29/06/2026
⚡ GDG Skopje at Google I/O Connect Berlin 🇩🇪
On June 25th at Wilhelm Studios, we were part of Google I/O Connect 2026. 🚀
Here is the developer-level breakdown of what we brought home from the engineering stages:
🧠 AI Stage
Agentic IDEs & "Vibe Coding" Escalation: We mapped out the transition from rapid exploration in Google AI Studio to autonomous development inside Google Antigravity 2.0 an agent-first IDE designed to handle complex multi-file codebase features. We also worked directly with the Gemini Interactions API and Managed Agents to abstract autonomous workflows.
Hardware-Level Scale & Cloud TPUs: pre-training with MaxText, post-training with Tunix, and high-throughput inference serving via vLLM.
On-Device Edge AI & LiteRT-LM: By deploying Gemma-4-E2B via the LiteRT-LM APIs, we engineered custom agentic "Skills" running on-device, bypassing network latency and cloud egress costs entirely. We also integrated advanced multimodal pipelines using the GenMedia suite (Veo, Nano Banana, Genie, and Lyria).
📱 Android Stage
Jetpack AppFunctions Framework: We dived into the new AppFunctions Jetpack API, which allows developers to securely expose core application features as executable capabilities directly to OS-level AI assistants.
Multimodal Media3 Processing: We used the Gemini API to perform semantic, time-coded analysis on raw video footage and passed those payloads directly to Jetpack Media3 Editing APIs to programmatically splice and concatenate frames on-device.
"Adaptive Everywhere" UI Engine & Wear OS 7: We built fluid, highly adaptive user interfaces using Jetpack Compose's newest Grid, FlexBox, and MediaQuery APIs to scale seamlessly across form factors (mobile, foldables, Wear OS 7, Android TV, and Android Automotive OS), while exploring Jetpack XR for spatial layouts and 3D assets.
🌐 Chrome Stage
Agent-Ready Web & WebMCP Architecture: By enabling chrome://flags/ -webmcp-testing in Chrome, we experimented with creating a closed feedback loop where coding agents interact with the DOM via a DevTools MCP/CLI connection.
☁️ Cloud Stage
Multi-Agent Backend Architecture: We unpacked the deployment and management of secure, multi-agent backends across Google Cloud’s container ecosystems, comparing Agent Engine / Agent Runtime, Cloud Run, and Google Kubernetes Engine (GKE) with strict security boundaries and enterprise data grounding.
Distributed Orchestration via MCP Servers: We built location-intelligence ADK agents powered by Gemini Pro that used remote, Google-hosted Model Context Protocol (MCP) servers to securely access BigQuery for demographic/transactional analytical payloads, alongside Google Maps for real-world location analysis.
Get your environments ready. 💻🔥
