I recently came across this post on LinkedIn:

I got this automated message from the "United Airlines" that said the message was created by GenAI. I'm assuming they built an agentic AI system to automatically trigger sending this message based on a flight delay.

Can you notice the issue with this message though? Curious what you think about United deploying this system which is clearly not 100% ready, but almost there. Would you do the same? Or wait to get the AI output more cleaner before productionalizing it?

This serves as a period reminder that you don't need to use an expensive, probabilistic system (genAI) when a simple, cheap, deterministic system (code) will do the job.

For those who don't know me: I work in AI. I am a big fan of AI. And I'm especially a fan of deploying AI into suitable use cases. Which means I've developed a nose for use cases that are clearly Using-AI-for-AI's-Sake. The automated message in the post below fits the bill.

This isn't limited to genAI, either. Remember a few years ago, when Twitter developed a facial recognition model to crop images to fit the timeline view? I wrote about it at the time but the gist is: why go through the hassle of developing an ML model when you could ... scale the image? A few lines of code. Near-zero CPU power. And you know exactly what you're getting every time.