshooting a lot easier and gives me more flexibility as I learn more or tools change. I also want to change the focus, through targeted sources and prompts, so that my agent focuses on teaching, not the latest trend. Finally, I switched from 1min.ai to Google Gemini, as n8n works better with Gemini and causes fewer issues overall, letting me focus on refining the model rather than tracking down system errors.
Building an AI Agent Read More »