I spoke at the very first "Build with AI" meetup by GDG Kuala Lumpur, held on April 9, 2026, on "Serverless Agents: Running on Cloud Run". This was my first time speaking at a GDG KL event, walking a room full of developers through the exact problem that kills most AI agent projects: they work great on localhost, then fall over the moment real users hit them. I showed how to take an agent from a laptop script to a secure, auto-scaling production service on Google Cloud Run, then proved it live with a real deployed app.
- DateApril 9, 2026
- EventBuild with AI (1st meetup), organized by GDG Kuala Lumpur
- TalkServerless Agents: Running on Cloud Run
- RoleSpeaker
- TopicsAI Agents, Model Context Protocol (MCP), Google Agent Development Kit (ADK), Google Cloud Run, Cloud Build, Artifact Registry, Secret Manager, Cloud Logging
The Talk

Talk Title Slide

Official Speaker Announcement, Build with AI Meetup #1
I opened by breaking down what actually makes an AI agent different from a chatbot: the ability to reason about a request, decide it needs a tool, call that tool to get real data, and only then answer, instead of just predicting the next words. From there I introduced MCP (Model Context Protocol), the "USB-C for AI" that gives agents one standard way to connect to outside APIs and data instead of a custom integration for every tool, and the Google Agent Development Kit (ADK), the framework that natively handles the reasoning loop, tool connectivity, and multi-turn session memory so you can skip the boring infrastructure and focus on the agent's logic.

Inside the ADK: the Runner, session & memory services, and the agent execution event loop
The second half moved to Google Cloud Run as the home for that agent: fully managed serverless, runs any container so it's language agnostic, instant autoscaling from zero to thousands of instances and back to zero, and a "pay-as-you-chat" cost model where you're only billed for the seconds Gemini is actually thinking. I walked through the full production architecture, load balancing, a React/Flask app on Cloud Run, and its connections out to Vertex AI, BigQuery, and Dataplex, and covered why Cloud Run in particular fits AI agents: high concurrency (one container handles hundreds of simultaneous requests) and extended timeouts of up to 60 minutes, so a long agent reasoning loop never gets killed mid-thought.

Production architecture: Cloud Run as the central hub between the frontend, Gemini, and enterprise data services
Live Demo: HostFlow Agent
For the demo, I used my own project, HostFlow AI, as a live, real-world case study. I showed the guest chat answering a WiFi question by pulling live property data out of Supabase and injecting it straight into the agent's context, meaning the exact same Cloud Run agent can serve hundreds of different properties just by swapping the injected context. Then I asked it for the nearest 24/7 fast food place while watching the Cloud Run logs live on screen: the Root Agent delegated to a PlacesAgent, which called the Google Maps API and returned real, live results instead of a hallucinated answer.
The finale was an emergency escalation test, typing "our Airbnb is on fire" and watching the agent skip any explanation, immediately hand off to an EscalationAgent, and fire a live Telegram alert to the host's phone in under a few seconds, all visible in the Cloud Run logs in real time. I then had the entire room scan a QR code and hit the live agent at once, watching the Cloud Run instance count scale up from zero in real time on the metrics dashboard as the traffic came in.

Live demo: the HostFlow AI guest agent answering questions and escalating an emergency in real time
I closed with a tour of the production tooling that makes this deployable and safe to run: Cloud Run Revisions for instant rollbacks and traffic splitting, Cloud Build for CI/CD triggered on every git push, Artifact Registry for versioned, reproducible container images, Secret Manager so zero API keys ever touch the codebase or the frontend, and Cloud Logging for full visibility into the agent's reasoning and tool calls, showing that this wasn't just a cool demo but an enterprise-grade, production-ready pattern.
Photos

Presenting Serverless Agents at the GDG KL "Build with AI" meetup

A packed room for the first "Build with AI" meetup