> Markdown version of [/events/world-congress-2026-europe/sessions/1298-future-of-mobile-ai](https://www.wearedevelopers.com/events/world-congress-2026-europe/sessions/1298-future-of-mobile-ai). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Future of Mobile AI. What On-Device Intelligence Means for App Developers


- **Date:** Friday, Jul 10, 2026
- **Time:** 12:20–12:50 (30 min)
- **Room:** Stage 12
- **Event:** World Congress 2026 Europe

## Recording

[Watch recording](https://www.wearedevelopers.com/videos/100264-future-of-mobile-ai-what-on-device-intelligence-means-for-app-developers)

## Description

Two years ago, adding AI to your app meant one thing: cloud APIs. You sent data to a server, waited for a response, paid per request, and hoped your users had good internet. Privacy? A terms-of-service checkbox.

That world is ending.

Today, you can run a large language model directly on a phone. No internet required. No per-request costs. Data never leaves the device. This isn't a research demo — it's production-ready technology that changes what's possible for app developers.

I built flutter_gemma, an open-source plugin that lets developers run AI  models like Gemma locally on iOS, Android, and Web. Through this work,  I've learned what on-device AI actually means in practice — not the marketing version, but the real tradeoffs, limitations, and opportunities.

In this talk, I'll share what I've discovered:

What's now possible — Running models like Gemma 3 on a smartphone. The hardware (NPU, Neural Engine) that makes it work. The formats (.task, .litertlm) that matter.

What changes for developers — New architectural patterns: offline-first AI, hybrid cloud/edge approaches. New decisions: which model size, which format, where to store gigabytes of weights. New skills: fine-tuning, conversion, optimization.

The honest tradeoffs — Not every phone can run every model. Smaller models are faster but less capable. Some support separate LoRA weights for easy updates, others require full model replacement. I'll explain what works where.

Where we're heading — Multimodal models (text + images) on device. Function calling — AI that controls your app. Personalization through on-device fine-tuning. Models designed specifically for edge, like Gemma 3n.

The future of mobile AI isn't about replacing cloud — it's about giving developers a new option. One that's private, fast, and works anywhere.

## Speaker

### [Sasha Denisov](https://www.wearedevelopers.com/@sasha-denisov)

CTO and Co-Founder of Brainform.ai


## Related talks at this congress

- [Edge AI on iOS: Beyond the Cloud, Designing the Next Generation of Intelligent On-Device Apps](https://www.wearedevelopers.com/events/world-congress-2026-europe/sessions/1256-edge-ai-on-ios) — Precious Osaro
- [AI in Your Browser: Exploring Chrome's Built-In LLM](https://www.wearedevelopers.com/events/world-congress-2026-europe/sessions/1176-ai-in-your-browser) — Daniel Ostrovsky
- [From Perception to Autonomy: Building Agentic Edge AI Robots with ROS 2](https://www.wearedevelopers.com/events/world-congress-2026-europe/sessions/1324-from-perception-to) — Moe Sani
- [Prompt, Render, Debug: AI-Powered React Development with Chrome DevTools](https://www.wearedevelopers.com/events/world-congress-2026-europe/sessions/1222-prompt-render-debug) — Suchitra Swain