Agentic AI Engineering Lab #5 – Robust Deployments & The Orchestrator Agent

In our last sprint, we established our cloud-native foundation by containerizing our agents and deploying them to Google Cloud Run. But manual deployments don't scale and require you to dig out the exact docker build and gcloud push commands from your shell history every time you tweak your agent’s prompt.

The Plan:

This session is about creating a truly robust, repeatable deployment process. We will wrap up our infrastructure setup by writing GitHub Actions that automatically build and push our agent containers to Google Cloud Artifact Registry on every commit. With a solid, "set-and-forget" pipeline in place, we'll shift focus back to agent behavior. We will build a simple orchestrator agent that uses the A2A protocol to dynamically delegate tasks to the worker agents living in the cloud.

This is a "Lab" session. We will:

• Automate for Robustness: Replace manual command-line wrangling with a GitHub Action workflow to build and push our Docker images to GCP on every push to main.
• Build the Orchestrator: Develop a lightweight central agent designed to interpret a complex user prompt and route sub-tasks appropriately.
• Live Multi-Agent Communication: Establish an active dialogue over the A2A protocol between your local orchestrator and your newly deployed cloud workers.

The Format

We run our sessions like an anthology series: Every meetup is a standalone sprint. While we often build on previous results, each session provides a complete experience and remains fully accessible to first-time attendees. This is a high-density micro-meetup limited to 6 seats.

What you'll build
By the end of the session, you'll have:

• A Repeatable CI/CD Pipeline: A set-and-forget GitHub Action that handles the Docker build and Artifact Registry push automatically, ensuring consistent deployments without the hassle.
• An Orchestrator Agent: A coordinator capable of breaking down requests and utilizing external agent tools.
• A Live A2A Network: A functioning multi-agent setup where your orchestrator successfully communicates with and aggregates results from your cloud-hosted workers.

Schedule (Saturday, 10:00 - 13:00 CEST)

• 10:00: Quick round of introductions.
• 10:15: Intro session: Robust CI/CD with GitHub Actions, Artifact Registry, and Orchestrator routing patterns.
• 10:30: Build time: CI/CD automation & building the Orchestrator.
• 12:00: Live demos + technical teardown.
• 12:45: Wrap-up + takeaways.

Prerequisites

• Accounts: A Google Account (for GCP) and a GitHub Account (for Actions).
• Laptop with dev setup: Docker and Git installed and working.
• Access to an LLM API: (Gemini, Claude, OpenAI, etc.).
• Mindset: This is a lab for engineers, not a tutorial. We learn by building, breaking, and iterating.

Logistics

• Location: We meet at my office in the Grunewald area near Martin-Luther-Krankenhaus. The exact address will be shared privately with attendees to maintain privacy.
  Capacity: Limited to 6 seats.