GDG Korea Android Meetup: How Are Android Developers in Korea and Abroad Using AI?
Sep 8 (Tue) 19:00
서울 강남구 영동대로 417 구글 스타트업 캠퍼스
AI-assisted development has moved far beyond simple code completion. It is beginning to reshape the way developers work altogether.
So how far are Android developers actually taking AI today, and how are they using it in practice?
At this GDG Korea Android Meetup, we’ll hear from Android developers in Korea and abroad about their real world experiences and explore different ways to integrate AI into the development process.
Speaker: Daniele Bonaldo (Mobile Consultant at Outfit Code, GDE for Android)
The rapid rise of AI coding assistants has led to a dangerous misconception: that anyone can effortlessly generate complex apps. The reality? Without strict architectural guardrails, pure “vibe coding” quickly degenerates into unmaintainable “slop.”
In this talk, we will explore the transition from simple AI autocomplete to true agentic development in Android Studio. We’ll see how to combine senior engineering experience with modern AI tooling to boost productivity without sacrificing quality.
Expect a deep dive into practical strategies, including how to use AGENTS.md and Android Skills to enforce consistent architectural constraints for AI agents, how to minimize context switching with MCP, and how to validate implemented features through natural-language Journeys.
Ultimately, this session explores how AI can evolve from a tool that simply writes boilerplate into a development partner for building high-quality, user-centric Android applications, while ensuring that developers remain the architects, guiding the AI rather than handing over control.
Speaker: Donghyun Ha (Remember)
As AI coding tools rapidly improve, another misconception has emerged: that building more tools and adding more automation automatically means we are getting better at using AI.
But unless we measure things separately, it is difficult to know whether better results came from our own system design or simply because the underlying model got better.
This talk is not about questioning the output produced by AI. Instead, it is about questioning the tools we build and the metrics we use to evaluate them.
Drawing on failures encountered while building search tools, server-side evaluation tools, and design automation systems, we’ll look at why tools end up going unused, how seemingly rigorous measurements can lead us to optimize for weaknesses that do not actually exist, and why simply documenting rules does not guarantee that AI agents will follow them.
The session will then introduce practical techniques for avoiding the trap of evaluating your own tools with your own tests, including pre-registration, third-party test creation, and ablation testing.
This session is for anyone who has built multiple AI tools but still finds themselves wondering: “Did any of this actually make things better?”