Post

Watching AI agents use the terminal makes me think the CLI was always a really bad UX for humans. LLMs remember so many more flags than me, and are far better at combining them, along with pipes, to get the information it needs.

naugtur

@naugtur.pl

9 Feb

They only know that because select humans got there before

Ingvar Stepanyan

@rreverser.com

9 Feb

Unlikely. I mean, they can just read and train on the whole official man pages, something most humans don't bother with or have capacity for.

Sam Rose

@samwho.dev

It depends on what exactly is meant but I think the take misunderstands how LLMs work. They’re capable of learning novel things and performing tasks they’ve never seen before just from inside the context window. They aren’t limited to only producing what they’ve seen.

9/2/26, 1:33 pm0 reposts1 like

Ingvar Stepanyan

@rreverser.com

9 Feb

Yeah that's what I'm saying too.

Sam Rose

@samwho.dev

9 Feb

There’s also reason to believe that they come up with generalised solutions to problems and can extrapolate those solutions to data they’ve never seen before. It’s not super well understood yet but there are papers out there analysing how LLMs perform modular arithmetic in a generalised way.