Kai

Apple Intelligence is actually powerful enough for some daily tasks.

by

When designing our AI features in Weather mini 3, we avoided third-party cloud services and heavy local LLM setups. Instead, we rely on what’s already built into users devices — no accounts, no downloads, no configuration. The goal is simple: a truly easy, “just works” experience that’s good enough to be useful in real life.

137 views

Add a comment

Replies

Best
Kai
Thanks!For Apple Intelligence, the on-device mode in 2025 is indeed not that powerful, comparing even to some local LLMs. However, when choosing the right task (for example, text-based tasks within token limits), the powerfulness is not the only thing that really matters to end-user experience: we need to consider speed, efficiency, privacy, and cost. I think this is where on-device Apple Intelligence really starts shining: no setup, no quota limits, no networking requirements... that means users really own this AI-related feature, and only then this can be a real potential to become a great feature for daily usages.
Zac Zuo

Love the "native" approach here. Avoiding heavy dependencies makes a lot of sense for a utility app.


From a developer's perspective, how do you evaluate the current state of Apple Intelligence? Does it effectively support the full user experience you envisioned for Weather mini?

And compared to cloud models, in what specific areas do you feel Apple's on-device capabilities are still falling short?

John Heeter

If the quality is not good, who cares if it's fast?

Kai

@jdheeter Totally.

We still care about quality, but our use case is narrow: parsing dates/cities and summarizing weather, not open-ended creative work. For that kind of simple text workflow, a smaller on-device model can hit “good enough” quality while staying instant and private.

John Heeter

@kailuo Perfect! Yeah you don't need a giant datacenter for that.