Most IDP projects fail because they try to boil the ocean. The org commits to a year-long platform roadmap, the plan calendar fills with workshops, and engineers route around the half-built thing.
A different model: ship a minimum viable platform in three weekends. Not three months — three weekends.
Weekend one: Backstage with a software catalog. Five services, three teams, every entity discoverable by URL. No bells, no plugins. Just a place where 'who owns this' has an answer.
Weekend two: a single paved path. One templated app: Next.js, Postgres, GitHub Actions, deployed to a managed Kubernetes cluster on-merge. One opinionated stack. One golden Helm chart. Anyone in the org can spin up a new service from this template in under five minutes.
Weekend three: observability defaults. The template ships with structured logging, OpenTelemetry tracing, a Datadog dashboard and a default alert policy. Engineers do not opt in to monitoring — they opt out, and they have to justify it.
After three weekends you have something that two or three early-adopter teams will use. Earn the right to grow from that. Add a second template (background workers). Add a deploy approval workflow. Add a cost dashboard. Each addition replaces a bespoke per-team thing. The platform grows by absorbing complexity that already exists, not by inventing new abstractions.
The single most common reason internal platforms fail is that the team building the platform forgets it is shipping a product. The customer is the engineer down the hall. The metric is whether they pick the paved path on their next service. Everything else is theatre.