Stop Building Franken-Platforms: The 5 Planes of Internal Developer Platform Architecture
Platform engineering is currently the fastest-growing discipline in cloud infrastructure. According to Gartner, over 80% of software engineering organizations will have established platform teams by 2026.
Yet according to Puppet's State of Platform Engineering research, over 70% of platform engineering initiatives fail to deliver measurable velocity in their first 18 months.
The primary cause of failure is not a lack of budget or tooling. It is an architectural failure mode known as the Franken-Platform.
What is a Franken-Platform?
A Franken-platform occurs when platform engineers assemble modern cloud tools without defining architectural boundaries or contracts between them:
-
The Portal Anti-Pattern: The team deploys Spotify Backstage or Port. To enable self-service, they write an action that runs an ad-hoc bash script executing
aws rds create-db-instancedirectly from inside the portal worker pod. Secrets are hardcoded, there is zero GitOps audit trail, and deleted repositories leave orphaned cloud resources running forever. -
The Monolithic GitOps Repo: Developers are forced to edit 600-line Helm
values.yamlfiles filled with VPC CIDRs, Kubernetes tolerations, and IAM role ARNs. A typo in a YAML indentation breaks deployment for multiple services. - The Isolated SRE Dashboard: Prometheus and Grafana are accessible only to the infrastructure team. Developers deploy memory leaks into staging with zero visibility until the monthly AWS bill spikes by 40%.
Gluing tools together with bash scripts does not create an Internal Developer Platform. It creates a second, more complicated layer of tribal complexity.
The 5 Planes Reference Architecture
Pioneered by the CNCF Platforms Working Group and PlatformEngineering.org, the 5-plane reference architecture decouples developer intent from infrastructure topology.
[ Plane 1: Developer Control Plane ] (Score.dev, Portals)
│ (Emits Workload Contract)
▼
[ Plane 2: Integration & Delivery Plane ] (Argo CD, GitHub Actions)
│ (Reconciles Desired State)
▼
[ Plane 3: Resource Plane ] (Crossplane, Kubernetes, Cloud Primitives)
▲
│ (Admission & Policy Intercepts)
[ Plane 4: Security & Governance Plane ] (Kyverno, HashiCorp Vault)
▲
│ (Streams Cost & SLIs back to Plane 1)
[ Plane 5: Monitoring & FinOps Plane ] (OpenCost, OpenTelemetry)
Plane 1: Developer Control Plane (Interface & Context)
The single pane of glass for developers. It includes the software catalog, documentation (docs-like-code), and workload specifications.
The Golden Rule: Developers declare what their application needs, never how cloud infrastructure runs.
# score.yaml
apiVersion: score.dev/v1b1
metadata:
name: order-service
service:
ports:
web: { port: 8080, targetPort: 8080 }
containers:
app:
image: ghcr.io/controlplanelab/sample-order-service:v1.2.0
variables:
DATABASE_HOST: ${resources.db.host}
resources:
db:
type: postgres # What is needed, not where it is hosted
Plane 2: Integration & Delivery Plane (Orchestration & State Sync)
Responsible for image compilation, cryptographic signing (Cosign), and continuous GitOps state reconciliation (Argo CD / Flux).
The Anti-Pattern: Never turn Plane 2 into a cloud provisioning engine. Plane 2 reconciles manifests into clusters; it should never directly invoke AWS/GCP APIs.
Plane 3: Resource Plane (Infrastructure & Operators)
The execution tier where compute, networking, and cloud services execute. Declarative operators like Crossplane translate workload claims into compliant cloud resources:
apiVersion: database.controlplane.io/v1alpha1
kind: PostgreSQLInstance
metadata:
name: order-service-db
spec:
compositionSelector:
matchLabels:
provider: aws-rds
tier: production-ha
parameters:
storageGB: 50
engineVersion: "15.4"
Plane 4: Security & Governance Plane (Automated Guardrails)
Replaces slow, bi-weekly Change Advisory Boards with automated admission controllers (Kyverno / OPA Gatekeeper). Workloads violating platform standards (running as root, missing CPU limits) are blocked in <50ms at admission time.
Plane 5: Monitoring & FinOps Plane (Closing the Feedback Loop)
Observability is not just for SREs. Tools like OpenCost and OpenTelemetry attribute real-time cloud costs per hour back to service owners in Plane 1, alerting developers before cost spikes become executive crises.
Download the Complete Reference Blueprint
All architecture diagrams, Score workload specs, Argo CD manifests, Crossplane claims, Kyverno policies, and OpenCost configs from this teardown are available in our open-source blueprints repo:
👉 GitHub Repository: ControlPlaneLab/blueprints (EP02-five-planes-idp)
👉 YouTube Video: The 5 Planes of IDP Architecture (Blueprint Teardown)
👉 Channel: Control Plane Lab (@ControlPlaneLab)
Top comments (0)