{"@attributes":{"version":"2.0"},"channel":{"title":"DEV Community: Kim","description":"The latest articles on DEV Community by Kim (@ki3ani).","link":"https:\/\/dev.to\/ki3ani","image":{"url":"https:\/\/media2.dev.to\/dynamic\/image\/width=90,height=90,fit=cover,gravity=auto,format=auto\/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F461692%2F1a5bae3f-55f9-4a97-b3a3-b99aec3197c5.jpg","title":"DEV Community: Kim","link":"https:\/\/dev.to\/ki3ani"},"language":"en","item":[{"title":"Building AI Agents with GKE","pubDate":"Mon, 22 Sep 2025 17:21:20 +0000","link":"https:\/\/dev.to\/ki3ani\/building-ai-agents-with-gke-5clk","guid":"https:\/\/dev.to\/ki3ani\/building-ai-agents-with-gke-5clk","description":"

This article was created as part of my official submission to the GKE Turns 10 Hackathon.<\/em><\/p>\n\n

The challenge was simple but profound: take a standard microservices application the Online Boutique and \"supercharge\" it with AI without touching a single line of the original code. The goal was to build an external, containerized \"brain\" that could add a new layer of intelligence.<\/p>\n\n

My answer? I built a fully autonomous, multi-agent AI team on Google Kubernetes Engine (GKE). Here's how I did it, and how GKE's features were fundamental to making this complex architecture work.<\/p>\n\n

\n \n \n The Foundation: GKE and Service Discovery\n<\/h2>\n\n

The entire project relies on multiple new microservices (agents, servers, UIs) communicating with each other and with the original Online Boutique app. GKE's built-in service discovery made this incredibly simple.<\/p>\n\n

By defining a Kubernetes Service<\/code>, I could give each component a stable DNS name. For example, the Marketing Agent's Service<\/code> looks like this:
\n<\/p>\n\n

\n
apiVersion<\/span>:<\/span> v1<\/span>\nkind<\/span>:<\/span> Service<\/span>\nmetadata<\/span>:<\/span>\n  name<\/span>:<\/span> marketing-campaigner-service<\/span>\nspec<\/span>:<\/span>\n  selector<\/span>:<\/span>\n    app<\/span>:<\/span> marketing-campaigner-agent<\/span>\n  ports<\/span>:<\/span>\n  -<\/span> port<\/span>:<\/span> 8080<\/span>\n    targetPort<\/span>:<\/span> 8080<\/span>\n<\/code><\/pre>\n\n<\/div>\n\n\n\n
This simple YAML meant that my Business Analyst agent could reliably send A2A messages just by calling http:\/\/marketing-campaigner-service:8080\/goal<\/code><\/a>. GKE handled all the complex networking behind the scenes.<\/p>\n\n
\n  \n  \n  The \"Senses\": An MCP Server with a Secure Sidecar\n<\/h2>\n\n
The first component I built was an MCP Server to act as the system's \"senses.\" It ingests real-time \"add to cart\" events from the app's frontend-proxy<\/code>. To connect securely to the Cloud SQL database, I used the Cloud SQL Auth Proxy<\/strong> running as a sidecar container, a best practice made easy by GKE's pod architecture.
\n<\/p>\n\n
\n
    spec<\/span>:<\/span>\n      serviceAccountName<\/span>:<\/span> mcp-toolbox-sa<\/span>\n      containers<\/span>:<\/span>\n      -<\/span> name<\/span>:<\/span> server<\/span> # My Python App<\/span>\n        image<\/span>:<\/span> ...<\/span>\n        env<\/span>:<\/span>\n          -<\/span> name<\/span>:<\/span> DB_HOST<\/span>\n            value<\/span>:<\/span> \"<\/span>127.0.0.1\"<\/span>\n      -<\/span> name<\/span>:<\/span> cloud-sql-proxy<\/span> # The Sidecar<\/span>\n        image<\/span>:<\/span> [<\/span>gcr.io\/cloud-sql-connectors\/cloud-sql-proxy<\/span>:<\/span>2.8.0<\/span>]<\/span>(http:\/\/gcr.io\/cloud-sql-connectors\/cloud-sql-proxy:2.8.0)<\/span>\n        args<\/span>:<\/span> [<\/span>\"<\/span>...\"<\/span>]<\/span>\n<\/code><\/pre>\n\n<\/div>\n\n\n\n
\n  \n  \n  The \"Strategist\": A Scheduled Agent with CronJob<\/code>\n<\/h2>\n\n
My Business Analyst agent doesn't need to run 24\/7; it just needs to wake up periodically to check for trends. A GKE CronJob<\/code> was the perfect tool for this. With a simple schedule<\/code> expression, I could define this \"sense-think-act\" loop to run automatically.
\n<\/p>\n\n
\n
apiVersion<\/span>:<\/span> batch\/v1<\/span>\nkind<\/span>:<\/span> CronJob<\/span>\nmetadata<\/span>:<\/span>\n  name<\/span>:<\/span> business-analyst-agent<\/span>\nspec<\/span>:<\/span>\n  # Run every 5 minutes<\/span>\n  schedule<\/span>:<\/span> \"<\/span>*\/5<\/span> <\/span>*<\/span> <\/span>*<\/span> <\/span>*<\/span> <\/span>*\"<\/span>\n  jobTemplate<\/span>:<\/span>\n<\/code><\/pre>\n\n<\/div>\n\n\n\n
After deploying, GKE takes care of the rest, spinning up and shutting down pods on schedule. Watching the pods transition to Completed<\/code> status was proof the autonomous loop was working.<\/p>\n\n
\n  \n  \n  The \"Executor\": Agent Collaboration via A2A\n<\/h2>\n\n
When the Analyst agent finds a trend, it delegates a task to the Marketing agent. This Agent2Agent (A2A) communication is just a simple, decoupled POST<\/code> request, made possible by the GKE Service<\/code> we defined earlier.
\n<\/p>\n\n
\n
\ndef<\/span> send_a2a_goal<\/span>(<\/span>self<\/span>,<\/span> product_name<\/span>):<\/span>\n    # GKE DNS resolves 'marketing-campaigner-service' to the correct pod IP\n<\/span>    [<\/span>self<\/span>.<\/span>marketing<\/span>](<\/span>http<\/span>:<\/span>\/\/<\/span>self<\/span>.<\/span>marketing<\/span>)<\/span>_agent_url<\/span> =<\/span> \"<\/span>http:\/\/marketing-campaigner-service:8080\/goal<\/span>\"<\/span>\n\n    a2a_message<\/span> =<\/span> {<\/span>\n        \"<\/span>goal<\/span>\"<\/span>:<\/span> \"<\/span>create_promotional_content<\/span>\"<\/span>,<\/span>\n        \"<\/span>payload<\/span>\"<\/span>:<\/span> {<\/span> \"<\/span>product_name<\/span>\"<\/span>:<\/span> product_name<\/span> }<\/span>\n    }<\/span>\n    [<\/span>requests<\/span>.<\/span>post<\/span>](<\/span>http<\/span>:<\/span>\/\/<\/span>requests<\/span>.<\/span>post<\/span>)([<\/span>self<\/span>.<\/span>marketing<\/span>](<\/span>http<\/span>:<\/span>\/\/<\/span>self<\/span>.<\/span>marketing<\/span>)<\/span>_agent_url<\/span>,<\/span> json<\/span>=<\/span>a2a_message<\/span>)<\/span>\n<\/code><\/pre>\n\n<\/div>\n\n\n\n
The Marketing agent, running in its own Deployment<\/code>, simply listens for these incoming requests and executes its own logic.<\/p>\n\n
\n  \n  \n  Agent 3: The AI Specialist (A2A & MCP Client)\n<\/h2>\n\n
This was the most ambitious step. I replaced the original recommendationservice with my own AI-powered Recommendation Agent. This agent is a showcase of advanced agent design, using both A2A collaboration and the MCP pattern. It acts as a gRPC server, and when it receives a request from the frontend:<\/p>\n\n
\"pods\"<\/a><\/p>\n\n