You haven't found AI's ceiling, you've found yours. With any software before ChatGPT and Claude, people understood they weren't experts.

Now everyone thinks they're a savant.

LLMs have a freakishly low floor and a *huge*, invisible ceiling that most people don't know is even there.

Anyone can type a sentence and get something impressive back in ten seconds, that's the floor, it's intoxicating.

Software like Photoshop and Excel show you their depth... panels, menus, functions you visibly haven't touched in years.

You KNOW you're not an Excel or Photoshop expert, even if you can play around the edges and do some stuff.

But your LLM chat box shows you nothing but a chat box, so you assume the thing *you* can do is the limit of what *it* can do.

I've lost count of the people I talk to who say 'I tried to get it to create X and it was a mess' and their take home is that AI can't do it... not that *they* can't get their AI to do it.

The gap between what you get back from AI and what's possible often isn't the tool, it's your craft, like when you're using any other software.