Fable is back — what using an expensive, deeply capable model teaches about discriminating AI use

Fable is back. And it’s changing how I think about thinking.

The most powerful model I use is also the one teaching me to slow down.

Last night I ran it against my whole operating system: goals, priorities, the actual strategy behind what I’m building. Not a single document, the whole thing. It caught tensions I’d been circling for weeks without naming them.

That’s not new for a strong model. What’s new is what it demanded of me to get there.

With Opus I fire off five sessions in parallel and crunch. Write the prompt, let it rip, check back in ten minutes, go again. It’s fast and cheap enough that sloppy prompting barely matters, you just get another shot.

Fable doesn’t work like that. It’s capable enough, and runs deep and long enough, that careless prompting wastes the capability itself. So I sit down and think before I type. What am I really asking. What do I actually want out of this. That’s the tell, not the output quality.

For me, that’s the shift. The scarce input isn’t compute anymore, it’s a clear question. My job is moving from executing fast to deciding where the deep thinking actually needs to happen.

Who else is experiencing something like this?