By skyforbes - I teach AI to ad agencies, and I'm refreshing curriculum.
- Yesterday's lesson was around building automations. It's for non-technical people, so instead of doing a Make.com UI lesson, I wanted to show how to leverage ChatGPT to output a "Blueprint."
- If you look at the forums, it's a known challenge to get these Blueprints right. I'm sure I could eventually get my custom GPT to do it, but I wasn't going to subject my students to it.
- So, I was focusing on more of a "get it out of your brain and into an LLM to make sense of it" direction.
Here's where it gets interesting. ChatGPT found a solve that I hadn't heard about. Have you tried "Make Maia (sounds like "me a?") It's a text to automation agent from Make.com that works like a text-based vibe coding platform. No, you hadn't heard? I hadn't either. But it's been about 2 months since I've built anything in Make, so that wasn't out of the question. There are 100 new tools every fucking day. It's pretty normal for me to miss one.
In the interest of time, I decided to build a Custom GPT that included Make Maia parameters without playing around with Make Maia first (spoiler alert…this is where I ignored my own advice…don't do this). The plan was, user provides a structured use case; GPT spits out an optimized prompt that Make Maia would use to deliver a functional workflow. We connect data sources/APIs/etc., and we're off to the races, right?
WRONG! Not off to the races. Not even hearing the starter pistol go off! Make Maia isn't live yet! ChatGPT hallucinated that Make Maia was a legitimate entity that I could use present day (Nov 24, 2025).
Absolutely infuriating.
So here's the lesson, as someone who considers himself a proficient GenAI user. You are not immune from being fooled. Check your work!!!
Hoping this helps somebody not waste 5 hours of their life!
