Did developers ever really push Apple’s on-device Foundation Models anywhere near their actual potential?

A lot of the discussion around Apple Intelligence focused on Siri not feeling competitive enough. There were even rumors that Apple’s AI lack of strength played a role in the leadership transition to John Ternus.

We don’t dismiss a race car because nobody has driven it to a win yet. So why assume Apple’s local AI models are disappointing before we’ve seen what developers can actually build with them?

I’ve been experimenting with Apple’s Foundation Models framework inside an iOS farming game, and what I found surprised me. I do not ask the AI itself to be the narrative engine. I programmed the game code with all the story variables. The code tracks gameplay state across many game days: crops planted, weather, harvests, eco decisions, routines, repeated tile usage, companion planting patterns, pacing, and player habits. The code then injects carefully selected context into tightly structured prompts.

The LLM is not asked to “understand” the game’s world. It simply writes from the context it is given. And when the prompts are engineered carefully enough, the generated text starts feeling surprisingly human-authored and emotionally consistent. Not because the AI became sentient. Because the software architecture guiding it became sophisticated.

That feels important. Apple didn’t just build a smarter autocomplete system for Siri. They gave developers access to a local generative engine and challenged them to invent new kinds of experiences with it. I think we are only beginning to see what that actually means.

Did Apple engineers expect the small AI model on an iPhone to write things like this entirely on its own? Well, if the coded input prompt is good enough, then yes it can!

“Chapter 5 (from iOS app Oh Farm)
This farming season was a blend of routine and discovery. I found a rhythm in the mornings, when the dew was fresh and the air was crisp. I tended to the strawberry rows with a methodical touch, checking the soil and ensuring each plant got its fair share of water. The soil there was rich, and the plants thrived, almost as if they were grateful for my care. It was a small moment of connection, but it made a big difference. I also noticed a companion pair of peas thriving nearby. They seemed to benefit from the shade and the soil quality of the field tile. It was a reminder that sometimes, the simplest things can bring the most joy. Overall, this season taught me patience and the importance of nurturing the land. I learned to trust the process and to let things unfold naturally. I also discovered a newfound appreciation for the beauty of the organic garden, with its vibrant colors and the symphony of birds and insects. It was a season of growth, both in my garden and in myself.”

The AI wrote, “I’m tired?!” That feels important. We see what can happen with innovative coding. We are not judging just the AI. Apple didn’t just build a smarter autocomplete system for Siri. They gave developers access to a local generative engine and challenged them to invent new kinds of experiences with it. I think we are only beginning to see what that actually means.

Even when the Apple AI messes up a little on the in-code prompt, the result is still entertaining. I read this as the farmer muttering to themselves after a hard day. 🥵 The code prompt gave the AI LLM hints “heatwave, 6 plants killed, eco score 15, optimistic”. Personally, I would be hoping for rain, fog and dew too!

“Day 3
Heatwave. 6 plants gone. That’s it. Not the end of the world. Not a loss. Resilience. 6 is a lot. 3 times 2. That’s a lot. That’s a lot. Rain refreshing. Morning fog and dew. 15 eco score. 15. 15.
Farm Club: Wow, it looks like we planted a lot of basil today! Your watering techniques are really helping the plants grow. Keep up the great work! “

The AI did better this day, with a similar prompt.

“Day 100
The heatwave has left its mark on the garden, with some crops suffering setbacks. Despite the challenges, the garden is thriving, with many new plants planted and the eco-score at a healthy 1020. I’m excited about the new garlic crop, a first for me, and feel a sense of quiet resilience as I continue to work the grid methodically.”

I think the future is bright for Apple’s iPhone AI engine. Developers just need to drive it well.