Why Platform Engineering Matters
Every engineering organization eventually hits the same wall: developers spending more time fighting infrastructure than building features. The promise of "you build it, you run it" is powerful — but without the right abstractions, it becomes "you build it, you suffer through it."
An Internal Developer Platform (IDP) is the answer. It provides golden paths — opinionated, well-supported workflows that cover the majority of use cases while still allowing escape hatches for edge cases.
The Architecture
Our platform sits on top of Kubernetes and provides three core layers:
-
Infrastructure Abstraction — Developers declare what they need (a database, a cache, a queue), not how to provision it. Crossplane translates these declarations into cloud resources.
-
Deployment Pipeline — A GitOps-based workflow using ArgoCD. Push to main, get a deployment. No Jenkinsfiles, no pipeline YAML. The platform handles rollouts, canaries, and rollbacks.
-
Observability by Default — Every service gets metrics, logs, and traces out of the box. No SDK integration required — the service mesh (Istio) and OpenTelemetry collectors handle it.
Golden Paths, Not Golden Cages
The key insight: a platform should make the right thing the easy thing, not the only thing.
We provide scaffolding templates (powered by Backstage) that generate a new service with:
- Dockerfile optimized for the language runtime
- Helm chart with sensible defaults
- CI pipeline that builds, tests, and pushes
- Pre-configured alerts and dashboards
But if a team needs something different — a custom runtime, a non-standard deployment strategy — they can always drop down to raw Kubernetes manifests.
Lessons Learned
Start with the developer experience, not the technology. We initially built the platform bottom-up: Kubernetes, then Helm, then ArgoCD. What we should have done is start with the question: "What does a developer need to do on day one?"
Measure platform adoption, not platform features. The number of CRDs you support doesn't matter. What matters is: how many teams are using the golden paths? How long does it take to go from "new service idea" to "running in production"?
Treat your platform as a product. It has users (developers), it needs a roadmap, and it needs feedback loops. We run monthly surveys and track NPS for the platform team.
What's Next
We're investing in two areas:
- AI-assisted operations — Using LLMs to analyze incident patterns and suggest runbook actions
- Cost attribution — Giving teams real-time visibility into their infrastructure spend, tied to business metrics
Platform engineering is not about building the perfect abstraction. It's about continuously reducing the friction between an idea and production.