// Articles
Guides, tutorials, and comparisons to help you build better backend applications.
The infrastructure behind a production AI agent: model access, state and memory, tool APIs, queues for long tasks, secrets, and tracing. What each piece does, what to use for it, and how to run the stack in your own AWS or GCP account.
Argo CD is the default GitOps controller for Kubernetes, and it still leaves you writing and maintaining every manifest it syncs. The main alternatives, from Flux to Spinnaker to push-based pipelines, and when to skip manifests entirely.
The AWS Activate credit tiers, who is eligible, how to apply, and how to spend credits on a production backend in your own AWS account instead of on idle resources.
Rolling, blue-green and canary deployments, the health checks and graceful shutdown that make them work, and the expand and contract pattern for database migrations that keeps old and new code running side by side.
What development, staging, preview and production environments are for, what has to match between them and what should differ, and how to keep them in sync without copying infrastructure config by hand.
Supabase is a Postgres database with auth, storage, realtime and generated APIs around it. Render is a host for long-running services, workers and databases. How they compare, when teams use both, and when the backend belongs in your own AWS or GCP account.
Publish and receive Pub/Sub messages from Go with the v2 client library, handle acks, flow control, retries and dead-letter topics, and get the same topics and subscriptions from Go declarations that run locally without the emulator.
Google Kubernetes Engine and Amazon EKS both run upstream Kubernetes, and they differ on how much of the cluster you manage, upgrades, networking, workload identity and what the control plane costs for backend teams.
Helm charts turn into Go templates over YAML with hundreds of values. Here are the alternatives for deploying your own services to Kubernetes, from overlays and typed code to generating the manifests from your application.
Dokku gives you git push deploys on one server, and these are the alternatives for when you want a web UI, several servers, zero-downtime container deploys, or managed databases and queues in your own AWS or GCP account.
Workers are fast at the edge and constrained everywhere else: memory per isolate, CPU time per request, partial Node.js compatibility and edge-native storage. Here are the alternatives for backends that need full runtimes, Postgres and background jobs.
Agent runs take minutes, call models many times and fail halfway. Here is how to move them out of the request into a queue and worker, track their status in Postgres, cap their cost, and run the setup in your own AWS account.