Fable’s intuition for iOS development is significantly worse than other areas I’ve used it in.
Great at infra, databases, web and more, just easily confused about how mobile apps work for some reason
T3 Stack creator Theo Browne tested Anthropic's newly available Fable 5 on varied coding tasks and found reliable strength on infrastructure, databases, and web work while the same model quickly lost coherence on iOS projects.
Fable’s intuition for iOS development is significantly worse than other areas I’ve used it in.
Great at infra, databases, web and more, just easily confused about how mobile apps work for some reason
Early users report that opaque Apple frameworks leave less reliable signal for the model to follow, producing faster drift than on web or backend assignments.
Launch materials emphasize long-horizon coding wins but omit mobile-specific evals, leaving builders to discover the iOS shortfall through direct use.
Positive users report strong experiences building iOS apps with Fable or similar models, while negative users criticize struggles with native mobile frameworks and iOS quirks.
No Digg Deeper questions have been answered for this story yet.
Fwiw, Julius is usually more right than me
The amount of world knowledge in Fable is insane. It’s navigating iOS simulators without any skills way better than codex with the OpenAI plugin 🤯
@theo Off the rails
Fable’s intuition for iOS development is significantly worse than other areas I’ve used it in.
Great at infra, databases, web and more, just easily confused about how mobile apps work for some reason

@theo fable's just an android developer, dw about it

@theo I'm personally starting to hate React Native, feels like anything that is truly interactive should stay the hell out of RN. Going to try Godot and see how that goes.

@theo this is true. ive tried it on swift ui and it gives me this 😂

@saaslanee @theo My experience is the exact opposite: several models worked well when doing RN and Expo work. I've had great success with Sonnet 4.6, Opus, GPT-5.3 Codex, and even Gemma 4. I only had issues when trying Haiku for RN work.
RN evals is handy when choosing https://rn-evals.vercel.app/

@theo I’ve had pretty good luck with it, it’s was able to make a great animation of a custom-ish input field
Just curious, are you using the MCP? I know you don’t like many MCP’s, but this one is really helpful for iOS dev, especially in the new beta

@theo GPT5.5 has been excellent for me for mobile app development

@theo @Chiebuniem_

@theo all models has been the same for mobile development specially react native and expo apps

Mobile is hard for every model; it feels like it needs quiet. A bit of guidance overall, but I've had great results in cleaning up the code and making architectural decisions. Just needs more guidance
I feel like most of it is just a lack of better tooling; sim-serve goes a long way, but not nearly enough

@aatxcus @theo uhm, what? show me your projects then

@theo I'm actually having *really* good experiences with it. What type of app are you trying to develop?

@theo I have yet to find model that can get RN Reanimated & GestureHandler right without heavy steering.

@theo Now who is right?

@theo yeah overusing geometry reader and lacks coherence on macos apps

@theo mobile has so many hidden rules that web dev just doesn’t prepare you for

@theo I've had good success with it for Android app development so far

@theo It’s weirdly good at gtm stuff though. So much better than opus/5.5

@theo 5.5 is really good