Cartwheel co-founder Andrew Carr compares AI agent Fable to an elite contractor that delivers exceptional code independently

@andrew_n_carr distilling tweets I see

Fable feels like a polish freelancer. Shows up, speaks in its own dialect, writes the best code you've ever seen, and then disappears

@andrew_n_carr I tried it, it wasn’t that good.. I don’t know, 🤷🏽‍♂️ it couldn’t match GPT-5.5 for me..

@andrew_n_carr This comparison really made me smile 😊

As a Polish freelancer, I only wish more of my clients appreciated my work the same way.

@steveruizok Great minds! I loved your presentation at the Lux thing

@migtissera @andrew_n_carr how would you extract a overacrhing chess strategy for a chess model instead of just a best move.

can it be that there are novel mental models to be extracted from expert chese models which allows us to make big jumps in terms of learning growth

@migtissera @andrew_n_carr i felt the thoughtfulness of fable is another level, especially with exploring novel ideas.

I do like the entropy of Anthropic models. They’re not RL’d on only a few (although could be many) dimensions. My understanding is that it’s an emergent factor from their constitutional training — I think it forces to keep some level of entropy preserved from the pre-training phase. But it’s also why it’s not super reliable like GPT-5.5 (on Codex for example). Claude Code is like a startup engineer sometimes on Red Bull and coke, lol.

But anyway, good that you had a decent experience with Fable. It was alright for me, but was unusable for most of my use cases.

@_ankur7 @andrew_n_carr Interesting.. what ideas were you exploring?

@andrew_n_carr Works for me. I need Fable back @AnthropicAI. Please?

@andrew_n_carr Almost mythopoeic, huh?

@migtissera @andrew_n_carr I hate all the Anthropic safety crap but that model was great.

@andrew_n_carr hit me up if you want the full package, minus the disappearing part ;)

@andrew_n_carr Unlike a reverse polish freelancers

Shows up, writes the best code youve ever seen in its own dialect, then leaves

Oh, I guess theyre mostly the same

@andrew_n_carr Haha

in addition to entropy, i felt it has some real belief in what it was saying.

that said i wouldn't disagree with your take necessarily. i guess creativity vs inaccuracies are probably positively correlated in humans too. suprising how much similar they are to us. thanks for sharing.

@andrew_n_carr Sorry for your past experience

@andrew_n_carr And then devs run their game localization by MT or AI instead of thinking for a moment, that running THE best language for writing prompts against THE machine, which will process these prompts, must inevitably cause syntactic fault recurrence.

@andrew_n_carr 🤣🤣🤣

@andrew_n_carr hilarious