You watch the demos.
Designers building components from text prompts. Developers scaffolding full services in minutes. PMs writing user stories that generate working prototypes.
It looks magical, effortless. It looks like AI is finally living up to the hype.
But when you try it with your team, it falls flat.
The output is messy, suggestions are off, and the results are underwhelming.
You start wondering — are we doing something wrong?
Are the tools overhyped?
Or is there something deeper?
Here’s the truth, no one tells you in the demo video:
AI works when you’ve already done the hard work.
Here’s, when it does’t:
1. Your Codebase Is a Mess
If your projects are big balls of mud — full of patches, lacking boundaries, built without a sense of architecture — don’t expect ChatGPT to save you.
It will happily suggest another quick fix. Another layer on top of the pile.
Want to face the truth?
Pack the project with Repomix, put it into Google AI Studio, and ask:
“Does this code follow clean architecture principles?”
Let it reflect the truth you already suspect — and maybe, finally, spark a conversation your team’s been avoiding.
The entire article and PDF/Notion/Google doc template are available as a paid subscription. Thanks for supporting Practical Engineering Management!